pdf2xml

INTERNATIONAL DEVELOPMENT RESEARCH CENTRE
Ottawa • Cairo • Dakar • Montevideo • Nairobi • New Delhi • Singapore

Published by the
International Development Research Centre
PO Box 8500, Ottawa, ON, Canada K1G 3H9
http://www.idrc.ca

Main entry under title :
Population and health in developing countries. Volume 1. Population, health, and survival at INDEPTH sites

Includes bibliographical references.
INDEPTH : International Network for the continuous Demographic Evaluation of Populations and their Health.
ISBN 0-88936-948-8

1. Public health surveillance — Developing countries.
2. Public health surveillance — Africa. 3. Health planning — Developing countries.
4. Public health — Developing countries — Statistics.
5. Public health — Africa — Statistics.
6. Developing countries — Population — Statistics.
7. Health status indicators.
I. INDEPTH Network.
II. International Development Research Centre (Canada)

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, or otherwise, without the prior permission of the International Development Research Centre. Mention of a proprietary name does not constitute endorsement of the product and is given only for information.

IDRC Books endeavours to produce environmentally friendly publications. All paper used is recycled as well as recyclable. All inks and coatings are vegetable-based products. The full catalogue of IDRC Books is available at http://www.idrc.ca/booktique.

FOREWORD

Traditional sources of health information collected from health facilities often serve as the basis for health-services planning and allocation of resources in many parts of the developing world. Yet, health-facility-based data provide only fragmentary and biased information. Not all population groups have geographic or economic access to health facilities. Those that do have such access are usually self-selected and are often those who visit health-care centres only when they suffer from a serious illness. A great majority of poor people may have less access to health-care facilities than those who are better off, and poor people often treat themselves or use nontraditional health care. Women may suffer gender disparities as well, with time and cultural constraints on the use of health-care facilities, particularly in rural settings. Services for children are also severely constrained. Thus, health-facility-based data are not representative of the health problems of all rural and urban communities and do not therefore reflect their health status.

This void of valid health information for a large segment of the world’s population makes it difficult for policymakers to formulate rational health policies to improve the health of these people. As the authors of this book argue, “the need to establish a reliable information base to support health development has never been greater” (INDEPTH Coordinating Committee, this volume, p. 1). Ideally, reliable health information should be population and community based, inclusive of all groups, and collected prospectively and continuously. Such an ideal is best met through demographic and health surveillance systems collecting demographic and health data on selected population samples. Often, randomly selected cross-sectional household surveys every few years complement these methods of research.

The premier example of such a system is the Health and Demographic Surveillance System (formerly known as the Demographic Surveillance System) of Matlab, Bangladesh, which started operations in 1963 as a major component of the

field research program of the International Centre for Diarrhoeal Disease Research, Bangladesh. It is recognized as the largest and longest sustained prospective longitudinal demographic and health surveillance of any population in the world. It has made significant contributions to health development in both Bangladesh and the rest of the world. The high cost of running such a system has delayed replication in other parts of the developing world. However, thanks to the fast-paced development of user-friendly computers, this constraint has been partially overcome.

Over the last decade, a growing number of community-based field stations have evolved in Asia and sub-Saharan Africa and started to generate reliable longitudinal population-based health and demographic data. This bodes well for countries with such stations, as it marks the first step toward rational health planning and meaningful health programs for the people of these countries. Recently, these stations joined to form a network called the International Network for the continuous Demographic Evaluation of Populations and Their Health in developing countries (INDEPTH), creating “a trans-continental resource of robust, longitudinal, health and demographic data in some of the most information deprived settings in the world” (INDEPTH Founding Document; http://www.indepth-network.org). In the span of a few years, INDEPTH has matured rapidly, succeeding in strengthening the capabilities of member sites and developing strategies to harness their potential to redress long-standing inequities in health. This development has been possible because of the dedication and hard work of a few individuals, and this monograph is clearly an indication of the high quality of the network’s work.

The emergence of INDEPTH should be welcome news to the donor community, where people often, and rightly, complain that the programs they fund in low-income countries are not usually based on the real needs of the people. By the same token, donors should come out strongly in support of INDEPTH, because they will be investing in an initiative that directly addresses one of the major constraints of development assistance. Researchers in program countries should also take advantage of the INDEPTH sites to promote essential national health research. The domination of health-facility-based biomedical research should give way to policy-relevant research with the likelihood of a more immediate effect on the health of the people in the countries in the program.

PREFACE

This monograph is the first in a series from the International Network for the continuous Demographic Evaluation of Populations and Their Health in developing countries (INDEPTH). It seeks to do several things. First, it seeks to compile, for both easy reference and comparative purposes, and in detailed and summary formats, the essential characteristics of each participating demographic surveillance system (DSS) site. Second, it seeks to present, for the first time, the mortality structure of each of these sites in a coherent and comparative format. Third, based on a network-wide analysis of the African site data, it proposes a methodology to generate, again for the first time, African model life tables that are based on objective empirical data.

The focus of this volume is the structures of populations at INDEPTH sites and the characteristics of their health and survival. The monograph is divided into three parts: Part I discusses core concepts and methods used in DSSs; Part II provides a comparison of mortality patterns in INDEPTH sites; and Part III presents profiles of INDEPTH sites.

As this is the first publication of its kind on DSSs in Africa and Asia, we thought it would be expedient to discuss core concepts and methods commonly used in most of the sites. Among the concepts discussed in Chapter 1 are the DSS area, longitudinality, DSS subjects, residency and membership, and core DSS events. Rates and measures generated using DSS are discussed in Chapter 2, with specific emphasis on the use of person–years lived in calculating rates. Chapter 3 discusses the DSS methods of data collection, starting with the initial census to establish the DSS population. This chapter discusses initial censuses, update rounds, and the vital events-registration system. It also puts emphasis on mortality monitoring and the tracking of migrants. The processing of DSS data is the main focus of Chapter 4. This chapter treats the important issues of quality assurance and control at the data-processing level. In Chapter 5, Part I ends with a discussion of the quality of DSS data, both in the field and at the data centre. This chapter then provides a detailed discussion of statistical and demographic techniques for analysis of DSS data.

Part II presents a comparison of mortality patterns of INDEPTH sites for the 1995–99 period. Chapter 6 starts with a discussion of crude overall mortality at INDEPTH sites. This chapter presents an INDEPTH population-age standard for sub-Saharan Africa (SSA) for the standardization of mortality rates, and it gives the reason for using this new standard instead of the United Nations models.

The INDEPTH age standard for SSA typifies the population in developing countries, with its very young age structure. INDEPTH sites have used this standard to compare mortality in SSA. This comparison highlights age-specific mortality, considering mortality in infancy, childhood, and adulthood. This discussion compares the INDEPTH standard for SSA with the Segi population and the new World Health

Organization standard population. The chapter ends with a presentation of basic life-table indicators for INDEPTH sites, based on their age-specific mortality rates over the 1995–99 period. Part II ends with Chapter 7, which analyzes more than 6.4 million person–years of observation at the African INDEPTH sites to identify mortality patterns. The emergent patterns are demonstrated to be substantially different from conventionally used model mortality patterns applied in Africa.

Part III presents profiles of 22 INDEPTH sites. The profiles are listed in alphabetical order, first according to region, and then according to country. These profiles are expected to stand for some time as the main reference source for basic details about INDEPTH sites and their DSS operations. Based on a structured template, each profile provides a site description, including the physical geography and population characteristics. It discusses DSS procedures at the site, including data collection and processing. Finally, each profile presents basic outputs, including demographic indicators. A summary matrix of all the DSS sites, presented in the introduction to Part III, provides the core details for each site.

ACKNOWLEDGMENTS

This volume is an outgrowth of the efforts of many people, both INDEPTH members and its collaborators, who gave of their time and expertise to writing these chapters. We would like to particularly thank the following for their invaluable contributions to the corresponding chapters:

We would also like to thank INDEPTH site members, whose names are mentioned in the site profiles, for coordinating the writing of their site’s profile. Special thanks go to Rose Lusinde and Don de Savigny for producing the map panels for the site locations and particularly to Kathleen Kahn and Don de Savigny for coordinating the formatting and editing of the 22 site-profile chapters making up Part III of the monograph.

The INDEPTH coordinators would like to express their gratitude to the INDEPTH editorial committee, led by Osman A. Sankoh, for its outstanding work in compiling this first monograph. We acknowledge with pleasure the willingness of individual site teams and their leaders to collaborate in sharing such rich data sets and experiences. We also recognize the contributions of all our investment partners — local communities, public-sector services, academic and research institutions, and donors — all of whom, often over prolonged periods, continue to support and sustain our efforts. We express particular thanks and appreciation to the many sponsors of INDEPTH, including the Rockefeller Foundation, the Navrongo Health Research Centre, the Population Council, the World Health Organization, and the Andrew W.

¹ Based on Benzler, J.; Herbst, A.J.; MacLeod, B. (in alphabetical order): A reference data model for demographic surveillance systems. INDEPTH 1999, http://www.indepth-network.org.

Mellon Foundation, for providing the funds needed to enable INDEPTH networking activities to function. We look forward to attracting new partners to join with us in advancing our mission, goals, activities, and products.

Finally, we thank internal and external reviewers for their invaluable comments, which increased the validity and clarity of many sections of the monograph.

INTRODUCTION

As we enter the new millennium, with the revolution of the information age still gaining speed, it seems inconceivable that large parts of the Earth’s population remain devoid of vital health information. For 1 billion people living in the world’s poorest countries, where the burden of disease is highest, no one registers those who are born or who die or ascertains the causes of their deaths. From the limited data available, the health profile of these populations can be likened to an iceberg: the bulk of reliable data on trends in age, gender, geographic variations, and burden of disease remains hidden. This great void in population-based information constitutes a major and long-standing constraint on the articulation of effective policies and programs to improve the health of the poor and thus perpetuates profound inequities in health. The need to establish a reliable information base to support health development has never been greater.

Recently, experience has emerged from a growing number of community-based field stations that have continuous monitoring systems for geographically defined populations. These field stations generate high-quality, population-based, longitudinal health and demographic data with the potential to fill this information void in the developing world. Since 1997 a number of organizations have made a systematic effort to harness and make more readily available the products of these disparate initiatives. A series of meetings were convened by the University of Witwatersrand (South Africa) (Agincourt Health and Population Programme); Department of Tropical Hygiene and Public Health, University of Heidelberg (Germany); the Rockefeller Foundation (Bellagio, Italy); and the Ministry of Health (Navrongo, Ghana) to examine the potential for harnessing these sites through a network. These activities culminated in a meeting convened in Dar es Salaam, Tanzania, 9–12 November 1998, to establish such a network.

Seventeen field sites drawn from 13 countries in Africa and Asia participated in this founding meeting. The name adopted for the network was the International Network for the continuous Demographic Evaluation of Populations and Their Health in developing countries (INDEPTH). Network membership has increased steadily since then and currently stands at 29 health and demographic evaluation sites in 16 countries (the 13 countries whose sites are profiled in this volume are shown in Figure I.1). The network’s founding document and constitution are available on the INDEPTH website (www.INDEPTH-network.org).

Figure I.1 Countries with DSS field sites participating in the INDEPTH network.

To achieve these goals and facilitate the effective interaction of INDEPTH sites, the network has identified the concept of flexible working groups focused on specific scientific issues or topics as a key mechanism. Seven working groups were initially established, with a focus on

Two further working groups have since been formed, focusing on adult health and ethical practice. Thus, through active and concerted efforts, the network is encompassing a critical agenda founded on traditional strengths in research on infectious diseases and nutrition, with a growing emphasis on reproductive health, and the network is extending this emphasis to chronic disease, injury, and related social phenomena such as rapid urbanization. A central objective is to use network sites to train local scientists in research and research management.

This monograph is the foundation for an INDEPTH series on various themes, including model life tables for Africa and Asia; cause-specific mortality in developing countries; migration patterns; trends in fertility; reproductive health (including HIV–AIDS); and health equity.

PART I
DSS CONCEPTS AND METHODS

Chapter 1
CORE CONCEPTS OF DSS

Introduction

During the past 30 years, demographic surveillance systems (DSSs) have been established in a number of field research sites in various parts of the developing world where routine vital-registration systems were poorly developed or nonexistent. Although these systems may have been developed differently in terms of their initial rationale, they are all required to track a limited and common set of key variables determining population dynamics and demographic trends. DSSs have similar approaches to defining key variables and their relationships and to developing systems for collection, storage, and analysis of these data. The core concepts presented here draw directly from the ideas and experiences emerging from INDEPTH DSS sites in Africa and Asia. It should be emphasized, however, that even though an effort has been made to standardize the definitions, many DSS sites still define some of the concepts differently.

Demographic surveillance systems

A DSS is a set of field and computing operations to handle the longitudinal follow-up of well-defined entities or primary subjects (individuals, households, and residential units) and all related demographic and health outcomes within a clearly circumscribed geographic area. Unlike a cohort study, a DSS follows up the entire population of such a geographic area.

In such a system, an initial census defines and registers the target population. Regular subsequent rounds of data collection at prescribed intervals make it possible to register all new individuals, households, and residential units and to update key variables and attributes of existing subjects. The core system provides for monitoring of population dynamics through routine collection and processing of information on births, deaths, and migrations — the only demographic events leading to any change in the initial size of the resident population. This core system is often complemented by various other data sets that provide important social and economic correlates of population and health dynamics. These may include information on events such as household formation and dissolution, acquisition and loss of economic assets, and growth or depletion of income.

In many population sites, the DSS may also provide a platform for other studies within the same geographic area. This support varies from one study to another and may include the provision of an initial sampling frame, adjustment for confounding variables, provision of additional explanatory variables, and measurement of the demographic impact of interventions.

Demographic surveillance area

The demographic surveillance area (DSA) is an area with clearly and fairly permanent delineated boundaries, preferably recognizable on the ground (for example, rivers, roads, and clearly demarcated administrative boundaries). The clear delineation of boundaries enables an unambiguous distinction to be made between individuals, households, and residential units to include in the DSS and those to exclude.

The area of a DSS site depends mainly on the size of the population required for demographic surveillance and related research activities (for a typical example, see “Establishing the monitored population” in Chapter 3). The size is also influenced by pragmatic considerations, such as the cost to the research centre and its capacity to manage the associated logistics and human resources. The DSA may expand or shrink over time in response to changing research needs or sources of funding. These changes usually introduce additional complexity, as they alter eligibility criteria and may make it difficult to maintain consistent definitions of internal and external migrations over the period of transition.

Longitudinality

Longitudinal measurement of demographic and health variables is one of the key characteristics of a DSS. This is achieved through repeated visits at more or less regular intervals to all residential units in the DSA to collect a prescribed set of attribute data on registered subjects, who are consistently and uniquely identified. This and recording events affecting these subjects during the interval between visits allow one to construct their history and differentiate DSS data from data collected in multiround surveys and other prospective studies that allow comparison over time only on an aggregated level.

Visits

DSSs collect data during rounds, or cycles, of visits to registered residential units in the DSA. The interval between visits depends on the frequency of the changes in the phenomena under study and on the length of recall intervals for the collected data, and thus on the research focus of each field site. However, like the size of the DSA and observed population, it also depends on funding and logistics. This interval varies from one site to another, ranging from 1 week to 1 year. However, for the majority of DSSs, observations are made at 3- or 4-month intervals. This is widely considered an appropriate interval to ensure comprehensive recording of births, deaths, and migrations, which is the minimum requirement for maintaining the coherence of any DSS .

When intervals between visits are long (a year or more), researchers commonly ignore migration events and instead conduct a full census at each new round. In- and out-migration flows are then inferred through reconciliation of unlinked census records after account is taken of births and deaths between censuses.

Data collected during each fieldwork round are not restricted to key demographic events but may also include the various attributes of the primary subjects. These attributes may be fixed (for example, ethnicity, gender) or changing over time (for example, marital or residential status).

Unique identifiers

Unique identifiers for primary subjects are an indispensable element of DSSs. All systems invariably formulate rules for assigning unique identifiers at the start of the DSS, but their methods for assigning these identifiers to DSS subjects may vary from one site to another. There are two main approaches. One common strategy is to transparently link the subjects in a single residential unit through a hierarchical system of unique numbers. These are built up from a unique number for the residential unit, followed by serial numbers for each of the households within it (where the notion of households applies) and then for each of the enumerated individuals within each household. In this system, the unique number for each individual in the DSS is a composite of the numbers for the residential unit, household, and household member. This may involve creating complex hierarchies, in which the unique number of the residential unit itself is a composite reflecting allocation to regions, areas, and villages (where they exist). This system requires thorough mapping of the DSA before enumeration. It also requires proper training of enumerators to avoid confusion in assigning identifiers. When mapping of the DSA is coupled with georeferencing of residential units, using geographic information system (GIS) technology, global positioning system (GPS) coordinates are assigned as location attributes of the residential units within the database.

The other strategy for assigning identifiers to individuals is to avoid any fixed link to residential units and households. In this system, identifiers for each subject are simply serial numbers incremented each time a new DSS subject is registered. This system requires providing field staff with block allocations of ID numbers with enough latitude to register new subjects. This approach should be coupled with computer generation of the identifiers to safeguard against the assignment of the same ID to multiple subjects on the ground. This strategy helps to preserve people’s anonymity outside their residential units, or when their attribute data are accessed through the database.

Primary DSS subjects

DSSs are typically structured around three main subjects (Figure 1.1) within the DSA. These subjects have both a conceptual and a logistical rationale. From a logistical point of view, it is not feasible to interview all individuals directly, and for this reason individuals are put in groups with physical and social meaning, and information is collected from credible and informed respondents within these groups. The reasons to distinguish between these subjects from a conceptual point of view will be dealt with in greater detail in the following subsections. The three main subjects are (Figure 1.1) as follows:

Figure 1.1. The three main DSS subjects.

Residential units

All DSSs identify residential units as a primary subject of interest, although they vary in the terms they use for these units (for example, compounds or homesteads) and may also differ slightly in their definition of them. Residency, or physical presence within a DSA at a fixed place of abode and for a sufficiently long period, is an essential prerequisite for the enumeration of individuals at risk for demographic events or disease exposure.

In most systems a distinction is made between places of residence and other structures, such as clinics, schools, churches, and stores. Identifying a unifying term for all these structural units may have conceptual merit, and some systems have attempted to do this, as these structural units share many characteristics and this approach simplifies the database hierarchy for handling this concept. In this system an inclusive term such as bounded structure may be used at a higher level and compounds (or homesteads) and facilities at the more specific level.

Households

The definition of household and its applicability both as a concept and as a separate DSS subject may vary greatly from one DSS to another. Households may simply be seen as fixed social subunits within residential units. In more complex systems, they may be seen as independent subjects able to change their place of residence while preserving their social identity, and they may have members who are resident elsewhere. In such a system, a clear distinction would be needed between residency, which defines the state of being physically present in a given residential unit for a defined threshold of time, and membership, which defines the state of belonging to a social group irrespective of physical presence. These concepts have a clear overlap with the related concepts of de facto population (persons who are physically present in a place) and de jure population (persons who usually reside in a given place), respectively. The concepts of residency and membership are discussed later in this chapter.

Individuals

The individuals are people of various ages, sex, and other personal characteristics who are residents or members of the DSS residential units or households, respectively. Their personal characteristics may be fixed (sex, date of birth) or change over time (age, marital status). Unless their changes are predictable (like the yearly increment of age), changing characteristics will need to be recorded repeatedly — or their changes will need to be recorded as events — to produce longitudinal trends.

Eligibility

Every DSS is required to define the population under surveillance. As most individuals within any population have places of residence and attachments to social groups, the task of defining the population begins with the identification of the residential units, households (where applicable), and individuals that will be visited and observed. Thereafter, a set of inclusion criteria must be applied to distinguish eligible from ineligible individuals or subjects within each subject category.

As residential units have fixed geographical positions in all DSSs, there are consistent and simple rules for their inclusion: they are included if they are situated in the DSA. In DSSs that deal with households as distinct (and potentially mobile) subjects, these households are eligible if (and while) they are situated in the DSA. This is what is referred to as household residency.

Rules for individuals, particularly in highly mobile populations, are more complex. The most typical approach is to simply base their eligibility on residence, that is, physical presence. Individuals are eligible if (and while) they are resident at eligible residential units. This is what is referred to as individual residency. Another approach,

based on social linkages, rules that individuals are eligible if (and while) they are members of eligible households. This requires careful and consistent definitions of household and membership and can allow individuals who are not resident to remain as members of the household and therefore to qualify for observation.

Residency and membership

Clear geographical boundaries for the DSA and well-defined physical boundaries for residential units are minimal prerequisites for following up DSS subjects consistently and arriving at numerators and denominators for rate calculations. In systems where residential units and households are separate subjects and there is a separate relationship between individuals and each of those subjects — expressed as residency and membership, respectively — these concepts become substantially more complex.

Observing an individual’s presence in, or absence from, a specific residential unit requires clear rules for residency status. The physical presence of an individual for a very short time may not be taken into account when the amount of time spent in the residential unit is computed. Conversely, the noncontinuous presence of an individual, with short periods of absence, may be considered continuous residency if he or she meets a threshold for inclusion.

Residency and membership statuses are assigned at the start of the DSS, based on prescribed eligibility rules. Thereafter, new residency episodes may commence as a result of births or in-migrations exceeding a prescribed threshold of duration, and current residency may end because of deaths or out-migrations, again exceeding a prescribed threshold of duration. New membership episodes may commence as a result of events that initiate a social relationship with a household, such as birth, marriage, adoption, or household formation, and may be terminated by events that end such a relationship, such as death, divorce, or household dissolution.

Core DSS events

To know the size of the registered resident population at any time, a DSS collects information about three core events that alter this size, namely, births, deaths, and migrations. These events are described by the following fundamental demographic equation:

where P is the population; B is the number of births; D is the number of deaths; I is the number of in-migrants; O is the number of out-migrants; and t₀, t₁ is the time interval of their occurrence.

An underlying principle for recording events in a DSS is that of a population at risk. Mortality, fertility, and migration rates are calculated by counting the number of deaths, births, or migrations occurring within a registered population exposed to the risk. For example, an individual who is not resident within the DSA is not considered at risk of dying within the area. Consequently, most DSSs do not observe nonresident individuals or households and do not record their events.

Births and fertility

Pregnancies and their outcomes for all women registered in the DSS are recorded regardless of the place of occurrence of such events. The recording of births has two purposes: for estimating fertility and for identifying a criterion for registering an individual. To estimate fertility, a DSS should record all pregnancy outcomes, including miscarriages (<28 weeks), induced abortions, stillbirths (≥28 weeks), and live births. All live births are then registered as individual members of the DSS, independent of subsequent survival. In some DSSs, fieldworkers take note of live births to visitors to the DSA to alert the data collector in the next round to register the mother (if she becomes eligible) and her child. This procedure is very helpful, as it greatly improves the accuracy of dates of birth of newly born babies and increases reporting of births from eligible mothers with frequent in- and out-migration.

Although most DSSs will report their estimates of the fertility of a specific age group of women, usually 15–49 years, they should also record births to women outside this age group.

The underreporting of pregnancies and their outcomes is a major problem across all DSSs. Some DSSs have used the recording of pregnancies during routine update visits to improve birth coverage. Pregnancy observation has also been used to increase the reporting of other pregnancy outcomes, particularly miscarriages, induced abortions, and stillbirths. However, this requires an update-visit interval of <5 months so that a notification of pregnancy can be obtained in one round, followed by the recording of the pregnancy outcome in the next visit.

Deaths and mortality

Deaths of all registered and eligible individuals are recorded, regardless of the place of death. It may be impossible to record the deaths of previously eligible individuals who then out-migrated. In this case, observation of their survival is censored at the time of migration. Information about the death of visitors to the DSA is sometimes collected, but it is only used in mortality estimates if a de facto population estimate is available for each day.

Underreporting of deaths is typically less of a problem than that of births, because a death is widely known and remembered. Exceptions are the deaths of young (and yet unregistered) infants, particularly perinatal deaths, if cultural beliefs or grief hinders reporting.

Some DSSs collect more detailed information about deaths to establish the cause of death, generally through the so-called verbal autopsies (VAs).

Migrations and mobility

Where nonresident household members are ignored, only external migration affects the size of the population, resulting in either the registration of a new in-migrant or the termination of follow-up of an out-migrant. However, recording internal migration is very important to ensure the accuracy and validity of DSS data. The DSS needs to identify internal migrations and migrants and collect supporting information to avoid double counting of individuals and to ensure that their exposure to the social and physical environment is correctly apportioned. Migrations influence the registration of births and deaths; for example, a death would not be recorded for an individual who out-migrated before his or her death.

Defining the circumstances under which a migration is acknowledged to have occurred is notoriously difficult, not only for DSSs, but even for vital-registration systems and censuses. Different DSSs have different criteria. One approach, generally known as the “50% rule,” considers individuals resident if they have spent most of the time between two data-collection visits within the DSA. Any former resident who has not spent at least 50% of the time in the DSA would be recorded as having out-migrated.

However, many rural communities have individuals who regularly and predictably change residence for seasonal work, employment, or educational opportunities. The terms circular and pendular migration are often used. In the Hlabisa DSS, a newly established system in an area of very high population mobility, individual residency has been replaced with household residency as a registration criterion. Consequently, although out-migrations are recorded, the fieldworkers do not automatically terminate follow-up observations.

Migration is a repeatable event — an individual may make several migrations over time, both internally and externally. To maintain longitudinal integrity of data concerning individuals, a DSS should establish whether an external in-migrant has previously been registered in the DSS. The individual’s current and previous records should be matched so that he or she is not handled as a new individual in the system but as an individual under observation for several periods.

Episodes

Episodes are a logical complement to events. They are meaningful and identifiable segments of time started and ended by events. The life of an individual, for instance, can be understood as an episode that started with the individual’s birth and ended with his or her death. In the same way, residential units or households can be said to be episodes that start when they are formed and end when they are dissolved.

The usefulness of the concept of episodes is not limited to primary subjects. It applies equally to associations between them and therefore provides a useful framework for handling residency, membership, marital status, and many other concepts. Episodes also make it much easier to formulate and implement validation rules regarding events.

Other events

In addition to births, deaths, and migrations, other events are of interest for our understanding of demographic, health, and social dynamics. One event on which data are commonly collected relates to nuptiality or marital status. Most DSSs collect information about events such as marriage, defined as an event that starts a marital relationship, and divorce, that is, an event that ends a marital union. Other events recorded by DSSs depend on their complexity and research interests but may include the change of a head of household, a household’s formation or dissolution, or the construction or destruction of building structures.

Nuptiality and conjugal relationships

DSSs collect data on nuptiality primarily because of the important influence of marital patterns on fertility. Marriage as a start of an episode is easily identified, although a period of sexual union may have preceded marriage. The ending of a conjugal relationship can be less clearly marked, because it may not always be the death of one of the partners or a divorce, but a period of separation. In DSAs where the nonmarital fertility rate is high, other conjugal relationships become important, and the systems record informal relationships as well as formal marriages. However, in taking on this broader approach to sexual relationships, the DSSs must overcome two hurdles:

Construction and disintegration of residential units

At any given time, new residential units may be under construction and other residential units may be at various stages of disrepair following natural disasters or abandonment. The physical state may be distinct from the functionality of the residential unit; that is, it is possible that a residential unit is physically intact but long abandoned, and apparently broken-down units may still have households and individuals living in them. It is also possible that broken-down or destroyed units may subsequently be rebuilt, when the owner returns.

As the state of the residential unit is often — if not always — a good indication of its functionality, a DSS should make provision to track both its physical state and function.

Events occurring in households

Similarly, households can go through important changes affecting their composition and socioeconomic and health conditions. New households may form within an existing residential unit when, for example, a son takes a wife and establishes a family of his own or when a polygynous man takes another wife. Separate households may merge to form a new household, or a complete household may move to settle at another residential unit. Households may lose one or more members over time and decrease in size, or they may completely dissolve through a process of slow attrition or a major environmental or social disaster.

In environments with substantial social flux and instability, it is important to keep track of these events and their effects on the formation and dissolution of households. This is essential if DSSs have conceptualized households as subjects in their own right. Because they also influence patterns of individual presence at a residential unit, these household changes have important implications for the composition of the residential unit as a whole.

Chapter 2
DSS-GENERATED MORTALITY RATES AND MEASURES

Introduction

This chapter provides definitions and explanations of key DSS-generated mortality rates and measures, as well as describing the methodology employed in calculating them. It is intended for readers unfamiliar with these rates and measures. Their calculation is basic, and the various formulas can be found in standard textbooks (see for example, Shryock and Siegel 1976; Kpedekpo 1982; Newell 1994). These measures have been briefly discussed in this chapter for quick reference, as they form the basis for standardizing the results across DSS sites. Perhaps the most important reason for discussing them is the opportunity it affords to discuss the classic controversy over whether to define some of them as rates or ratios (for example, infant mortality, under-five mortality, and maternal mortality). Furthermore, this chapter provides an explanation of the need for a standard population and introduces the INDEPTH standard population for Africa south of the Sahara, discussed in greater detail in Part II.

Rates and ratios

Rates and ratios are frequently used in measuring demographic events. Rate refers to the frequency of events. A rate is estimated by taking the number of events in a given period and dividing it by the population at risk during that period. Pressat (1985, p. 194) stated that the term rate

A ratio is the proportion between a numerator and a denominator that are related (for example, under-five child deaths per 1000 under-five person–years lived in a given year).

Crude death rate

The crude death rate (CDR) is defined as the number of deaths in a given period divided by the total population. Although the CDR can be computed for any segment of time, the period usually used is a year, and the denominator used in the rate calculation is the midyear population. The midyear population is the size of the population (or any specified group within the population) at the midpoint of a calendar year. This midpoint is often calculated as the arithmetic mean of the size of the population at the beginning and end of the year. Conventionally, the rate is expressed as a number per 1000 individuals.

In the case of a population under continuous surveillance, with possibly high in- and out-migration rates that may yield a strong variation in population size, the use of exact person–years lived is preferred. Person–years is the sum, expressed in years, of the time spent by all individuals in a given category of the population (Pressat 1985). Specifically, these years express the periods that eligible individuals spent in the DSA. Times or periods spent outside the DSA due to migration or death are excluded.

Age-specific death rate and ratio

Because of the differentials in exposure to the risk of dying, epidemiologists and demographers often use age-specific death rates (ASDRs) and sex-specific death rates, instead of the CDR. ASDRs are the most commonly used. The ASDR for an age group is defined as the number of deaths in the age group in a specific period divided by the total number of person–years lived in that age group during that period and multiplied by 1000. Demographers often use a slightly different notation. They express the ASDR of a particular age group as the deaths among individuals in that age group in the year, divided by the mid-year population of that age group and then multiplied by 1000. Five-year age groups are common, although age categories vary according to the purpose of study.

The following discussion of infant, under-five, and maternal mortality measures highlights the classic controversy over whether to define these measures as rates or ratios. The denominator used in calculating a measure determines whether it is a rate or a ratio. As stated earlier, the measure is a rate when the total number of individuals at risk is used as the denominator, and it is a ratio when some other event is used as the denominator.

Infant mortality

It is usually difficult to estimate the number of person–years lived for children <1 year old (infants). Consequently, the total number of live births is often used as the denominator to calculate the infant mortality rate. The total number of deaths among children <1 year old in a calendar year is divided by the live births in the same year, multiplied by 1000. Calculating the infant mortality rate in this way makes it more appropriately referred to as a ratio.

Infant deaths are unevenly distributed through the first year of life. A high proportion of infant deaths usually occurs in the first month of life. Of these deaths, a high proportion occurs during the first week of life; and of these, a high proportion

occurs during the first day. The conventional infant mortality rate or ratio may usefully be broken up into rates or ratios covering the early stages of life and a rate or ratio for the remainder of the year. The one for the first period is called the neonatal mortality rate or ratio, and that for the second period is called the postneonatal mortality rate or ratio. These concepts are briefly defined in the following paragraphs.

Neonatal mortality is defined as the number of deaths of infants <4 weeks old (or <1 month old) during a year. It is calculated by dividing the deaths of infants <28 days old during a year by the live births in the same year and multiplying by 1000. Early neonatal mortality is calculated by dividing the deaths of infants <7 days old during a year by live births in the same year and multiplying by 1000. Late neonatal mortality is calculated by dividing the deaths of infants 7–28 days old in a year by live births in the same year and multiplying by 1000. Postneonatal mortality is calculated by dividing the deaths of infants 4–51 weeks old during a year by live births in the same year and multiplying by 1000.

Infant mortality can also be expressed as a probability of dying before reaching the age of 1 year. Perinatal mortality is calculated by dividing the sum of stillbirths in the year and the deaths of infants <7 days old during the year by the sum of stillbirths in the year and live births in the same year.

Under-five mortality

Some consider the under-five mortality as a ratio expressing the number of deaths of children <5 years old divided by the number of live births in a year and then multiplied by 1000. Others treat it as a rate, calculating it by dividing the number of deaths of children <5 years old by the total number of person–years of children <5 years old and multiplying by 1000. When under-five mortality is presented as a probability of dying before age 5, it is expressed as ₅q₀.

Maternal mortality rate and ratio

Most DSSs record all pregnancies and their outcomes as well as deaths. As such, they have the potential to provide accurate, up-to-date estimates of maternal mortality rates and ratios. The maternal mortality ratio is conventionally defined as the number of deaths due to puerperal (pregnancy-related) factors per 100 000 live births. But strictly speaking, this is referred to as a ratio because the denominator is not the persons at risk of experiencing the event. In view of this, the following are the methods for estimating maternal mortality ratios and rates. The maternal mortality ratio is calculated by dividing the number of pregnancy-related deaths in a specified period by that of live births in the same period and multiplying by 100 000. The maternal mortality rate is calculated by dividing the number of pregancy-related deaths in a specified period by person–years lived by women of childbearing age and multiplying by 1000.

Maternal mortality can also be estimated by relating maternal deaths to women of reproductive age or to all pregnancies, including stillbirths and abortions.

Standardization

Age-standardized death rate

Crude mortality rates are inappropriate for comparing different populations within the DSS sites because of the different age structures within the sites. On the other hand, a single parameter is required for simple comparison. Therefore, standardized rates are used, in which the age-specific mortality rates are combined using a standard population. An INDEPTH standard population for sub-Saharan Africa (SSA) has been developed (see Table 6.2). More details on the INDEPTH standard population are provided in Chapter 6. The Segi (1960) and the new World Health Organization (WHO) standard age distributions are also shown in Table 6.2.

Age-specific rates are weighted averages of rates, where the weights are obtained as a proportion of the standard population in the respective age group. The summation goes over all age groups.

Confidence intervals for rates

Estimates of the mean and standard deviation of a population are usually needed if it is impossible to deal with the entire population. The standard deviation of a distribution of sample means is referred to as the standard error of the sample. It measures how precisely the sample mean estimates the population mean. For example, with a 95% confidence interval, about 95% of the sample means obtained by repeated sampling would lie within two standard errors below or above the population mean. Based on the sample mean and its standard error, a range of likely values can be constructed for a population mean that is not known. This range is referred to as a confidence interval. More precisely, there is a 95% probability that a particular sample mean lies within 1.96 standard errors above or below the population mean.

Confidence intervals can be calculated for the ASDRs. The variance of the CDRs or the ASDRs is used instead of the means. Estève et al. (1994) discussed the method in detail. For a small number of deaths or for small populations, however, confidence intervals for ASDRs are not reliable, because the formula used to calculate them is too imprecise. The question is then one of how large the numbers of deaths and populations must be to give reliable results. It is difficult to supply a rule of thumb, and as Estève et al. (1994, p. 58) noted,

Chapter 3
DSS METHODS OF DATA COLLECTION

Introduction

Knowledge of the methods for collecting or compiling data at the DSS sites is essential because these methods influence the ways that data are processed, analyzed, and interpreted. The most common demographic methods used in data collection are censuses, sample surveys, and vital-events registration systems. The last method, however, is nonexistent or only partially applied in many developing countries. Given the paucity of vital-events registration and knowledge on population or health-status trends in such settings, demographic and health surveys have been introduced for health planning, practice, evaluation, and allocation of resources. Demographic estimates undertaken in developing countries have employed both indirect and direct methods, using retrospective single-round surveys and prospective multiround ones (Tablin 1984).

Indirect estimation methods rely on information obtained from subjects not directly at risk of a particular demographic phenomenon. The indirect methods can be used to estimate levels and trends of fertility, mortality, and migration where data sources are defective or incomplete. An example of an indirect method is the estimation of infant and child mortality from proportions of surviving children or the estimation of adult mortality from those orphaned. Indirect estimation methods are also used to assess data collected using conventional methods. Such data are compared with other information to infer a certain pattern, on the basis of certain assumptions. If this pattern is reproduced then data can be further inferred. Indirect estimation may, in addition, involve fitting of demographic models to fragmentary and incomplete data (Pressat 1985). The results obtained are used to estimate a particular parameter.

Direct methods use data on the people at risk to establish a demographic measure and pattern. These methods rely on data obtained from censuses, surveys, and recorded data on the components of change — that is, births, deaths, and migration. Data obtained from these methods are used directly to provide estimates of demographic phenomena, such as fertility, mortality, and migration. An example of a direct method is the use of the number of children born to women of a particular age group to estimate age-specific fertility rates.

In single-round surveys, a population is enumerated once during a survey, and retrospective data are gathered on past events (Kpedekpo 1982; Tablin 1984; Newell

1994), such as a birth or death that occurred in the last year (or a life and maternity history). This method may result in overestimation or underestimation of events, as a result of memory lapse. Respondents may exclude events from the reference period. It has been argued that an underestimation of 30–40% is likely using this method (Tablin 1984). Some examples of single-round surveys are the World Fertility Survey and the Demographic and Health Surveys.

Prospective surveys involve repeat visits (longitudinal data collection) to the same respondents or the same study area (Pressat 1985). All DSS sites employ this method of data collection. This does not mean, however, that the methodological approach is the same across all sites. Sites each have unique features, as shown in the various site chapters of this monograph. The purpose of this chapter is therefore to provide a general description of the data-collection methods used by the DSS sites. The data-collection methods are described to provide a quick reference for the reader, rather than describing experiences with data collection. Periodically, specific examples are provided from sites for clarification.

Establishing the monitored population

Selection and establishment of the DSA are prerequisites of any DSS site, but no specific sampling method has to be employed in the selection of an area. Depending on the nature of the study, sites employ probability or nonprobability sampling methods, or both, in drawing their sample population. Once an area has been selected the community has to be mobilized to prepare it to participate in the research and ensure its compliance. Mobilization activities involve conducting sensitization meetings with influential opinion leaders, such as councillors and village, hamlet, or religious leaders. During these meetings, the DSS staff presents and clarifies the project’s objectives and expected output and outlines its anticipated activities. Other sensitization methods include drama and sports activities involving the project staff and the community.

As DSSs are longitudinal studies, staff also have to maintain the community’s compliance with DSS activities longitudinally, and this means that mobilization of the community is not limited to the initial stages but has to be a continuous process. Compliance is maintained in a variety of ways across sites, including giving feedback to the community through presentation of results in simple tables or graphics, production and circulation of a newsletter, meetings with the key informants at regular intervals, and presentations of findings to health-management teams.

In terms of the minimum and maximum population size under DSS, there is no consensus. DSS sites can have a variety of population sizes under surveillance. For example, Butajira DSS (Ethiopia) began with a sample of 28 616 people (Berhane et al. 1999), whereas Navrongo DSS (Ghana) and Rufiji DSS (Tanzania) had, respectively, 124 857 and 85 102 people 1 year after they began operations (Binka et al. 1999; Mwageni and Irema 1999). The Adult Morbidity and Mortality Project (AMMP, Tanzania) has three sites and more than 300 000 people under surveillance (TMH 1997). The site chapters give more details on the sample sizes of the various DSS sites.

Planning for data collection

Any data-collection exercise requires advance planning and recruitment and training of field staff, such as enumerators and supervisors. It also involves the designing and printing of DSS forms and the preparation of field or training manuals. DSS enumerators are normally recruited from among those local individuals who meet minimum qualifications set for specific projects. Training focuses on proper ways to use DSS forms, conduct interviews, and handle various field forms. Field or interview manuals are used for training and are eventually provided to all field staff as reference materials during data collection. The training manuals clearly indicate the duties and responsibilities of the field staff. In addition, the staff may receive training on how to use or operate field equipment, such as motorcycles. The field staff are given periodic training on field operations to keep up to date on data-collection techniques.

Initial census

Data collection to establish the baseline population begins with a census, conducted by trained enumerators living in the study area. As stated earlier, they are trained on how to use DSS forms and conduct interviews. The initial census establishes the foundation for a longitudinal surveillance system and helps obtain background data on the subjects. Data are collected using standard questionnaires, with closed- or open-ended questions, or both. Separate questionnaires are used to collect household and individual data. The structured questionnaires comprise at least two sections: the header, for recording the unit of interest; and the main part, for recording basic information (see example 1 in Appendix 1).

The type of data collected during the initial censuses depends on the specific objectives of the site. In many sites, data are collected on variables such as household composition (household head, relation to household head, etc.), culture (religion and ethnicity), demographic data (age, sex, marital status), and socioeconomic data (education, occupation, etc.). In addition, the DSS can collect data on behavioural issues (alcohol consumption, smoking, etc.), housing, health-care use, and environmental conditions (source of drinking water, sanitation facility, etc.).

For identification purposes, each household and individual registered is assigned a unique number within its village and his or her household, respectively. A series of numbers for each individual may be used to identify the village, the household, and the individual within the household. The number allocated to the individual is permanent. In some systems, if an individual moves to a new area, the number is still used to identify that person. In this way, it is possible to monitor migrants, as will be shown.

Update rounds

The longitudinal system of data collection continues then with periodic visits to registered households. The purpose of the visits is to record vital changes or events since the previous visit. These may include births or other pregnancy outcomes, marital status (marriages, divorces, separations, reconciliations), deaths, and migrations. Regular data collection is undertaken to maintain accurate denominators for estimation of

age-, sex-, and cause-specific death rates. The DSS approach has no specific interval for periodic visits to the registered households (Indome et al. 1995). Yet, it is important to ensure that the interval chosen between interview rounds is consistent for any given household or area. Provided they are consistent, periodic-visit cycles may range from 1 to 12 months.

During the periodic visits or updates, the status of each individual is verified using the household-registration or -record books (see example 2 in Appendix 1) or forms. The registration books are computer printouts of information on households and their members collected in the initial census. They are systematically arranged by household to facilitate further visits or household contacts. These books can be printed in rows and columns to maintain several rounds of data collection. The information on rows may correspond to individual members, as well as details of a household, whereas the columns have spaces for filling in vital events detected in each DSS round. However, all vital events have to be registered on specific event forms. These forms may include observation of pregnancies, births, deaths, and marital changes (see examples 3–5 in Appendix 1). These are forms used in the Butajira, Navrongo, and Rufiji DSSs.

All errors that the interviewers note during update rounds they correct accordingly in the respective book, along with filling out the changes form. The changes form requires the unique number of the household or individual, the change to be made, the original information, and the correction. Corrections that may require filling in the changes form include those for age, name, sex, missed members of a household, and relationship to the head of household. Eventually, these forms are taken to the data centre for correction of databases. This means that in DSS sites data are collected in conjunction with data-management operations (details on data management are provided later in this monograph). In most cases, the fieldwork and computer cycles coincide. Figure 3.1 summarizes the linkage between field and computer operations in Rufiji DSS. This linkage aims at maintaining the integrity of data, as well as ensuring timely reporting of findings. Upon completion of interviews in the household (during the initial census or updates), the forms are taken to the computer centre for data entry. Errors noted during quality control (for details, see Chapter 5) or data entry are verified, reported to the field staff for diagnosis, and later corrected in both the household-registration book and the computer databases.

Updating of vital events is not the only activity carried out during these periodic visits. During update rounds, enumerators register new people or households. These include the migrants, the newly married, and any individuals missed during the initial census. The longitudinal system allows individuals to enter or exit the DSS at any time. They enter through births or in-migration and exit through deaths or out-migration (Figure 3.2). As these individuals are under surveillance, it is possible to estimate the total time spent by each individual in the study population. This time contribution is called person–years of observation and is used as a denominator to estimate rates of events (such as fertility, mortality, and migration). Details on the uses of person–years of observation appear elsewhere in this monograph.

The periodic visits to registered households make DSS self-checking, allowing data collected in one round to be checked and corrected in successive rounds. This reduces the risk of omitting, forgetting, or misreporting variables or events. During the rounds it is also possible to select subsamples (nested studies) on which to collect

data on specific items at marginal extra cost and without disturbing the original purpose of the surveillance. However, where the population is very mobile, a major problem of multiround surveillance is tracking subjects.

Figure 3.1. The linkage between field and computer operations at the Rufiji DSS site, Tanzania.
Source: After Binka et al. (1999). Note: HRB, household-registration book.

Figure 3.2. Prospective monitoring of demographic events.
Source: After Berhane et al. (1999).

Recording demographic events

Monitoring of births and deaths in developing countries is very crucial, as these two events are easily omitted from routine statistical records and systems (Binka et al. 1999). This can lead researchers to underestimate their occurrence. A good recording system is needed to capture such events. Key informants can do this. Key informants are usually senior or respected members of the community (such as village or hamlet leaders) within the DSA. Key informants fill in their registers whenever an event has occurred, and they report this to the supervisors who visit them on regular basis. Ideally, being part of the community themselves, these people should not be individuals who have to find out about these pregnancies, births, and deaths but those who would hear about them in their course of normal life. As an incentive, a common practice is to pay key informants token fees for reporting such events, once they are confirmed by the system. An example of the system for recording events, as practiced in the Rufiji DSS, is summarized in Figure 3.3.

Figure 3.3. Vital-events reporting system at the Rufiji DSS site, Tanzania.
Source: After TEHIP (1996).

In the vital-events reporting system of the Rufiji DSS, key informants observe and record any birth or death occurring in the study area. This information is passed on to the DSS key-informant supervisor (or enumerator, who informs the key-informant supervisor). Within 2 weeks, the key-informant supervisor visits the households where a birth or death has been reported and contacts the data centre for verification of the event. If the information is correct, the key informant is paid a token fee. The key-informant supervisor then administers a verbal autopsy (VA) with one of the deceased’s relatives (who is well informed of the trend of illness of the deceased) for all reported deaths. Enumerators also check births and deaths during fixed enumeration rounds.

Monitoring mortality

Documentation of causes of death has contributed to progress in knowledge of epidemiology and public health. Such documentation allows researchers and policymakers to assess the health status of a population, assign health priorities, study time trends in mortality from specific causes, and evaluate health interventions. Documenting deaths is a common practice in developed countries, where most deaths occur in a medical environment, postmortem autopsies are both feasible and culturally accepted, and vital-events registration is mandatory and complete. In developing countries, however, many deaths occur in the home, with limited or no medical attendance, and postmortem autopsies are rarely possible or complete and vital-events registration is impractical. To assess the cause of death, one must rely on an alternative source of information, that is, an attending relative’s description of symptoms and events preceding death.

The VA is an indirect method employed in DSS sites to ascertain the causes of death from close associates whom the DSS interviewers question regarding their knowledge of the symptoms, signs, and circumstances leading to the death. Retrospective interviews of individuals who were there and can describe what happened during the hours, days, or months preceding a death are done, and then a most likely cause of death is inferred from the sequence and combination of symptoms and events. Specially designed forms (questionnaires) are used to suit the population of interest (TMH 1997). For example, if the study of interest is the mortality patterns of children <5 years old, then a form is designed and structured to cover all signs and symptoms of illnesses that affect mostly children of this age (see example 6 in Appendix 1). There are also special interview forms for deaths of children <31 days old and for deaths of those ≥5 years old. The DSSs use trained medical personnel or laypeople to conduct VAs.

VAs are used in health-care projects involved in research and evaluation of health services. As earlier described, key informants record deaths that occur in their area in a mortality register; this is reported to the interviewers who will conduct the VA. The interviewers make appointments to visit the houses of the bereaved families. On the appointment day, an interviewer visits the house and administers a VA with the caretaker or a close family member of the deceased. The VA questionnaires are designed to suit the settings of the area under surveillance (TMH 1997). Such information as name, age, sex, occupation, and other risk factors is usually collected, in addition to an open history of events leading to the death, previously diagnosed medical conditions, and signs and symptoms that appeared before death. The interviewer

can use the questionnaires to record information on use of health facilities before the death, reasons for using or not using a particular health facility, the caretaker’s perception of cause of death, and confirmatory evidence of a cause of death (if available). The cause of death is determined from a combination of these signs and symptoms.

Causes of death from the VA questionnaires can be reached by either asking physicians or using computer algorithms, depending on the design and structure of the questions. If physicians are asked to do this, then usually two physicians independently code the VA forms and determine the cause of death, using some kind of agreed classification (for example, the WHO International Classification of Diseases [ICD] for causes of diseases). In the case of discrepancies, a third physician is asked to code the forms. Computer algorithms are based strongly on the checklist of signs and symptoms recorded on the form. If discrepancies are noted at this level, then the cause of death is categorized as unknown. Discrepant VA forms produced by the algorithm are taken to physicians for diagnosis and coding. Usually, forms with discrepancies are fewer than others.

Tracking migrants

Migration is a complex subject, with a variety of definitions (Pressat 1985; Newell 1994). As such, the definition relies more on the way data are collected and the purpose for which they are collected. Generally, migration refers to movement of people (groups or individuals) that involves a permanent or temporal change of their usual place of residence (Pressat 1985). Migrants are therefore people who change their usual place of residence. According to Kpedekpo (1982), classification of migrants can be based on the following criteria:

Data on migration can be collected in several ways. Censuses, sample surveys, and continuous population registers are the most common (Shryock and Siegel 1976). Censuses and surveys can provide migration data directly (by asking questions about, for example, the number of moves, duration of residence, date of exit or entry, and previous residence) or indirectly (by estimating migration from total counts of population and natural increase of two censuses or counts). The problem with these methods is their failure to detect multiple moves or those that people cannot remember. In addition, past migrants are grouped together with most recent ones. Also, the indirect method requires very accurate data for the two censuses.

A migration history is another way to collect data on migrants. DSS sites collecting migration data employ this method. This is a continuous way of giving data on previous residence of individuals with dates of their moving out and in. In this way, migrants are linked to the database. Special in- or out-migration forms are used to track down migrants (see examples 7 and 8 in Appendix 1). The in-migration form requires more details than that for out-migration. In addition to personal particulars

of an in-migrant (sex, date of birth, education, occupation, etc.), information on the date of and reasons for the migration and the place of origin are also gathered. If in-migration involves a household, a household questionnaire is also used to record household characteristics. On the out-migration form, information is recorded on the date of and reasons for the migration and the destination.

DSS sites do not record all the moves but only those within a certain period. For example, the Navrongo DSS considers an individual an in-migrant if this person is in the same place of residence for 3 months (Binka et al. 1994), whereas Rufiji DSS uses a 4-month criterion for the same purpose (TEHIP 1996). The opposite applies to an out-migrant. The purpose of setting these criteria is to find a proxy to determine the residency status of individuals. This status enables estimation of the individual’s overall time contribution to supply denominators for calculation of other demographic measures, such as mortality and fertility.

Additional rounds of data collection

The previous sections have focused on collection of data for demographic variables — mainly, births, deaths, and migrations. All these can be considered extradynamic events, as they change frequently within a year. Other variables are constant or change slowly, such as socioeconomic aspects like education, occupation, housing conditions (floor, roofing material), health-care use (like vaccination), and environmental conditions (like source of drinking water and sanitation facilities). Such information can be collected once in a year, preferably at the beginning of each calendar year.

A DSS can have other nested studies to capitalize on its population database and organizational infrastructure. Such studies employ a variety of designs, such as cohort, cross-sectional, and case referent, depending on the specific primary purpose of each study, and these studies are usually linked to the longitudinal surveillance system. The Butajira DSS, for example, used its database as a sampling frame for a study population and used the routine surveillance to follow subjects in various studies of acute respiratory infections (Berhane et al. 1999). In Tanzania, a new study aimed at monitoring a program for antimalarial combination therapy uses the Ifakara, Morogoro (AMMP), and Rufiji DSAs.

Such nested studies in the DSS sites take advantage of the existing infrastructure and field organization for data collection. Sometimes these new studies may employ supplementary personnel trained to collect information specific to each study. As a result, many DSS sites become pools of trained field staff.

Geographic information systems

A GIS is a computer-assisted information-management system for geographically referenced data. It integrates the management (that is, acquisition, storage), analysis, and display (mapping) of geographic data (Loslier 1995). The GIS contains two integrated databases, namely, spatial (location information) and attribute (characteristics of the spatial features). The spatial database comprises digital coordinates obtained from maps, using GPS. These coordinates can take a variety of forms, such as points (dispensaries, hospitals, schools, households), lines (roads, railways, rivers), or polygons (wards, towns, villages, hamlets). The attribute database can include information such

as population size or density and number of health facilities or personnel. The GIS can create a link between spatial data and their associated descriptive information. Its strength lies in its capacity for integration and analysis of data from many sources, such as population, topography, climate, vegetation, transportation network, social services, and epidemiological characteristics.

Many DSS sites use GPS to determine locations and boundaries of phenomena of interest, including boundaries of settlements, households, and villages, and to map health services in terms of access and coverage. Thus, Navrongo DSS used GPS coordinates to assess the child-mortality impact of insecticide-treated bednets in 96 clusters of contiguous compounds (Binka et al. 1996). The data collected using GPS are joined to spatial imagery with GIS. In this way, it is possible to combine and analyze the occurrence of features with various locations. Nouna DSS in Burkina Faso has a GIS with data on all households in 49 villages and information on such features as health facilities, sources of water, roads, schools, and religious places (Sauerborn and Kouyaté 2000¹).

Conclusion

This chapter has presented a general picture of the major data-collection activities at the DSS sites. The data-collection process has been presented in terms of sequence of events carried out in DSS sites. It discussed the people involved in data collection and the tools used in obtaining information. (Part III will describe specific data-collection methods the DSS sites employ, including sampling procedures, type of information gathered, and key functions and responsibilities of the staff.) This chapter has also shown the potential of DSS sites to contribute reliable demographic and health-related data. Given developing countries’ lack of complete vital-events registration systems and the costs of and long intervals between national censuses, the DSS approach is probably one of the best options for improving the quality of data. The DSS data-collection procedures are linked to data-management and quality-control procedures, which are the items discussed in detail in the next two chapters.

¹ Sauerborn, R.; Kouyaté, B., ed. 2000. Nouna Research Centre, a platform for interdisciplinary field research in Burkina Faso, West Africa. Internal report.

Chapter 4
PROCESSING DSS DATA

Introduction

Compiling longitudinal population information poses unique data-management challenges. Projects must maintain changing individual-level information on the composition and household structure of a large, geographically defined population. Events that arise — births, deaths, migrations, etc. — must be linked to individuals and other entities at risk of these events. These events affect not only demographic rates, for instance, but also relationships within and between households. As event histories grow, records of new events must be logically consistent with those of events in the past. Seemingly obvious checks on data to meet minimal standards of integrity can result in hundreds of lines of code.

Relating critically needed auxiliary data to dynamic population registers poses further challenges. Morbidity and cause-of-death data must be entered, linked, and stored. Most DSS projects also maintain socioeconomic data such as on marriage, family relationships, and economic conditions, owing to the strong correlation between health and socioeconomic status. These must be logically consistent with other longitudinal data on the population at risk and relationships among individuals under surveillance. Moreover, projects are often launched to assess the impacts of health technologies, service strategies, or policies, and this necessitates data entry, management, and checking procedures for the internal consistency of service information, as well as procedures to link this information to demographic histories. Variance in exposure to interventions must be monitored at the individual level, in conjunction with precise registration of demographic events and individual risk. Maintaining a detailed record of demographic events, relationships, and exposure to risks or interventions requires complex data-management operations, with a carefully controlled field-operation infrastructure to oversee and support data collection and entry, and a comprehensive computer system for the data-management operation.

Data-management systems required for this operation typically encompass thousands of lines of computer code. A key contribution of the INDEPTH network has been technology-sharing to offset the complexity of developing a data system and creating a reference data model for storage of DSS data. This generic model for data storage facilitates cross-site comparative analyses of the type described in this volume, as it standardizes data rules and concepts across sites. Future work of the network will address the need for generic analytical and data-management software compatible with the reference data model.

This chapter outlines features of this reference data model that pertain to the INDEPTH DSSs. In the not-too-distant past, developing DSS software was difficult, time-consuming, and prone to conceptual and programmatic errors. Software generators and object-oriented tools for software development greatly simply the task of developing a complex system, once common principles of software structure are instantiated in a common applications framework. The mechanisms of INDEPTH have marshalled these software innovations to meet the collective needs of member stations. The reference data model will facilitate exchange of information, swift formulation of site-specific data management software and common software for data analysis, and simplified technical assistance and capacity-building operations.

Background

The work of the INDEPTH Technical Working Group (TWG) has been informed by the achievements, limitations, and future needs of projects in Bangladesh, Burkina Faso, Ghana, Indonesia, Mali, Senegal, South Africa, Tanzania, and Uganda. One of the earlier systems, the Bangladesh DSS in Matlab District, was developed in the 1960s and has since been used for a wide range of studies of demographic dynamics, family planning, epidemiology, health-services research, and other issues (Rahman and D’Souza 1981; D’Souza 1984). Although the Bangladesh DSS has redeveloped its computer operations several times, its field operations have provided a model for a wide range of DSS applications in developing countries. The Bangladesh DSS precisely defined eligibility rules for members of a population under study; this, combined with a data system with rigorous logical-consistency checks, has provided high-quality data for many research papers. A number of software systems have been written, based on experiences with the Bangladesh DSS, including the Sample Registration System (Leon 1986a, b, 1987; Phillips et al. 1988; Mozumdar et al. 1990) and the Indramayu Child Survival Project of the University of Indonesia (Utomo et al. 1990). The DSS in Niakhar, Senegal, most recently described in Garenne (1997), has also influenced the technical design of a number of systems, including those of PRAPASS in Nouna, Burkina Faso (Sauerborn et al. 1996), and Agincourt, South Africa (Tollman et al. 1995). Garenne (1997) described the concept of entry–exit files (similar to the concept of “episodes” described here) as a means of modeling both intervals of residence at a location and intervals of relationships. Garenne also provided useful observations regarding the implementation of field and software systems for longitudinal population studies.

To develop its data model, TWG synthesized the experience of these disparate applications. The model specifies a demographic “core” common to field stations doing longitudinal research on populations (MacLeod et al. 1991; Phillips et al. 1991). Sites have developed software systems to manage this demographic core, maintain a consistent record of significant demographic events in the population of a fixed geographic region, generate registration books that the fieldworkers use, and compute basic demographic rates, such as birth, mortality, and total fertility. These core capabilities establish a computational framework to which projects add their site-specific data and consistency specifications. The concept of a core also entails some generic principles of data collection and management that apply to all INDEPTH sites.

The INDEPTH concept of a data core

All participating sites in INDEPTH collect and maintain a common core of data. Attempts to standardize data processing have led to the concept of a “core system” that provides many of the common software requirements of field research laboratories and can be extended and modified to tailor software to various specifications. This concept is based on the principle that certain characteristics of households, household members, relationships, and demographic events are common to all longitudinal studies of human populations, and software required to collect, enter, and manage data can therefore be generic to a family of applications. TWG has identified these features of a core system common to all DSS operations. In this framework, the core system maintains a consistent record of baseline and longitudinal data on all households, household members, and their relationships in a geographically defined population, including births, deaths, migrations, and marriages. The core system maintains information on events and observation dates to give each entity in the study corresponding “person-day” counts of risk for demographic events. Core computer operations structure data and maintain logical integrity on the following basic elements of a household unit:

Although these are seemingly trivial items, mundane relationships tend to become complex and unwieldy when arrayed as a logical system of longitudinal population data; and portraying even simple relationships requires rigorous standards to avoid error. For example, to be counted as a death in a resident population, a concerned household member must be resident in the study area at the time of death; a live birth to a woman 5 months after she gave birth to another child would be an inconsistent event. A central contribution of TWG has been to clarify such minimal system logic so that the system prevents errors resulting in violation of business rules and rendering data useless.

Most INDEPTH sites have also developed software for reporting outcomes and managing data:

Tailoring the core system

Given the basic core model for data structure, each site has developed site-specific applications using building blocks of the core framework, which allow software developers to construct additional modules for project-specific data. At nine INDEPTH sites, standard tools of database-management packages have been used for an INDEPTH product known as the household-registration system (HRS) for the core specification.¹ Other INDEPTH sites have developed project-specific core capabilities to maintain the logical integrity of birth, death, migration, and marriage data over time and in a format consistent with the reference data model. Each site modifies the core to accommodate new cross-sectional data, special longitudinal modules, or variable classes or labels investigators want to add to field registers, along with logic to maintain the integrity of new variables.

The tools of commercially available database packages greatly facilitate the process of core modification. Standard features of commercially available database systems include those for easily adding data to the core system. For example, the HRS is built from the form menu (data-entry screen) and database builders of the Microsoft FoxPro system. These builders encourage and facilitate an object-oriented software-development approach through easily understandable mouse and menu procedures. To make changes to the core, a programmer locates the database table, menu, or form

¹ The HRS formed the basis for INDEPTH software systems in The Gambia, Ghana (Binka et al. 1995), Indonesia, Mali, Mozambique, Tanzania (three sites), and Uganda. Applications involve a wide range of INDEPTH studies, including family-planning research, malaria interventions, child and maternal health, and correlates of HIV transmission. The current INDEPTH data model improves on the original HRS and other INDEPTH systems by allowing investigators to track nonresident individuals; include more general relationships, rather than just marital relationships; and separate membership in social groups (such as the household or family) from the location.

object to be changed, then works with the small pieces of code, called code snippets, which are “attached” to the object. Some code snippets control the timing of the entry of data for a variable; others enforce rules of consistency. Some INDEPTH sites, such as Hlabisa, are developing similar capabilities, using systems in SQL Server and Access.

The reference data model

As explained in Chapter 3, a DSS tracks the presence of individuals in a defined study area. These individuals can enter and leave the study area in a small set of well-defined ways (for example, entering through birth or in-migration and leaving through death or out-migration). The INDEPTH reference model uses events to record the ways individuals enter (or return to) and leave the study area over time. Thus, events bracket the residency of any individual in the study area. In general, they occur in pairs, with one event (such as presence in the study area) initiating a state and another event (such as migrating out or death) terminating that state. Use of episodes in the reference model makes this pairing of initiating and terminating events explicit. The concept of episodes is diagramed in the centre section of Figure 4.1.

When a DSS tracks episodes, the concept of the “time resolution” of this tracking is very important. Below a certain time threshold, movements into or out of a particular place are not recorded. If a person leaves the physical location in the morning to go to the market and returns in the afternoon, this is not reflected in the DSS. If this period of absence increases beyond a certain threshold (6 weeks, 3 months, or some other period), it turns into an episode to be recorded in the DSS. This threshold varies from project to project, but the project always makes it explicit. The time resolution for “in” episodes should be consistent with the time resolution for “out” episodes, that is, the time before a visit becomes residency or the time after which an absence becomes an out-migration.

DSSs are concerned not only with the physical location or residence of individuals but also with their membership in social groups (such as households) and their relationships with other individuals (such as marital unions or parenthood). Many DSSs also need to reconstruct genealogies and to record isolated events, such as pregnancy outcomes or births and deaths external to the study area.

To support field operations and routine cleaning of data, a DSS must also keep track of where, when, and by whom a particular event was recorded. In this respect, the reference model provides a number of fields to facilitate construction of a good-quality data set. Another challenge for demographic field operations is to correctly identify migrating individuals. To resolve this problem, the reference data model includes fields to designate the place a migrant is moving to or coming from.

The INDEPTH reference model meets these requirements through its use of the following entities and the relationships between them (see Figure 4.1):

Figure 4.1. Reference Demographic Surveillance Data Model.
Note: LMP, last menstrual period. Mandatory fields and entities are in in bold.

In summary, Figure 4.1 illustrates the entities and relationships of the INDEPTH reference data model. Mandatory fields and entities are displayed in bold type, whereas optional fields and entities are displayed in normal (nonbold) type.

The role of the reference data model in maintaining data integrity

As explained in Chapter 3, any DSS must maintain a large volume of data over an extended period. Unless specific measures are taken, the integrity of the data will suffer, along with the accuracy and reliability of the information in the system. INDEPTH has taken steps to foster common standards for data integrity, based on a well-defined relational model. Although not all systems have the same measures to protect data quality, the following have been proposed or used at one or more INDEPTH site:

(The actual values used to indicate the standard values depend on the data type of the field and the natural value range for the data item. Care should be taken to exclude these values from quantitative analysis of the data.)

Extending the core

Although the INDEPTH reference data model covers aspects common to all INDEPTH DSSs, it makes no attempt to specify all site-specific needs. However, it is designed to accommodate new components to meet the needs of a wide spectrum of longitudinal studies, without losing its clear overall structure. Several ways are presented in this section:

Social groups can be related to other social groups, or “first-level” social groups like households can be members of “second-level” social groups like clans or other types of networks. DSSs designed to track the interaction of households might define relationship and membership episodes for social groups, to store information about this topic.

Households are normally associated with only one homestead, even if the members of the household reside in more than one physical location. When social groups are used to record households, this association can be depicted by an episode that records the start and termination of occupation at a physical location. Households also normally have a head of household. This head may change with time, but the household will still retain its identity, and head of household can be recorded either as an updatable attribute (“Current head of household”) or as a member of the social group. If the temporal dimension is important, the extension can be specified as an episode linking the household to an acting head of household.

In summary, the reference data model provides a structure to accommodate great flexibility in the design of longitudinal studies, and for this reason, INDEPTH includes sites engaged in various study designs, with a wide range of data-management needs. Despite this diversity, the model has a core of logic and structure lending integrity to operations and providing a crucial foundation for technical collaboration among sites.

Conclusion

This chapter has described the data model that INDEPTH has developed as the guiding framework for processing data at member sites. It makes attributes common to most health and family-planning studies explicit. As well, it serves as a structural framework for the addition of project-specific data. Much work still needs to be done to develop this model and a common data-processing system for INDEPTH core operations. However, the common framework for data management has already facilitated data sharing within the network, and nearly one-half of all INDEPTH sites use a common software system for core operations. If this use of generic software is more broadly accepted, the INDEPTH data model could serve as the basis for sharing system development, capacity-building, and collaborative research.

Chapter 5
ASSESSING THE QUALITY OF DSS DATA

Introduction

In a DSS, errors occur at all stages of the operation. These may take the form of coverage errors, resulting from omission or repeated counting of persons, or content errors, arising from incorrect reporting of the characteristics of respondents. To establish whether the data are of reasonable quality, INDEPTH sites use a variety of evaluation procedures at the field, data-processing, and analysis stages.

Assessing data quality in the field

It is important to note, at the outset, that however comprehensive the data-checking procedure is, it cannot substitute for careful, methodological, and conscientious interviewing (Shackleton 1998). During training, fieldworkers are made to understand that it is their primary responsibility to ensure accuracy and completeness of data. In addition, field monitoring of data quality is ensured through regular supervisory visits, form checking, and reinterviews.

Supervised visits

The field supervisor’s role is to ensure that each fieldworker conducts interviews of optimum quality. An effective way for a field supervisor to do this is to join up with the fieldworker and observe one or more of the fieldworker’s interviews. The frequency of the supervisory visits varies from site to site and may be daily, weekly, or fortnightly. Such visits are normally unannounced. They are intended to help monitor the performance of the fieldworker from several perspectives. The first is to check whether the fieldworker is actually making the field visits. The supervisor then observes interviews and discusses defects in interviewing techniques. Where necessary, a supervisor makes an effort to help resolve any problems the fieldworker may have. Supervisors pay particular attention to the sequence of the interview process, to prevent omission of questions and make sure fieldworkers follow a logical and systematic format for interviews.

Form checking

First and foremost, fieldworkers are expected to check their own work as part of their daily routine. Ideally, they should do this before leaving the site of an interview so that they can correct errors immediately. Key checks at this stage are to verify the number of event forms, ensure no omission of questions, and provide valid codes for questions. At some sites, each fieldworker gives completed forms to another team member to check before handing them over to the supervisor at the data centre. In addition to observing field interviews during visits, supervisors review samples of completed field questionnaires to identify inconsistencies and assess their completeness. They point out any obvious error for correction. In DSS activities, there is generally a high probability of obtaining missing information on a revisit. Therefore, to maximize the chance of identifying errors before the form leaves the field and minimize the effort required to do the revisit, the team supervisor carefully checks each form again. Here, the checking is more comprehensive and includes validity of dates, consistency of household relationships, and sensibleness of linked fields. Any error detected is returned to the fieldworker for correction, and if needed the fieldworker does a revisit to make corrections.

Duplicate visits

To further check on the reliability of the information, supervisors also carry out random field checks on compounds or households. On these visits, they re-administer portions of the questionnaires. The responses are compared with those obtained by the fieldworker, to provide an idea of the degree of accuracy of the data. At some sites, those responsible for the spot checks also ensure that all neighbouring households are registered. In addition to making random field visits, quality-control supervisors reinterview a 3–15% sample of all compounds or households at the site. They compare the data obtained from the re-enumeration with those of the fieldworker to determine whether the original interview was actually conducted. This also helps to reveal any systematic errors made by the interviewer and provides data for calculating error rates. It must be emphasized, however, that not all errors are completely attributable to the fieldworker, as they may arise if, for example, a different member of the compound or household serves as the respondent. At some DSS sites, efforts to improve on coverage include an independent annual listing of all households, which is then cross-checked against DSS households.

Assessing data quality at the data centre

General procedures

At the data centre, some sites have a second level of supervision: field headquarters’ staff (senior supervisors) thoroughly examine completed questionnaires to identify errors missed by both interviewers and supervisors and ensure that data for individual respondents are consistent. The next stage involves computer editing, using computer programs with built-in checks to assess the validity of responses, either during or following data entry. These built-in consistency checks help to flush out illogical

responses, invalid codes, double entries, and items with missing values. Verification of data is also carried out to detect systematic data-entry errors. This procedure helps to assess the performance of individual data-entry clerks and determine whether the general error rate of data entry is within acceptable limits. At the beginning of each data-collection and data-processing cycle, a verifier repeats the work of a data-entry clerk, until the clerk is qualified in terms of the maximum allowable error rate. Thereafter, only a sample of the work is verified to ensure that the clerk keeps up an acceptable level of accuracy.

Statistical techniques

Matching of records

The statistical procedure to determine the completeness of coverage and reliability of the data is to reinterview and to match individual records case-by-case from two data sources. To evaluate net coverage error, events from the DSS are matched one-on-one with corresponding records from the re-enumeration of 3–15% of the original population. The proportion of records in the re-enumerated sample that were missed in the registration process provides an estimate of the overall coverage error. To assess the accuracy of the data, records from the two data sources are matched, based on a central variable, such as age. By matching individual records from the reinterview with those from the DSS, it is possible to determine the number of individuals omitted from, or erroneously included in, each age group in the DSS. The assumption is that the probability of event omission from the quality-control sample is much lower than, and independent of, the probability of omission from the DSS, although surveys do have correlation biases. Another statistical approach for evaluating coverage and content errors is to compare both absolute and relative numbers from successive periods of the DSS to identify deviations from expected patterns. Occasionally, aggregate figures from the DSS are also compared with those from an independent source to test for consistency.

Population pyramid

The population pyramid is a graphic representation of a population’s age–sex distribution. It is another method to assess the quality of age reporting and is used to give a detailed picture of the age–sex structure of the population. The basic form shows bars corresponding to age groups or single-year age distributions in ascending order, from youngest to oldest. These distributions may be in either absolute numbers or percentages calculated from the grand total for the population. In growing populations, the pyramid is expected to be triangular, with concave sides (that is, it narrows rapidly from the base up). Thus, the shape of the pyramid helps to reveal irregularities, such as age shifting and age heaping, in the age–sex structure of the population.

Alternative techniques

Undercounts and misplacements of events are very often encountered in DSS activities. Other errors resulting in the misclassification of population characteristics also occur. Even with the best quality assurance, it is impossible to overcome all these

errors in the field and at the data centre. Several standard statistical and demographic methods are available to DSS sites for evaluating the accuracy of data.

Age preference

The degree of age preference can be used to test for deficiencies in the DSS data. Although age is the most important variable in demographic analysis, it is typically prone to errors of recall and other types of biases. Age misreporting takes two basic forms: “heaping,” or digit preference, and “shifting.” In less literate populations, the reporting of events, especially births, is usually clustered at certain preferred digits, as a result of ignorance, genuine reporting errors, or deliberate misreporting. Thus, it is common to find concentrations of people at ages with numbers ending in digits 0 and 5 and, to a lesser extent, 4, 6, or 9. Indexes such as Whipple’s index and Myers’ blended index have been developed to statistically assess the extent of age preference, based on the assumption that the population is rectangularly distributed over some age range (Shryock and Siegel 1976). Whereas Whipple’s index is a measure of preference for ages ending in 0 and 5, Myers’ index provides an overall measure of age heaping, as well as an index of preference for other terminal digits.

To measure the extent of heaping on digits 0 and 5, Whipple’s index employs the assumption of rectangularity over a 10-year range and compares the population reporting ages ending in 0 and 5 in the range 23–62 years. The index varies between 100, indicating no preference for digits ending in 0 or 5, and 500, indicating that only digits ending in 0 or 5 were reported. A United Nations-developed scale can be used to evaluate the reliability of any data set based on the estimated Whipple’s index, as follows: <105 = highly accurate; 105–109 = fairly accurate; 110–124 = approximate; 125–174 = rough; 175 = very rough.

The Myers’ blended index involves determining 10 times the proportion of the population reporting in each terminal digit for any 10-year age group. This yields an index of preference for each terminal digit representing the deviation from 10% of the total population reporting the particular digit. The overall index is derived as half the sum of the absolute deviations from 10% and is interpreted as the minimum proportion of individuals for whom an age with an incorrect final digit is reported. The index is 0 when no age heaping occurs and 90 when all age reports have the same terminal digit.

Sex ratios

Another way to appraise the accuracy of data is to examine the general and the age-specific sex composition of the population. The measure usually examined is the sex or masculinity ratio, which is expressed by the following equation:

where P_m and P_f are the number of males and females, respectively. The point of balance for this measure is 100 and is interpreted as the number of males per 100 females. In real life, however, most vital events can be predictably proportioned between males and females. Generally, males outnumber females at birth, but higher rates of male mortality with advancing age offset this pattern. A sex ratio at birth,

therefore, usually ranges between 95 and 102. Thus, failure to observe these typical sex distributions may signify either errors in the data or unusual population characteristics. To obtain a more accurate assessment, researchers normally compare the sex ratio estimated from the data with that obtained in previous years.

Age ratios

Another way to evaluate DSS data is to compare age ratios with expected or standard values. Age ratios are defined here as the ratio of the population in a given age group to one-third the sum of the populations in that age group and in the preceding and following groups, multiplied by 100. The age ratio is expressed for a 5-year age group as follows:

where ₅P_a is the population in the given age group; ₅P_a–₅ is the population in the preceding age group; and ₅P_a+₅ is the population in the following age group. In the absence of extreme fluctuations in the past vital events, the age ratios should be about equal to 100, based on the assumption that coverage errors are about the same for all age groups and that complementary errors in adjacent age groups offset age-reporting errors. The average absolute deviation from 100 of the age ratios, over all ages, gives the age-accuracy index, or overall measure of the accuracy of the age distribution: the lower the age-accuracy index, the more accurate the age data.

Comparison with population models

Yet another way to assess DSS data is to compare the actual percentage distribution of the population by age with an expected age distribution corresponding to a population model, such as that of the “stable population.” With negligible migration and fairly constant fertility and mortality, the age distribution of a population will assume a definite, unchanging form. Thus, the percentage age distribution of a population with a fairly stable structure can be used to evaluate the accuracy of the reported age distributions. For each age group, an index may be calculated by dividing the percentage in the age group in a given country by the corresponding percentage in the stable population. Deviations from 100 signify under- or over-enumeration of the relative age groups. The stable-population model (with zero population growth) and the quasi-stable population model (similar to the stable-population model but with moderately declining mortality) may also be used to assess DSS population age–sex structures.

Conclusion

Right from the start of data collection, the DSS sites use various procedures to ensure sound data, including thorough, manual editing of the questionnaires in the field and at the data centre, partial or complete reinterviewing of a sample of respondents, and computer checks. At the analysis stage, depending on data requirements, specific techniques are applied to assess whether the data conform to an acceptable pattern. It is worth noting here that not all DSS sites have daily work routines. A few sites carry out only annual censuses. However, evaluations of DSS data at many sites suggest that the data are of reasonable quality and that they indicate an improvement over time.

PART II
MORTALITY AT INDEPTH SITES

Chapter 6
COMPARING MORTALITY PATTERNS AT INDEPTH SITES

Abstract

Empirical mortality life tables are chronically lacking for Africa. This chapter presents such tables for 19 INDEPTH sites for the 1995–99 period, with 17 of these in Africa. The data compiled for the calculations represent 4 194 627 person–years of exposure and 56 977 deaths. To compare the overall levels of mortality at the various sites, an INDEPTH population standard was developed and used to standardize observed crude death rates for Africa. Finally, the age- and sex-specific patterns and rates of infant, child, and adult mortality are provided for each DSS site, and mortality clusters are identified.

Introduction

Mortality data from Africa

Accurate data on mortality in Africa are still scarce. Until recently, the main tools for overcoming this shortcoming have been indirect demographic-estimation techniques and model age-specific mortality schedules produced by Brass et al. (1973) (the Brass relational system); Coale and Demeny (1966) (the CD model life-table system); and the United Nations (1982) (the UN model life-table system). The Brass relational system is based on empirical data collected in West Africa during the middle of the 20th century. In contrast, neither the CD nor the UN model life-table system is built using significant amounts of data collected from Africa. Moreover, all three of the systems are based on data that are 30–50 years old. Given the dramatic demographic changes that have affected Africa in the past 20–30 years and the fact that two of the systems are based largely on data collected from other regions and the third is based on data from only one region of Africa, it may be problematic to use them in the current African context. No doubt, the World Fertility Survey (WFS) and the Demographic and Health Surveys (DHS) have remedied in part the above situation by increasing our knowledge of the level trends and differentials in infant and child mortality in the developing world (Cleland and Scott 1987). However, complete mortality life tables

cannot be constructed from WFS and DHS data without relying on indirect methods. Finally, several African countries have since independence undertaken national censuses, but mortality data from these sources are often plagued with underreporting and need to be adjusted using hypotheses that are not always realistic.

Mortality data from INDEPTH sites

Data collected at DSS sites are often dismissed because they are collected from small areas, a fact presumed to make the resulting mortality measures neither accurate nor representative. The modest population size of a DSS site does not really constitute a major flaw, however, as even sites monitoring small populations can produce robust measures of age-specific mortality when data are aggregated over several years. Moreover, data collected over long periods from the same population living in the same area can reveal important age-specific trends in the risk of death. Furthermore, when data from a number of widely dispersed sites are brought together, they provide a measure both geographically and temporally representative of mortality conditions. Currently, only DSS sites provide data of use in depicting the temporal and geographic contours of mortality patterns in Africa.

Each DSS site monitors a well-defined, prospectively linked population over a period of years. The longitudinal nature of the DSS ensures that demographic events (such as births, deaths, and migrations) and person–years of exposure are accurately recorded. Keeping the data-collection rounds short, usually 3–4 months, minimizes the likelihood of “losing” a respondent or failing to observe an event. Consequently, the data presented here are of unusually high quality with respect to coverage, completeness, and accuracy of age.

This chapter presents data for age-specific counts of deaths and person–years of exposure at 19 INDEPTH sites in the period 1995–99. The data are used to construct life tables describing the mortality conditions at each of the sites in this period. The levels of child, adult, and overall mortality are compared across the sites, and standardized CDRs are presented for wider comparison. The next chapter presents a detailed examination of the age patterns of mortality revealed in these data.

Age-specific mortality rates and life tables

Data

The data used in this chapter come from sites for which information on mortality was available for at least a full year during the 1995–99 period (Table 6.1). The overall average length of the observation period for the contributing sites is 3.7 years. In total, the data yield 4 194 627 person–years of exposure, during which 56 977 deaths occurred. An average of 16% of the person–years exposed were lived at ages younger than 5 years old, and an average of 37% of the deaths also occurred between birth and 5 years of age. The CDR for both sexes combined ranges from a low of 7 per 1000 in Agincourt, South Africa, to 39 per 1000 in Bandim, Guinea-Bissau.

Table 6.1. Summary of mortality data from 19 INDEPTH sites, 1995–99.

Note: CDR, crude death rate (actual number of deaths per 1000 population); PY, person–years.
^a Reporting in midyear to midyear annual periods resulted in a 5-year reporting period running from 15 July 1994 to 15 July 1998.
^b Comparison area.
^c Treatment area.

Method of analysis

Although many sites reported data for longer periods, the following analysis is restricted to the 1995–99 period. The aim here is to present the mortality profile of the INDEPTH sites for a recent period for which there was a maximum number of contributing sites.

Life tables were constructed in the standard fashion (Preston et al. 2001). For each site, _nM_x, the age-specific mortality rates for the age group x,x +n were calculated as the ratio of deaths, _nD_x, to person–years exposed, _nPY_x, in the same age group. When calculating _nq_x, the probability of dying in age group x,x+n, one assumes that the average age at death, _na_x, equals half of the age interval, except for ages <5 years. In the age intervals 0–<1 and 1–4 years, the values of _na_x are calculated using the relationships developed by Coale and Demeny, based on the their West model life-table system (Preston et al. 2001). The open age interval encompassing ages ≥85 years is closed in the usual way, by letting _nL₈₅ equal the ratio of l ₈₅ to ∞M₈₅. Standard errors are calculated using formulae developed by Chiang (1984).

Crude death rate

To examine the overall level of mortality reported at each site and to compare those across sites, we calculated the age-standardized crude death rate (ASCDR) and life expectancy at birth. The CDR is the overall death rate obtained by taking the ratio of the total deaths in the population to the total person–years of exposure over a given period. Life expectancy at birth is the number of years a newborn is expected to live if

Table 6.2. Standard age distributions.

^a Standard age distribution proposed by INDEPTH for sub-Saharan Africa.
^b Standard age distribution proposed by Segi (1960).
^c Standard global age distribution proposed by WHO (see Estève et al. 1994).

at each age he or she is subjected to the age-specific mortality rates under consideration. Both measures reflect the total risk of death faced by the population as a whole.

The CDR can also be expressed as the age-weighted average of age-specific mortality rates. As a result, the CDR is a function of both the age structure of the population and its age-specific mortality rates, and variations in either schedule, from one site to another, may yield spurious differences in CDRs. Because diverse populations may have significantly different age distributions, the CDR cannot be directly compared across different populations. To remove the influence of the age structure and make such a comparison possible, it is necessary to substitute a standard age distribution in place of the population’s true age distribution when calculating the CDR. The result is an ASCDR. There are several widely used standard age distributions, including the Segi and WHO standard age distributions (see Segi 1960; Estève et al. 1994). Both of these standards reflect populations with fairly low fertility and mortality. Consequently, they give significant weight to the middle years of life. All of the INDEPTH sites record information from fairly young populations with high fertility and mortality. Under those conditions, the population has proportionally more young people, giving it a “younger” age distribution. When the Segi or WHO standard age distributions are applied to the INDEPTH data, they give too much weight to the high mortality rates prevailing at middle and older ages and too little to mortality at younger ages. Consequently, the absolute level of the ASCDRs produced using those standards significantly overestimates the true level of mortality at the INDEPTH sites.

To address this problem and create ASCDRs that more accurately reflect the true level of mortality at the INDEPTH sites, we calculated the INDEPTH standard age distribution. We constructed an average age distribution for each site over the period 1995–99 by taking the weighted average of the person–years of exposure in

each age group across all of the years for which data had been reported. The weight for each year is the total number of person–years reported for all ages during that year. We calculated the INDEPTH standard age distribution by taking the weighted average of the individual site average age distributions in each age group. In this case, the weights are the total number of person–years in each of the individual site average age distributions. The result is displayed in Table 6.2, along with Segi and WHO standards.

In Figure 6.1, the younger age distribution of the INDEPTH standard, which is typical of developing countries, is contrasted with the much older population structures of the Segi and WHO standards.

Figure 6.1. CDR and life expectancy at birth.
Source: Segi and WHO standards (see Segi 1960; Estève et al. 1994). Note: WHO, World Health Organization.

Table 6.3 displays the CDR for each site and the ASCDRs calculated using both the INDEPTH and Segi standard age distributions along with the values for life expectancy at birth taken from Tables 6A.1–6A.19 (see Annex). Differences in the ASCDRs are the result of differences in the underlying age-specific mortality schedules measured at each site. Because they control for the age distribution of the population, both of the ASCDRs may be directly compared across the sites.

The INDEPTH standardized CDRs range from about 7 to about 33 per 1000 for males and from about 5 to about 27 per 1000 for females, revealing a very wide range of mortality at the INDEPTH sites. The figures for life expectancy at birth vary in a

relationship that is loosely inverse to the values of the CDR (Figure 6.2), and they cover a similarly wide range: from 66 to 39 years for males and from 74 to 40 years for females. The data from Bandim are anomalous and reflect some unresolved questions about the way in which they were collected and reported.

Some geographic clustering occurs. Agincourt, in South Africa, is grouped with the two sites in Bangladesh: the Matlab comparison and treatment areas. Also together at the low end of the spectrum are three rural sites in Tanzania: Hai, Rufiji, and Ifakara; and one site in Senegal: Mlomp. In the middle of the pack are three sites in West Africa: Nouna, Oubritenga, and Farafenni. At the high end is a mixture of sites from West, East, and southern Africa. The absolute level of mortality varies considerably over space, with sites located close to each other having similar levels of mortality, but with a wide range of mortality levels measured in all major regions of Africa.

Table 6.3. Crude death rates and life expectancies at birth for 19 INDEPTH sites, 1995–99.

Note: ASCDR, age-standardized crude death rate; CDR, crude death rate (actual number of deaths per 1000 population); e₀, life expectancy at birth.
^a Standardized with INDEPTH standard age structure.
^b Standardized with Segi standard age structure (see Segi 1960).
^c Treatment area.
^d Comparison area.

For the most part the sex differentials are small, but they generally favour females, as expected. Two of the sites in southern Africa with significant male migration — Agincourt, South Africa, and Manhiça, Mozambique — register substantial sex differentials, standing out in contrast to the rest of the sites. Bandim, in West Africa, also records a very substantial sex differential, but as noted above there may be a methodological explanation for this.

Figure 6.2. ASCDR and life expectancy at birth.
Note: ASCDR, age-standardized crude death rate; comp., comparison area; e₀, life expectancy at birth; treat., treatment area.

Child mortality

The measures of child mortality displayed in Table 6.4 are the life-table probabilities of dying in a specified age group: ₁q₀ for ages 0–<1 year, ₄q₁ for ages 1–4 years, and ₅q₀ for ages 0–<5 years — all taken from the life tables in Tables 6A.1–6A.19 (see Annex). The conventional infant mortality rate is also included. The life-table measures represent the probability that a child who survives to the beginning of the specified age interval will die before reaching the end of that interval. A value of 0.1 for ₁q₀ indicates that 10% of newborns will die before their first birthday, and correspondingly a value of 0.25 for ₄q₁ indicates that 25% of the children reaching their first birthday will die before reaching their fifth birthday. We chose to present these measures because they are intuitive and powerful and represent the fundamental probability of death, rather than a potentially ambiguously defined and difficult-to-interpret rate or ratio to live births, which would be affected by differentials in fertility between sites.

As shown in Figure 6.3, a wide range occurs in the level of child mortality. The probability that a newborn dies before reaching its fifth birthday ranges from 32 to 255 per 1000 for males and from 34 to 217 per 1000 for females. The Agincourt site in South Africa has recorded a comparatively very low level of child mortality. In another cluster, composed of the Matlab sites in Bangladesh, Mlomp in Senegal, and Hai in Tanzania, all have reported low levels of child mortality, but not nearly as low as the level reported from the South Africa site. The next higher cluster is composed of sites

Table 6.4. Infant and child mortality at 19 INDEPTH sites, 1995–99.

Note: IMR, infant mortality rate (number of deaths of infants <1 year old per 1000 live births in a given year); NA, not available; ₁q₀, probability that a newborn will die before reaching its 1st birthday; ₄q₁, probability that a child that has reached its 1st birthday will die before its reaching its 5th birthday; ₅q₀, probability that a newborn will die before reaching its 5th birthday; ₁q₀/₄q₁, ratio of probability of death faced by children before and after their 1st birthday.
^a Treatment area.
^b Comparison area.

from various regions of Africa, including Dar es Salaam, Tanzania; Butajira, Ethiopia; Ifakara, Tanzania; Nouna, Burkina Faso; and Manhiça, Mozambique. Following after, with ₅q₀ very close to 175 per 1000 for males and females, are Farafenni, The Gambia; Rufiji, Tanzania; Navrongo, Ghana; Gwembe, Zambia; Morogoro, Tanzania; and Oubritenga, Burkina Faso. The three remaining sites — Niakhar, Senegal; Bandim, Guinea-Bissau; and Bandafassi, Senegal — all have substantially higher values of ₅q₀, closer to 225 per 1000. A wide range occurs in the level of child mortality, but except at the very lowest and very highest levels, no geographical clustering is apparent. The lowest levels are definitely found in South Africa and Asia, and the highest levels are reported from West Africa.

It is also worth noting the very high levels of ₁q₀ reported from Rufiji,¹ Tanzania, and Bandafassi, Senegal. Both of those values are extraordinarily high and indicate that the conditions for infants in those areas are among the most unfavourable anywhere on the globe. Table 6.4 also displays the ratio of ₁q₀ to ₄q₁, to elucidate the changing risk of death children face before and after their first birthday. This ratio reveals that children in Rufiji who survive to age 1 year face a probability of death improved by nearly a factor of four, whereas children in Bandafassi face a nearly constant probability of dying throughout the first 5 years of life.

Sex differentials in child mortality are fairly small and do not appear to consistently favour one sex over the other. Interestingly, this pattern is broken by four sites: Manhiça, Mozambique; Rufiji, Tanzania; Niakhar, Senegal; and Bandafassi, Senegal. In the last two cases, there is a clear differential favouring females, as there is in Manhiça. In contrast, Rufiji records a substantial differential favouring males.

Figure 6.3. Child mortality.
Note: Cont., control area; ₅q₀, probablity of dying between birth and <5 years of age; treat., treatment area.

Adult mortality

In keeping with the life-table treatment of child mortality, the index chosen for adult mortality, ₃₀q₂₀, is the probability that a person who has survived to age 20 will die before his or her 50th birthday. Values for ₃₀q₂₀ taken from Tables 6A.1–6A.19 (see Annex) are displayed in Table 6.5 along with values of ₅q₀ and the ratio of ₅q₀ to ₃₀q₂₀. The information on child mortality is included to allow the calculation and display of the relationship between child and adult mortality for each site, embodied in the ratio of ₅q₀ to ₃₀q₂₀.

Very substantial ranges occur in the level of adult mortality: 63–501 per 1000 for males and 59–421 per 1000 for females. A value of 500 per 1000 for ₃₀q₂₀ indicates that fully half of the people who survive to age 20 do not live to reach their 50th birthday. Additionally, a number of sites record substantial sex differentials in adult mortality — Mlomp, Senegal; Agincourt, South Africa; Navrongo, Ghana; Hai, Tanzania; and Manhiça, Mozambique, in particular. Also apparent is the opposite differential, in which female rates exceed those of males in two sites: Rufiji, Tanzania, and Dar es Salaam, Tanzania. HIV–AIDS and maternal mortality may play roles. Without more information from the sites, we are unable to explain these differentials.

For the first time, Agincourt, South Africa, does not define the low end of the range. Where adult mortality is concerned, the Matlab sites in Bangladesh clearly stand out, with substantially lower risks of death than anywhere else, and in both these sites a very small sex differential favours females. In both cases, nearly 95% of adults

¹ Rufiji is the newest INDEPTH site and is reporting data for its first year of operation (see Table 7.2). The apparent high risk of death for infants revealed by the data from Rufiji may be in part an artifact, resulting from an age-reporting bias for an infant’s date of birth in first-year DSSs. This is due to the fact that in the first year of any DSS, unlike subsequent years, a large portion of the infants would be born before the DSS started and their birth dates would be subject to maternal recall error. These errors decrease for infants born during the DSS, as such infants become registered soon after birth. This start-up bias would have less of an effect on under-five mortality rates.

Table 6.5. Adult mortality and child–adult mortality ratio at 19 INDEPTH sites, 1995–99.

Note: ₅q₀, probability that a newborn will die before reaching its 5th birthday; ₃₀q₂₀, probability that an adult who has survived to age 20 will die before reaching his or her 50th birthday; ₅q₀/₃₀q₂₀, ratio of the probability that a newborn will die before reaching its 5th birthday to the probability that an adult who has survived to age 20 will die before reaching his or her 50th birthday.
^a Treatment area.
^b Comparison area.

reaching age 20 years survive to their 50th birthday. The next cluster appears at between 150 and 200 per 1000 and includes sites ranging from Mlomp in Senegal to Rufiji in Tanzania (Figure 6.4). In all of these cases, the sex differential is small, except for Agincourt, South Africa, and favours females in all cases except for Rufiji, Tanzania. The last cluster covers a wide range: about 250–475 per 1000. This group includes the remainder of the sites and is marked by the very high risk of adult mortality in Bandim and the substantial sex differentials in Navrongo, Ghana; Hai, Tanzania; and Manhiça, Mozambique.

As was the case with child mortality, the geographic clustering clearly separates the Asian sites from the African sites, but beyond that, there does not appear to be any substantial geographical clustering of similar risk of adult mortality within Africa. The cluster with moderate risk includes sites from all major regions of Africa, as does the high-risk cluster.

The relationship between child and adult mortality reveals three distinct groups: sites in Asia, sites in West Africa, and sites in the rest of Africa. The Asian and some of the West African sites clearly record levels of child mortality that are higher than the corresponding levels of adult mortality. Mortality at all ages is relatively low in Asia, so this finding is primarily the result of exceptionally low adult mortality. In four West African sites — Niakhar and Bandafassi, Senegal; Farafenni, The Gambia; and Oubritenga, Burkina Faso — this is the result of unusually high child mortality, coupled with substantial adult mortality. It is our guess that in these cases malaria is the primary reason why child mortality is so high, but this must be confirmed with more information from those sites.

Figure 6.4. Adult mortality.
Note: Cont., control area; ₃₀q₂₀, probability of dying between ages 20 and 50 years; treat., treatment area.

Discussion

The data presented here are the first large compilation of high-quality data collected over a large area of Africa at intensively operated longitudinal field sites. In light of the general lack of high-quality information describing contemporary mortality in Africa, this is a unique and useful collection of data. The level of mortality varies considerably across the sites that have produced these data, and all but one or two appear to have produced very reasonable age-specific mortality schedules. A great deal of additional analysis will be applied to these data in the near future. The first extension of the basic description of the levels and age patterns of mortality presented here is the identification and thorough examination of the common underlying age patterns of mortality embodied in these data, presented in the following chapter.

ANNEX: LIFE TABLES

Table 6A.1. Life table for the Agincourt DSS site, South Africa, 1995–99.

Note: _nD_x, observed deaths between ages x and x+n; _nd_x, number dying between ages x and x+n; e_x, expectation of life at age x for the life-table population; l_x, number of survivors at age x in the life-table population; _nL_x, person–years lived by the life-table population between ages x and x+n; _nM_x, observed mortality rate for ages x to x+n; NA, not applicable; _nPY_x, observed person–years between ages x and x+n; _nq_x, probability of dying between ages x and x+n; SE_{l_x}, standard error in l_x; SE_{_nM_x}, standard error in _nM_x; SE_{_nq_x}, standard error in _nq_x; SE_{e_x}, standard error in e_x; T_x, person–years lived by the life-table population at ages older than x.

Table 6A.2. Life table for the Bandafassi DSS site, Senegal, 1995–99.

Table 6A.3. Life table for the Bandim DSS site, Guinea-Bissau, 1995–97.

Table 6A.4. Life table for the Butajira DSS site, Ethopia, 1995–96.

Table 6A.5. Life table for the Dar es Salaam DSS site, Tanzania, 1994/95–1998/99.^a

Table 6A.6. Life table for the Farafenni DSS site, The Gambia, 1995–99.

Table 6A.7. Life table for the Gwembe DSS site, Zambia, 1991–95.

Table 6A.8. Life table for the Hai DSS site, Tanzania, 1994/95–1998/99.^a

Table 6A.9. Life table for the Ifakara DSS site, Tanzania, 1997–99.

Table 6A.10. Life table for the Manhiça DSS site, Mozambique, 1998–99.

Table 6A.11. Life table for the comparison area of the Matlab DSS site, Bangladesh, 1998.

Table 6A.12. Life table for the treatment area of the Matlab DSS site, Bangladesh, 1998.

Table 6A.13. Life table for the Mlomp DSS site, Senegal, 1995–99.

Table 6A.14. Life table for the Morogoro DSS site, Tanzania, 1994/95–1998/99.^a

Table 6A.15. Life table for the Navrongo DSS site, Ghana, 1995–99.

Table 6A.16. Life table for the Niakhar DSS site, Senegal, 1995–98.

Table 6A.17. Life table for the Nouna DSS site, Burkina Faso, 1995–98.

Table 6A.18. Life table for the Oubritenga DSS site, Burkina Faso, 1995–98.

Table 6A.19. Life table for the Rufiji DSS site, Tanzania, 1999.

Chapter 7
INDEPTH MORTALITY PATTERNS FOR AFRICA

Abstract

Mortality data from Africa compiled by the INDEPTH Network and including over 6.4 million person–years of exposure are used to identify new mortality patterns. Seven age patterns of mortality emerge from these data, two of which clearly show excess mortality due to HIV–AIDS. The emergent patterns are compared with the existing model mortality patterns produced by Coale and Demeny (CD) and the United Nations (UN) and are demonstrated to be substantially different. The principal-components technique is used to calculate 15 principal components that account for all of the variation in the data. It is demonstrated that the components are sufficiently general to accurately reproduce the existing CD and UN model mortality patterns. The resulting component model of mortality is demonstrated through the construction of a hypothetical set of life tables combining the HIV–AIDS pattern of mortality with an underlying pattern of mortality that is not affected by HIV–AIDS. This general technique yields mortality patterns that might prevail if the population described by the underlying mortality pattern were infected with HIV–AIDS.

Mortality models and Africa

An individual’s probability of dying depends primarily on sex, age, health, genetic endowment, and the environment, all of which determine the risk of falling victim to illness or accident. The primary determinants of mortality interact in complex ways and depend in turn on a large and variable set of complex social determinants. As a result, it has not been possible to formulate a general, theory-driven model of individual risk of death. In lieu of a good general model, two widely used sets of model life tables are the CD model, created by Coale and Demeny (1966), and the later UN model (United Nations 1982). In both cases, a large set of empirical mortality rates are summarized to yield a small number of characteristic age patterns of mortality. Coale and Demeny identified four patterns, which they called North, South, East, and West to reflect the fact that each pattern is representative of the mortality pattern in a

particular region of Europe. For a similar reason, the UN’s patterns also bear regional names: Latin America, Chile, Far East, South Asia, and General. The UN General pattern is, as its name suggests, a general pattern that is not specific to a single location.

Each of the eight existing model mortality patterns (excluding the UN General pattern) results from the characteristic epidemiological profile of the region it represents. For example, the UN South Asia pattern describes an age pattern of mortality with “very high rates under age fifteen and very high rates again at the oldest ages, with correspondingly lower mortality for the prime age groups.” This pattern is ascribed to “high incidences of infectious, parasitic and diarrheal diseases at the youngest ages and high mortality from diarrheal and respiratory diseases at the oldest ages” (United Nations 1982, p. 13).

For large areas of the developing world, accurate information describing the mortality of the population is not available because vital registration systems are incomplete and inaccurate. Where that is true, model mortality patterns are used to substitute for real information. Two important examples are population projections and estimates of child mortality. All population projections must include both existing mortality conditions and educated estimates of future mortality regimes. The Brass estimators of child mortality (United Nations 1983), widely used in areas where accurate data on child mortality are unavailable, rely on estimates of the age pattern of child mortality, and in most cases a model mortality pattern is used for that purpose. Moreover, model mortality patterns are used to evaluate data, to produce smoothed or corrected versions of faulty data, and to extend or fill in the age range of incomplete data. Demographers working in regions where mortality data are inaccurate or incomplete depend heavily on model mortality patterns to allow them to evaluate the data they have and to make reasonable estimates and predictions.

None of the data used to create either of the widely used collections of model mortality patterns came from sub-Saharan Africa. Consequently, it is not evident that the existing model mortality patterns adequately describe the age patterns of mortality in Africa, and it is only because there is nothing else that they are applied to African populations at all. Furthermore, the emergence of the HIV–AIDS pandemic in Africa has radically altered the age pattern of mortality in large areas of the continent. Because the existing model mortality patterns do not contain an AIDS pattern of mortality, they are no longer appropriate under any circumstance where AIDS is a significant cause of death or where AIDS is anticipated as a significant cause of death in the near future. This is an even more serious problem than it might first appear because of the crucial role that model mortality patterns play in routine demographic work in Africa — precisely because of the substantial lack of comprehensive, accurate mortality data.

This chapter presents seven age patterns of mortality derived almost exclusively from data collected in Africa, including two patterns resulting from excess mortality caused by AIDS. A 15-factor model is constructed to summarize the data, and that

model is used to isolate the AIDS-related component of mortality in the AIDS pattern. Last, the AIDS component is superimposed in various amounts on one of the patterns to generate a coarse set of model life tables that illustrates the effects of AIDS mortality.

Mortality data

To allow maximum flexibility in analysis, individual sites provided counts of deaths and person–years observed in standard 0 to 85+ age groups by sex for single years of observation for as many years of observation as possible. The majority of sites were able to provide data in this format, although one or two provided time-aggregated data. Table 7.1 summarizes the data for this work.

The overall aim of this work is to identify age patterns of mortality for Africa and Asia using longitudinal data from INDEPTH field sites. To adequately capture the variation in mortality over time, the data from each site are grouped into 3-year intervals, or as close to 3-year intervals as possible and practical, to yield 70 site–periods. The annual data in each of those periods is aggregated to yield 70 site–period data sets for each sex: 140 site–period data sets in all. Table 7.2 shows the periods chosen for each site.

Table 7.1. Temporal aspects of INDEPTH mortality data.

Note: NA, not applicable.
^a Comparison area.
^b Treatment area.

Table 7.2. Periods chosen for analysis from each site: site–periods.

Note: Numerals label the periods chosen at each site. Observations at the AMMP sites in Tanzania — Dar es Salaam, Hai, and Morogoro — go from midyear to midyear and are reported in midyear to midyear intervals instead of calandar-year intervals. In each of these cases, seven 1-year periods are reported.
^a Comparison area.
^b Treatment area.

Principal-components analysis

Data summary

The goal is to identify a compact representation of the information contained in a large set of observations of similar items. Principal-components analysis transforms the observations to produce an equal number of components. These can reproduce all of the original observations when combined in the appropriate proportions. The components differ from the original observations in that they capture as much variation as possible in as few components as possible. The first component accounts for the maximum variation that one component can account for. After the analyst removes the variation associated with the first component, the second component accounts for as much of the remaining variation as can be accounted for with one component. The analyst continues this process until all the variation in the original data set has been accounted for and the number of components equals the number of original observations. The important consequence is that the majority of the variation in the data set is accounted for in the first few components.

In this way a large set of observations may be summarized using a small number of components. After deciding how much of the original variation must be retained, the analyst may choose to discard the higher order components accounting for the residual variation.

Component model of mortality

The component model of mortality constructed here makes no substantive assumptions regarding the underlying form of the age-specific mortality schedule. The model makes the general assumption that an arbitrary age-specific mortality schedule can be decomposed into a small number of individual components and a negligible residual term. Additionally, it is assumed that a small number of components together form a universal set of age-specific mortality components and that, when combined in the appropriate proportions, they can reproduce any age-specific mortality schedule. For the purposes of this work, these assumptions encompass only the complete set of mortality data examined here; however, it has been demonstrated that the “universal” mortality components generated from the INDEPTH data are capable of reproducing all of the CD and UN model life-table mortality schedules to within a very small tolerance.

Assume there are n separate components of the age-specific mortality schedule and g age groups. Let m represent the g × 1 vector of age-specific logit (_nq_x) values, and let C represent the g × n matrix whose ith column is the g × 1 vector containing the ith component of mortality. Let a be an n × 1 vector of coefficients that determines how much of each component is used to generate the mortality schedule, and let r be a g × 1 vector of residuals, one for each age. Then equation [7.1] is a compact representation of the full-component model of mortality:

where m, C, a, and r are as defined above. Expanding this around the row for the 20–24 age group reveals

where ₅C_ⁱ20 is the value of the ith component for the 20–24 age group; a_i is the value of the coefficient on the ith component; and r₂₀ is the value of the residual for the 20–24 age group. Each of the column vectors contains g elements, one for each age group.

Once the matrix C has been identified through principal-components analysis (described below), the model may be used in many ways. First, it is informative to examine the shape of the components themselves. The primary component (accounting for the bulk of the variation in the data) represents the common underlying shape of the schedule as a function of age. The second and higher order components define age-specific variations on the basic shape. Moreover, it may be possible to associate certain substantive interpretations with the components; for example, one may appear to affect the balance between child and adult mortality, and one may appear to contribute to or remove from a particular age group affected by a specific condition, such as maternal or AIDS-related mortality.

Estimates of the coefficients a that transform the components into a given mortality schedule may be obtained through an ordinary linear least-squares regression of the mortality schedule against the components C. The residual identified in the regression is equivalent to r, and the regression coefficients are the elements of the vector a with the addition of an extra element to store the constant estimated in the regression. Let a′ be the (n + 1) × 1 coefficient vector with the additional element to store the constant generated in the regression model, and let C ′ be the g × (n + 1) matrix of components with one additional column containing all ones to accommodate the constant in a′. The constant is interpreted as a measure of the overall level of the mortality schedule, whereas the coefficients indicate how much of each age pattern (component) is necessary to reproduce the overall age pattern in the original data. Interpreted in this way, the regression controls for level and provides an estimate of how much of each component is contained within the data, or how important each individual age pattern is in generating the age pattern observed in the data. Equation [7.2] represents the regression component model of mortality:

where m, C′, and a′ are defined as above. Expanding this around the row for the 20–24-year age group reveals

where ₅C_ⁱ20 is the value of the ith component for the 20–24 age group; a_i is the value of the coefficient estimated on the ith component; and a_c is the constant term estimated in the regression, taking the same value for all age groups. Each of the column vectors contains g elements, one for each age group.

Ignoring the residual and postmultiplying C′ by a′ (equation [7.2]) yields the original mortality schedule purged of the residual r. Together with C′, the (n + 1) × 1 vector a′ contains all the information needed to reproduce the original mortality schedule to within r. In most cases the number of components (n + 1) necessary to adequately encode the mortality schedule is much less than g, the number of age groups. As a result, a′ is a compact representation of the mortality schedule that encodes the fundamental shape of the schedule without the “noise” associated with the high-order components and the residual term. Additionally, by adjusting the constant term contained in the last element of a′, it is possible to arbitrarily set the level of the mortality schedule without affecting its age pattern.

The individual coefficient vectors associated with each mortality schedule represent the most important dimensions of the mortality schedules and can be compared and grouped without worrying about the high-order noise associated with the individual schedules. Moreover, by comparing only the coefficients corresponding to the components and ignoring the constant, it is possible to compare individual mortality schedules based only on their individual age patterns and not on differences in their level. Correspondingly, by comparing only the constants associated with two mortality schedules, the influence of the age pattern is effectively removed (controlled for), and it is possible to compare the mortality schedules based only on their level.

Principal components of INDEPTH mortality data

For each sex, logit (_nq_x) values are calculated for the standard 0–85 age groups (18 in all)¹ in each of the site–periods according to equations [7.3] and [7.4]. This yields a 70 × 18 data set consisting of one column for each site–period and one row for each age group, with each cell containing a value of logit (_nq_x) corresponding to the specified site–period and age group.

where _nq_x is the life-table probability of death between ages x and x + n for those who survive to age x; _nM_x is the observed mortality rate (the ratio of deaths to person–years lived) for those between ages x and x + n; and _na_x is the average proportion of years between ages x and x + n lived by those who die in that age interval.²

¹ 0, 1–4, 5–9, 10–14, 15–19, 20–24, 25–29, 30–34, 35–39, 40–44, 45–49, 50–54, 55–59, 60–64, 65–69, 70–74, 75–79, 80–84.

² Without substantially more data tabulated by single year of age it is impossible to directly calculate or estimate the values of _na_x . Moreover, except for the youngest ages, the value of _na_x is always near 0.5. At the youngest ages, the values are much closer to 0.25. Additionally, the life table is not highly sensitive to the exact values chosen as long as they are close to 0.25 for ages <5 years and close to 0.5 for ages >5 years. In this work, the value of _na_x used for ages >5 years is 0.5. For ages <5 years, the values for _na_x are for males 0.33 for ages 0–1 years and 0.25 for ages 1–4 years; and for females, 0.35 for ages 0–1 years and 0.25 for ages 1–4 years. These are loosely adapted from the CD West model life-table system (Coale and Demeny 1966).

Table 7.3. First 15 principal components of INDEPTH male mortality.

Note: Units of logit (_nq_x).
^a Percent of variance explained by the component.
^b Cumulative percent of variance explained by the component(s).

The factor³ and score routines provided with the STATA statistical software package release 5.0 (StataCorp 1997)⁴ are used to calculate the principal components of the 70 × 18 covariance matrix⁵ associated with the data set described above. Each age group (row) in the data set is given a weight equal to the total number of person–years of observation for the age group summed across all site–periods. Fifteen of the resulting 70 principal components are retained, and for both males and females those 15 components account for greater than 99.99% of the variation in the data.

Male

The first 15 principal components of INDEPTH male mortality are contained in Table 7.3, and the first 5 components are shown in Figure 7.1. The primary (first) component, PC1, obviously represents the underlying age pattern of mortality, and together PC2–PC4 modify the age pattern in a way that is consistent with mortality caused by AIDS. PC2 in particular has the shape necessary to account for increased mortality between the ages of 20 and 64 years. PC3 and PC4 allow modifications between the ages of 20 and 49 years and during childhood.

Figure 7.1. First five principal components of INDEPTH male mortality. The first five principal components explain 98.94% of total variance.

³ The factor routine is used with the options [pc] to request principal-components analysis; [covariance], to specify that the covariance matrix is analyzed; and [weight], to specify the weighting.

⁴ Mention of a proprietary name does not constitute endorsement of the product and is given only for information.

⁵ The covariance matrix is used so that the observations are not standardized before the calculation. The resulting principal components refer to the unstandardized observations and can be directly recombined to produce age-specific mortality schedules that need no further transformation, except for the inverse logit, to produce values of _nq_x.

Table 7.4. First 15 principal components of INDEPTH female mortality.

Note: Units of logit (_nq_x).
^a Percent of variance explained by the component.
^b Cumulative percent of variance explained by the component(s).

Figure 7.2. First five principal components of INDEPTH female mortality. The first five principal components explain 98.40% of total variance.

The primary component crosses the x-axis between ages 5 and 9 years and again between ages 30 and 34 years, with the result that as the coefficient of the primary component increases, child and adult mortality increases while the mortality of teenagers and young adults decreases. Consequently, the first coefficient determines the ratio of child and adult mortality to teenage and young-adult mortality. This is likely due to the fact that mortality of the very young and elderly is more sensitive to adverse (or advantageous) conditions than the mortality of the generally healthy and robust teenagers and young adults.⁶ Naturally then, this balance accounts for a great deal of the variation in the data and is therefore encoded in the first component. Remember that the overall level of mortality is governed by the value of the constant term in equation [7.2], so the coefficient of the first component is really only responsible for the age balance, not for the absolute level of mortality at any age.

Female

The first 15 principal components of INDEPTH female mortality are contained in Table 7.4, and the first 5 components are shown in Figure 7.2. In broad terms they are very similar to the male components. However, the primary component contains a significant positive bulge between ages 20 and 44 years, which is absent on the male primary component (see Figure 7.3). The most likely explanation for this is that it accounts for the maternal mortality experienced in the female population. Additionally, the second component describes a somewhat narrower, younger pattern of deviation that at its peak is of slightly greater magnitude than that for the males (see Figure 7.4). This likely results from the general fact that the effect of AIDS on female mortality occurs at a younger and more focused age than its effect on male mortality.

⁶ It is also worth noting that the impact of the first component is not constant with age: when the value of the first component is close to zero, the absolute impact is much smaller than when the value of the first component is more distant from zero. An examination of the curve reveals that the absolute effect of the first component increases significantly with age past 39 years.

The third and fourth components are virtually identical for males and females, except at older ages. The data for older ages will not be interpreted, because they are more likely to be inaccurate and the differences are large only for the oldest ages.

Male and female principal components contrasted

Figures 7.3–7.6 plot the first four principal components of INDEPTH mortality for the males and females together, to clearly demonstrate the differences between the male and female components. These differences are discussed briefly above.

To examine the generality of the INDEPTH components of mortality, the existing CD and UN model mortality patterns (at levels corresponding to a life expectancy at birth of 55 years) were regressed against the INDEPTH components of mortality in a simple linear ordinary least-squares regression. The regressions were run against all 15 of the INDEPTH components, the first 10, and finally the first 5. In each case, the fit statistics were examined and the predicted mortality patterns were calculated and visually compared with the fit patterns. Table 7.5 displays the R ² fit statistic for those regressions. Using all 15 components produces near-perfect fits that are able to faithfully reproduce the existing patterns in all respects. Reducing the number of components used has the expected effect of reducing the quality of the overall fit and failing to correctly model the high-frequency variation in the model patterns. Using 10 components still produces a very reasonable fit, and using 5 or 6 components is acceptable in most circumstances; however, with a small number of components, a substantial “smoothing” occurs, as a result of the lack of high-frequency components. This is actually useful if the aim is to capture the fundamental shape of the mortality curve or if the data are “dirty” and the analyst needs the data to fit the basic shape but can ignore the smaller bumps and dips, which may be meaningless.

Figure 7.3. First principal component of INDEPTH male mortality and female mortality contrasted. The first principal component explains 87.12% (male) and 82.49% (female) of the total variance.

Figure 7.4. Second principal component of INDEPTH male mortality and female mortality contrasted. The second principal component explains 8.89% (male) and 11.76% (female) of the total variance.

Figure 7.5. Third principal component of INDEPTH male mortality and female mortality contrasted. The third principal component explains 1.53% (male) and 1.91% (female) of the total variance.

Figure 7.6. Fourth principal component of INDEPTH male mortality and female mortality contrasted. The fourth principal component explains 0.77% (male) and 1.50% (female) of the total variance.

INDEPTH mortality patterns

The overall aim of this work is to identify common age patterns of mortality in the INDEPTH data. The resulting patterns provide a distilled representation of the important mortality conditions experienced by the populations from which the data were collected. Moreover, some understanding of the age patterns of mortality in Africa, based on empirical data from Africa, is invaluable to demographers and planners of all kinds, who must account for present and future mortality in much of their work.

Component-clustering method

The most critical task in identifying the common underlying mortality patterns is to identify clusters of similar patterns — in this case, clusters of site–periods with similar age patterns of mortality. This is a particularly difficult exercise that necessarily involves some subjective input from the analyst.

A given age pattern of mortality can be observed at various levels resulting from the fact that there may be causes of mortality that affect all ages in roughly the same way and consequently do not produce an age pattern. Given that, mortality schedules may cluster along two dimensions: age pattern and level. The age pattern of a mortality schedule contains a lot of information regarding the epidemiological profile of the population and is consequently of primary interest here.

One of the substantial advantages of the component model of mortality is the distilled, parsimonious representation of a mortality pattern that results from regressing it on the components. The vector of regression coefficients contains independent information on the age pattern and level of the mortality schedule. That fact allows the creation of clusters of age patterns without respect to level.

Table 7.5. The R ² values from linear regressions of existing model mortality patterns on the INDEPTH components.

To create the age-pattern clusters, all 70 of the INDEPTH mortality schedules for both males and females are regressed against the appropriate sex-specific components of INDEPTH mortality. The coefficients corresponding to the first 4 principal components are retained, and the other 11 plus the constant are discarded. The first four principal components account for 98.32% of the variation in the male data and 97.66% of the variation in the female data, making them sufficient to capture all but the finest nuances in the age pattern of mortality. These form a collection of 70 4 × 1 coefficient vectors for each sex.

The agglomerative hierarchical clustering algorithm provided with the S-PLUS 2000 Professional statistical software package (release 3) is used to identify clusters of similar coefficient vectors for each sex.⁷ The Ward method used here is described by the provider of the software as follows (MathSoft Inc. 1999, p. 102):

⁷ S-Plus’s “agnes” routine was used with options: metric = euclidean, standardize = true, and linkage type = word.

Table 7.6. INDEPTH mortality age-pattern clusters

Table 7.6. (concluded)

^a Comparison area.
^b Treatment area.

Detailed discussions of clustering techniques and this particular algorithm are found in Kaufman and Rousseeuw (1990), Struyf and Hubert (1997), and MathSoft Inc. (1999).⁸ This routine was applied separately to the male and female data sets, each consisting of four columns (one for each coefficient described above) and 70 rows (one for each site–period).

Clusters

The method described above identified five robust clusters in the male data and seven robust clusters in the female data, presented in Table 7.6. Because females are subject to the additional risk of maternal mortality, their age patterns are always more complex, and so it is not surprising that two more clusters were identified in the female data. Categorizing the male data into the seven female clusters produces seven male clusters that can be directly compared with the female clusters.

In many cases, periods from the same site are grouped in the same cluster, reassuring us that the clustering algorithm is identifying and grouping fundamentally similar mortality schedules. Where periods from the same site are assigned to various clusters, mortality has been changing significantly over time, and the mortality schedules from two periods are substantially different.

Mortality patterns

After the clusters are identified, a characteristic age pattern of mortality is identified for each cluster. In keeping with the use of the component model of mortality, we then calculate, for each of the 15 coefficients derived from the regression of the individual site–period mortality schedules on the 15 components of INDEPTH mortality, the weighted average across the site–periods in each cluster. The weights used are the person–years of observation in each site–period. This yields the average amount of

⁸ A number of clustering techniques were applied to both the raw and the transformed data and to the coefficient vectors, including agglomerative hierarchical clustering, partitioning around K-means, partitioning around K-medoids, fuzzy partitioning, and divisive hierarchical clustering. Three different statistical software packages — STATA (StataCorp 1997), S-PLUS (MathSoft Inc. 1999), and MVSP (Multi-Variate Statistical Package [KCS 1998]) — were used, and in each case all of their clustering routines were tried. All of the methods produced essentially the same clusters but differed in the clarity of their output and in how they managed ambiguous cases. The agglomerative hierarchical algorithm provided with S-PLUS was eventually chosen, based on its clear and robust theoretical underpinnings and the fact that its output is easily understood and interpreted. Moreover, it appeared to provide the most robust clusters and the most efficient means of categorizing ambiguous cases.

each of the 15 components and the constant needed for each of the mortality schedules in a given cluster. When these average values are combined with the components through equation [7.2], the result is the weighted average mortality schedule for each cluster. By varying the constant, the analyst can adjust mortality schedules to an arbitrary level, and for convenience’s sake, the seven INDEPTH mortality patterns presented in Table 7.7 are adjusted to a level that yields a life expectancy at birth of 55 years. Table 7.7 organizes the male and female patterns into the seven female-derived clusters. This arrangement facilitates comparison of the male and female patterns. The five male-derived patterns are retained when the male data are organized into the female-derived patterns; this simply creates two sets of two slightly redundant male patterns. The author verified this by producing the male patterns based on both the male- and female-derived clusters.

Table 7.7. INDEPTH mortality patterns.

Note: Units of logit (_nq_x).

Figures 7.7 and 7.8 plot the seven INDEPTH age patterns of mortality for males and females, respectively. Figures 7.9–7.15 compare each of the seven INDEPTH age patterns of mortality for males and females. The patterns are arbitrarily named 1–7,⁹ and a discussion of the patterns accompanies the plots.

Figure 7.7. Seven INDEPTH age patterns of mortality for males, adjusted to yield a life expectancy at birth of 55 years.

Figure 7.8. Seven INDEPTH age patterns of mortality for females, adjusted to yield a life expectancy at birth of 55 years.

⁹ This is done to avoid the potential stigmatization from use of more descriptive names.

Figure 7.9. INDEPTH mortality pattern 1, adjusted to yield a life expectancy at birth of 55 years.

Pattern 1

The first pattern (Figure 7.9) is similar to the CD North and UN Latin American model life-table age patterns of mortality (see “Comparisons with the Coale and Demeny and United Nations model life tables,” below). There is no indication that HIV–AIDS affects pattern 1, and the male and female age patterns are similar, with the exception of a bulge in the female pattern during the reproductive years, presumably caused by maternal mortality. Pattern 1 is primarily derived from sites in West Africa over the entire period covered by the INDEPTH data set. HIV–AIDS has not yet become as significant a problem in West Africa as it is in Central and southern Africa, so a large impact of AIDS is not expected to be seen in the data from West Africa. It is worth noting that child mortality between the ages of 1 and 9 is significant and substantially elevated above that shown by the most similar existing models, below. This is in keeping with the fact that malaria is a significant cause of death in West Africa, and it has a large impact on those age groups.

Figure 7.10. INDEPTH mortality pattern 2, adjusted to yield a life expectancy at birth of 55 years.

Pattern 2

Pattern 2 (Figure 7.10) is the only pattern to contain significant contributions from Asia, and it is in fact dominated by data from the Matlab project, in Bangladesh. The only other site to contribute data to this pattern is the Mlomp site, in Senegal. Again, the male and female patterns are similar, with the exception of maternal mortality. However, pattern 2 is strikingly different from all of the others in that the mortality of children, teenagers, and young adults is comparatively very low, and correspondingly the mortality of older adults is comparatively high. In keeping with the fact that the data contributing to this pattern come from Bangladesh and Senegal, it is not surprising that there is no evidence at all of an HIV–AIDS impact. Pattern 2 is very similar to the UN South Asia pattern, as it should be, coming largely from South Asia (see below).

Figure 7.11. INDEPTH mortality pattern 3, adjusted to yield a life expectancy at birth of 55 years.

Pattern 3

The sites contributing to pattern 3 (Figure 7.11) are almost exclusively located in southern and East Africa: South Africa and Tanzania in particular. This pattern obviously contains some influence of HIV–AIDS, but not nearly to the degree observed in pattern 5. The South African data come from the Agincourt site, where mortality is extraordinarily low compared with the other INDEPTH sites in Africa and where HIV–AIDS is recognized but not yet impacting the population in the catastrophic way that it is in other parts of southern and East Africa. The remainder of the data come from the Dar es Salaam site, where there appears to be a greater impact of HIV–AIDS. This pattern is most similar to the UN Far East pattern of mortality, corresponding to the fact that infant and child mortality are very low compared with mortality at older ages. A noteworthy feature of this pattern is the fact that infant and child mortality does not appear to be substantially elevated, as might be expected when HIV–AIDS is an important contributor to mortality.

Figure 7.12. INDEPTH mortality pattern 4, adjusted to yield a life expectancy at birth of 55 years.

Pattern 4

Pattern 4 (Figure 7.12) is a variation on pattern 1, with the important difference manifested in the 35-69 years age range. At all other ages, patterns 1 and 4 are negligibly different, except that infant and child mortality in pattern 4 is consistently slightly lower than in pattern 1. But between ages 35 and roughly 69 years, pattern 4 reveals significantly higher mortality than pattern 1. This pattern is most similar to the UN General pattern for females and UN Latin America for males. As was the case with pattern 1, most of the data contributing to pattern 4 come from West Africa.

Figure 7.13. INDEPTH mortality pattern 5, adjusted to yield a life expectancy at birth of 55 years.

Pattern 5

The HIV–AIDS pattern of mortality is most clearly visible in pattern 5 (Figure 7.13). The data contributing to pattern 5 are derived from the three Tanzanian sites run by the Adult Morbidity and Mortality Project in Dar es Salaam, Hai, and Morogoro. A very striking bulge appears in the mortality of males between the ages of 20 and 54 years and for females between the ages of 15 and 49 years. Additionally, the female bulge is significantly narrower and more pronounced, corresponding to the fact that the female population is infected earlier and within a tighter age range. This pattern is not particularly similar to any of the existing model patterns, but it is most closely matched with the UN General (female) and UN Latin American (male) model patterns. Pattern 5 differs from pattern 3 mainly in the shape of the HIV–AIDS impact. The effect is more diffuse with age in pattern 3, meaning that mortality is elevated through a broader age range, the magnitude of the elevation is more consistent, and the differences between the sexes are less apparent. Pattern 3 is derived largely from the Dar es Salaam data, and this may reflect the fact that the epidemic is more mature in Dar es Salaam and has consequently had enough time to infect a wider age range of people of both sexes. As with pattern 3, it is worth noting that infant and child mortality do not appear to be substantially affected in a manner comparable to adult mortality, and this is in contradiction to what is known about HIV prevalence and vertical transmission. Further investigation is needed to determine why this effect is not prominently measured in these data.

Figure 7.14. INDEPTH mortality pattern 6, adjusted to yield a life expectancy at birth of 55 years.

Pattern 6

Pattern 6 (Figure 7.14) is one of the two additional patterns identified in the female data. It is an interesting pattern that reveals very high mortality of children and teenagers, together with comparatively low mortality of infants and adults of all ages. This pattern is exhibited at sites in northeast and West Africa, with most of the data coming from Ethiopia. Without additional information, it is impossible to speculate on what may be producing this unique pattern. The male pattern is most similar to the CD North model pattern, and the female pattern is closest to the CD West model, both of which embody high mortality in the same age ranges. They deviate from those patterns in that infant mortality is substantially lower than would be found in either model pattern, and child and adolescent mortality is significantly higher: this might be called the “Super North” pattern.

Figure 7.15. INDEPTH mortality pattern 7, adjusted to yield a life expectancy at birth of 55 years.

Pattern 7

Pattern 7 (Figure 7.15), the other additional pattern identified in the female data, is also of interest. It is derived from two sites in Central and West Africa. The reason why it was identified in the female data is obvious: a very substantial bulge appears in the female age pattern between ages 25 and 44 years. This most likely results from very serious maternal mortality, the risk of which increases with age. The site in Zambia is a rural site without easy access to modern medical facilities, and this may contribute to an unusual risk of maternal mortality. The corresponding male pattern is similar to pattern 6, and both are similar to the CD North model pattern. The CD North model pattern contains higher child and teenage mortality, coupled with comparatively low mortality at older ages. This is consistent with the fact that malaria is an important contributor to mortality at both sites.

Comparisons with the Coale and Demeny and United Nations model life tables

The INDEPTH mortality patterns are explicitly compared with the existing CD and UN models to ensure that they are indeed new patterns and to demonstrate exactly how they differ from the existing model mortality patterns. The method used is a simple minimum sum of squared differences. Each INDEPTH mortality pattern is compared with all of the existing CD and UN model mortality patterns: CD patterns North, South, East, and West; and UN patterns Latin America, Chile, South Asia, Far East, and General. For each pair of patterns, the difference between the two is calculated

for each age group, and those differences are squared and summed to yield the sum (over all ages) of the squared differences (SSD) between the two patterns. For each INDEPTH pattern, the SDDs derived from the seven comparisons are ranked, and the members of the pair with the smallest SDD are considered to be most similar. All of the mortality patterns used in the comparisons are adjusted to a level corresponding to an life expectancy at birth of 55 years.¹⁰ The SDDs are presented in Table 7.8, where both the minimum and the next greater SDDs for each comparison are identified.

For each INDEPTH pattern, the age-specific deviations from the closest fit existing model pattern are calculated and presented in Figures 7.16 and 7.17. Those figures clearly reveal that all of the INDEPTH patterns are systematically different from the existing model mortality patterns. Both figures reveal clear peaks in the deviations for children (1-14 years) and young to middle-aged adults (25-49 years). Interestingly, infant and child mortality between the ages of 1 and 4 years is generally lower than the corresponding pattern. The peak in the deviations during childhood may be due to malaria and other diseases that have a large impact on children in Africa but not elsewhere in the world, and it is clear that continued investigation is needed to identify all of the factors contributing to childhood deviations. The peak during the adult years is most pronounced for patterns 3 and 5, which are the two patterns affected by HIV–AIDS, and it is reasonable to assume that this peak is primarily due to the impact of HIV–AIDS. It is curious to note that infant and child mortality in patterns 3 and 5 does not appear to be elevated in a manner corresponding to the increase in adult mortality. This suggests that the HIV–AIDS epidemic does not have an enormous impact on infant and child mortality or that all of the data used to generate patterns 3 and 5 are defective with regard to infants and children. It seems unlikely that all the data would be defective, along with being defective to the same degree, and this points to the need for considerable investigation of the impact of HIV–AIDS on infant and child mortality.

Table 7.8. Sum of squared differences comparing INDEPTH and existing mortality patterns.

Source: CD model (North, South, East, West) is from Coale and Demeny (1966); UN model (Latin America, Chile, South Asia, Far East, General) is from United Nations (1982).
Note: CH, Chile; FE, Far East; GL, General; LA, Latin America; SA, South Asia. Bold, minimum; italic, next best.

¹⁰ The level of the INDEPTH patterns is set by adjusting the constant term in the component model of morality, and the CD- and UN-model mortality patterns used in the comparisons are generated by the United Nation’s computer program for the analysis of mortality data, MortPak-Lite (United Nations 1988), at a level corresponding to a life expectancy at birth of 55 years.

Figure 7.16. Age-specific deviations of INDEPTH male mortality patterns from those of best-fit existing models [logit (_nq_x)], adjusted to yield a life expectancy at birth of 55 years.

Figure 7.17. Age-specific deviations of INDEPTH female mortality patterns from those of best-fit existing models [logit (_nq_x)], adjusted to yield a life expectancy at birth of 55 years.

Demonstration of the HIV–AIDS model life-table system

Model life-table construction

The component model of mortality is capable of generating (and fitting) a very wide range of arbitrary mortality patterns. This makes it particularly well-suited for the creation of model life tables. To demonstrate how the component model can be used to create a set of model life tables, we use the INDEPTH mortality components to isolate (in a set of coefficient deviations) the general age pattern of the impact of HIV–AIDS, and then add that impact in increasing quantities to the INDEPTH pattern-1 mortality schedule, thus creating a set of life tables with decreasing life expectancies at birth corresponding to an increasing impact of HIV. The result is a set of life tables with the underlying age pattern defined by INDEPTH pattern 1 but with various levels of HIV–AIDS mortality added to that.

Figures 7.18 and 7.19 display the male and female INDEPTH pattern-5 mortality schedules with and without what is presumed to be the increase in mortality due to HIV–AIDS. Figure 7.20 presents the male INDEPTH pattern-1 mortality with and without an increase in mortality over the infant and childhood ages.¹¹ In each case, the difference between the two curves is fitted against the first 15 components of mortality (for the appropriate sex) to yield the coefficients presented in Table 7.9.

Figure 7.18. INDEPTH male mortality pattern 5, without and with the presumed increase in mortality due to HIV–AIDS (the HIV–AIDS bulge).

¹¹ There is no empirical pattern used to create the increase in infant and child mortality. It is simply created so that it could be included in the model life tables.

Figure 7.19. INDEPTH female mortality pattern 5, without and with the presumed increase in mortality due to HIV–AIDS (the HIV–AIDS bulge).

Figure 7.20. INDEPTH male mortality pattern 1, with and without HIV–AIDS mortality for infants and children.

Table 7.9. Coefficient values estimated in fit of HIV-derived deviations in logit (_nq_x) on the mortality components.

The model life tables are constructed to produce a family of life tables with the underlying mortality of INDEPTH pattern-1 mortality. The HIV–AIDS pattern of mortality is added to each of the members of the family in amounts sufficient to reduce the life expectancy at birth in 5-year increments. Equation [7.5] is a simple extension of the component model of mortality that describes the relationship used to accomplish this. In this case, the (n + 1) × 1 vector d′ of HIV–AIDS coefficient deviations is multiplied by α and added to the (n + 1) × 1 vector of coefficients, a′.¹² The scaling factor α determines how much of the HIV–AIDS pattern to add to the basic pattern of mortality represented by the vector of coefficients, a′. Once that addition has been accomplished, the resulting vector is premultiplied by the matrix of components C′ to yield the logit-transformed probabilities of dying, logit (_nq_x). The relationship governing the HIV-augmented model life table is given by the following equation:

where m, C′, a′, α, and d′ are as defined above. Expanding this around the row for the 20–24 age group reveals

¹² Remember that the prime (′) indicates that the matrices and vectors include the column and row needed to store the constant and its coefficient. Also, n is the number of components used, and g is the number of age groups.

where ₅C_ⁱ20 is the value of the ith component for the 20–24 age group; a_i is the value of the coefficient on the ith component; α is a single scalar applied to the vector of coefficient deviations; d_i is the coefficient deviation for the ith component; a_c is the constant term, which takes the same value for all age groups; and d_c is the deviation for the constant term. Each of the column vectors contains g elements, one for each age group.

Once the logit (_nq_x ) values have been calculated, the inverse logit produces values for _nq_x to be substituted into a life table and used to calculate its other columns, including life expectancy. The model life tables are calculated through an iterative, target-seeking process that varies α until the desired value for the life expectancy is attained (see Figures 7A.1–7A6 and Tables 7A.1–7A.4 in the Annex).

Conclusion

Data describing mortality at 19 sites in Africa and Asia are used to identify seven new age patterns of mortality, six of which originate solely from Africa. A component model of mortality is constructed from the raw data and used to identify clusters of similar age patterns of mortality, and these patterns are compared with the existing CD and UN model life-table age patterns of mortality and demonstrated to be systematically and individually different from the existing models. This finding supports the notion that unique age patterns of mortality occur in Africa and that routine demographic and epidemiological estimations calculated from African data must take this into account. To make these data useful to practicing demographers and epidemiologists, a set of model life tables based on these patterns must be constructed. INDEPTH is pursuing the construction of a set of INDEPTH model life tables for Africa, using the component model of mortality and based on the age patterns of mortality presented here.

ANNEX: AIDS-DECREMENTED MODEL LIFE TABLES

Figure 7A.1. Male life-table probability of dying (_nq_x), decreased by AIDS mortality in 5-year increments (initial life expectancy at birth, 45 years).

Figure 7A.2. Female life-table probability of dying (_nq_x), decreased by AIDS mortality in 5-year increments (initial life expectancy at birth, 45 years).

Figure 7A.3. Male life-table probability of surviving (P_x), decreased by AIDS mortality in 5-year increments (initial life expectancy at birth, 45 years).

Figure 7A.4. Female life-table probability of surviving (P_x), decreased by AIDS mortality in 5-year increments (initial life expectancy at birth, 45 years).

Figure 7A.5. Male life expectancy (e_x, or average remaining lifetime for a person who survives to the beginning of the indicated age interval), decreased by AIDS mortality in 5-year increments (initial life expectancy at birth, 45 years).

Figure 7A.6. Female life expectancy (e_x, or average remaining lifetime for a person who survives to the beginning of the indicated age interval), decreased by AIDS mortality in 5-year increments (initial life expectancy at birth, 45 years).

Table 7A.1. Model life tables for INDEPTH pattern 1: life expectancy of 60.0 years decremented by HIV–AIDS mortality.

Note: e₀, life expectancy at birth (number of years a child is expected to live as calculated at the time of birth); e_x, life expectancy at age x; P_x, probability of surviving at age x; _nq_x, probability of dying between ages x and x+n.

Table 7A.2. Model life tables for INDEPTH pattern 1: life expectancy of 55.0 years decremented by HIV–AIDS mortality.

Table 7A.3. Model life tables for INDEPTH pattern 1: life expectancy of 50.0 years decremented by HIV–AIDS mortality.

Table 7A.4. Model life tables for INDEPTH pattern 1: life expectancy of 45.0 years decremented by HIV–AIDS mortality.

PART III
INDEPTH DSS SITE PROFILES

INTRODUCTION

Part III of this monograph contains detailed profiles of all INDEPTH DSS sites contributing mortality data to the comparative analyses presented in Part II. Part III is intended as a reference for those who wish to know more about any particular site and the context of the population it monitors, the specific methods it uses, and some of its additional demographic outputs.

Information for each site is provided in separate profiles, presented in a standardized format to allow easy comparisons of selected features between sites. A map panel is also provided to help the reader appreciate the exact location of each DSS site.

The first section of each profile provides full details of the physical and human geography of the DSS area. These details include the following:

The first section of each profile also provides population characteristics of the DSS site, including

The second section of each site profile is dedicated to the site itself. This section contains introductory information on the site, including

In the second section of each site profile, you will also find details on the site’s procedures for DSS data collection and processing, including field procedures such as the following:

The third section of each site profile provides the basic outputs of the DSS site, including demographic indicators generated by the site, such as

A graphic of the current population pyramid for the site is also provided, along with a table with the age- and sex-specific all-cause mortality by 5-year age groups. Some sites also provide tables of age-specific fertility and migration rates.

The site profiles are sequenced, first according to geographic area, then by country alphabetically, and finally by site alphabetically. Hence, they appear in the following order:

For quick reference, Table III.1 provides a matrix of key features of all the sites in these profiles.

Table III.1. Key features of INDEPTH DSS sites profiled in Part III of this volume.

Source: INDEPTH data.
Note: Most demographic indicators are from 1995–99, but see Part II, Chapter 6, or Part III, for exact dates for data for each site. Also, demographic indicators may differ here and in Part III, because the latter reports values from the most recent years or uses different denominators. GIS, geographic information system; HRS, household-registration system; mgmt, management; ORP, Operations Research Project; pop., population; SE, socioeconomic; URI, update-round interval; VA, verbal autopsy.
^a P, peri-urban; R, rural; U, urban.
^b Control area.
^c Treatment area.

Note: ASCDR, age-standardized crude death rate (annual number of deaths per 1000 population standardized using the INDEPTH African standard age structure for all African sites and Segi standard age structure[see Segi 1960] for all Asian sites), avg., average; CBR, crude birth rate (annual number of births per 1000 population); CDR, crude death rate (annual number of deaths per 1000 population); CRNI, crude rate of natural increase (crude birth rate minus crude death rate per 100; ignores migrations); Dep. ratio, dependency ratio (ratio of the sum of the populations <15 and >64 years old to the population 15–64 years old 100); e₀, life expectancy at birth; F, female; IMR, infant mortality ratio (number of deaths in infants between birth and 1 year old per 1000 live births); M, male; MMR, maternal mortality ratio (number of pregnancy-related deaths in women 15–49 years old per 100 000 live births); NA, not available; SR, sex ratio; TFR, total fertility rate (average number of children per woman 14–49 years old); U5M, under-five mortality.
^d Number of deaths in children under 5 years old per 1000 under-5 person–years.
^e Number of deaths in children under 5 years old per 1000 live births.
^f Probability of dying before reaching the age of 5 years per 1000 children.

Chapter 8
BUTAJIRA DSS, ETHIOPIA

Site description

Physical geography of the Butajira DSA

The Butajira Rural Health Programme (BRHP), centred on the Butajira DSS, is in Meskan and Mareko District, Gurage Zone, in the Southern Nations, Nationalities and Peoples Regional State in Ethiopia (Figure 8.1). The estimated area of the district is 797 km², of which Butajira covers about 9 km². The area is 130 km south of Addis Ababa and 50 km west of Zway in the Rift Valley, latitude 8.2°N and longitude 38.5°E. Climate varies from arid lowland areas at altitudes of around 1500 m above sea level (asl) (tropical climate) to cool mountainous areas of up to 3500 m asl (temperate climate). The main wet season occurs between June and October, with the remaining months predominantly dry. Daytime temperatures are typically 20–30°C, with nighttime temperatures falling close to freezing at higher altitudes. The lowland areas are drought prone and have been affected during the main droughts in Ethiopia.

Figure 8.1. Location of the Butajira DSS site, Ethiopia (monitored population, 40 000).

Population characteristics of the Butajira DSA

¹ Department of Community Health, Faculty of Medicine, Addis Ababa University, Ethiopia.

²2 Department of Public Health and Clinical Medicine, Umeå University, Sweden.

The district’s population is currently an estimated 260 000, with a density of around 325 people/km². The DSA covers a sample within the district, following 10 communities initially sampled from the entire district, using a probability-proportional-to-size technique. The DSS population is currently an estimated 40 000 (in 2000). Nine of the 10 sites are rural, and 1 is located in Butajira.

The main ethnic group is Gurage, which is further divided into minor ethnic groups or tribes. The Meskan, Mareko, Sodo, Siliti, and Dobi are major tribal groups. Two-thirds of the people follow the Islamic religion, and orthodox Christianity is the second-dominant religion. The major language is Guragigna, with variations between tribal groups. Amharic, the national language, is widely spoken and is also an important written language. Only a minority of individuals, generally in the younger age groups and in the urban area, understand foreign languages such as English. The main occupations are farming in rural areas and small-scale business in town. The district has 30 schools: 1 technical school, 1 high school, and 28 primary schools. About 77% of the population is illiterate. Illiteracy is greater among females and the rural population.

Most houses in rural areas are round, thatched huts, built from timber and mud. In the town, housing is typically dense and crowded, usually with zinc roofs. The main water sources in rural areas are rivers and wells; in town, people have tap water, but it is not piped to every household. Sanitation in general is poor, and only a few houses have latrines. An all-weather gravel road connects Butajira with Addis Ababa, the capital, but other villages in the district are generally connected to the town only via dry-weather roads. Butajira has 24-hour electrical and telephone services, but neither extends to rural areas. The only health centre in the district is in Butajira, and it has so far provided the highest level of health care available, the nearest referral centre being some 100 km away. At present, a new district hospital is being constructed and commissioned in the town. A growing number of private clinics and dispensaries are also available in the district.

Health problems have been associated predominantly with infections and maternity, and difficult access to health care in the rural areas has often exacerbated the problems. Also, a trend has been observed toward a higher incidence of noncommunicable diseases.

The population has suffered from a number of events at the national level, including the long-term Ethiopian civil war (up to the defeat of the Mengistu regime in 1991) and the more recent border conflict with Eritrea. Although the study area had no direct involvement in these conflicts, conscription programs and the diversion of national resources to the military have had detrimental effects nationwide. Similarly, severe and recurring droughts have had considerable effects on rural populations at various times, particularly in lowland areas.

Butajira DSS procedures

Introduction to the Butajira DSS site

The BRHP was initiated in mid-1986 with a complete census of the 10 sampled kebeles (kebele is the smallest administrative unit in Ethiopia) in the District of Meskan and Mareko. Soon after, by January 1987, a DSS with continuous registration of vital events was initiated. The major aims were to develop and evaluate a system for continuous registration of births and deaths to generate valid data on fertility and mortality and provide a study base for essential health research and intervention (Berhane et al. 1999).

The BRHP DSS is primarily a collaborative research project undertaken by the Department of Community Health, Faculty of Medicine, Addis Ababa University, Ethiopia, and the Division of Epidemiology, Department of Public Health and Clinical Medicine, Umea University, Sweden. The collaboration started as a doctoral-study project (Shamebo 1993). Later, it grew into a departmental collaboration and included the development of the study-base infrastructure and involvement of a multidisciplinary group of researchers. The original DSS population in 1987 was around 28 000 and grew over 10 years to about 37 000 active individuals, with more than 60 000 individuals involved at some time during this first 10 years of monitoring.

Studies in Butajira have been conducted in a set of nine randomly selected (probability-proportional-to-size technique) rural kebeles (known as “peasants’ associations”) and one urban kebele (the Urban Dwellers’ Association). Monthly visits to each household have provided the data. The DSS operates as a dynamic open-cohort system. The individual person–years are aggregated to serve as denominators for calculation of various health and demographic indices. So far, three complete censuses of the population (in 1986, 1995, and 1999) have been done. The extent of similarity between the 1994 national census and the DSS database illustrates the quality of the continuous registration system. Currently, the surveillance interval is changing from monthly to quarterly. Custom-made software, based on the dBase system, is used to handle the data.

The BRHP registers births, deaths, marriages, new households, out- and in-migrations, and internal moves (migration within BRHP DSS kebeles). Household and environmental variables were measured during the censuses. The study base is now well established and is being used for other more focused studies on essential health problems of the country, using qualitative, as well as quantitative, research methods. So far, research on childhood respiratory illnesses, other infectious diseases, reproductive health, and mental health has been conducted using the study-base infrastructure.

The intensity and diversity of the research activities have also required a wider participation of multidisciplinary researchers. The participating researchers have backgrounds in obstetrics (Andersson 2000; Berhane 2000), pediatrics (Muhe 1994), epidemiology and biostatistics (Shamebo 1993), sociology, psychiatry (Alem 1997), nursing, and public health. At present, more than 50 field staff are working in the DSS.

This work has contributed to human-resource development and research capacity-building at the Faculty of Medicine, Addis Ababa University. Several training opportunities have been offered at masters and doctoral levels. The doctoral training is conducted in a sandwich model that allows researchers to stay close to their mother institutions and carry on their routine responsibilities. This model of training has also significantly reduced the risk of “brain drain,” which may occur when people are sent

for long-term training abroad. Doctoral training is offered by European universities, mainly by Umeå University, Sweden. The public health master’s program is conducted by the Department of Community Health, Faculty of Medicine, Addis Ababa University.

Butajira DSS data collection and processing

The Butajira area was originally selected for the establishment of the DSS for several reasons. At 130 km from Addis Ababa, it was considered beyond the direct influence of the municipal area but not too far from the university. In the mid-1980s, civil war raged in northern Ethiopia — hence a location to the south was preferred in the interests of long-term continuity. The area also offered a diversity of developmental, geographic, ethnic, and religious parameters within a fairly discrete area. As time passed, the extent of this diversity and its major consequences for many population parameters became increasingly apparent.

Field procedures

INITIAL CENSUS — The initial census of the population in the selected villages was done in 1987 to obtain the baseline population and establish a system of DSS with continuous registration of vital and migratory events at household level. The total population was 28 780. Any adult member of the household >15 years old was eligible to respond in the monthly household interviews. These were carried out by a team of secondary-school graduate enumerators who were based in the kebeles. Each vital event was registered on a separate form at the household level. Basic demographic, social, housing-condition, and health-care-use characteristics were recorded for each household on its entry into the DSS and then during each reenumeration (Berhane et al. 1999).

REGULAR UPDATE ROUNDS — As it happened, the first overall update of the 1987 census was not done until 1995, which was, in retrospect, too long. A further update round was then conducted in 1999.

CONTINUOUS SURVEILLANCE — From the time of the 1987 census until 1999, continuous surveillance was carried out during monthly visits to each household. However, in the light of experience, both here and elsewhere, quarterly household visits were phased in during 1999 and 2000.

FIELD SUPERVISION AND QUALITY CONTROL — Data-quality-assurance mechanisms have been instituted at several points to ensure the integrity of the data. The most critical of these mechanisms is field supervision. Field supervisors daily supervise data collection and check each completed data form. They also make random visits to selected households each month, using a weekly distributed timetable. Research assistants supervise the data flow from the households to the computer system. They also check the data at the field level, for randomly selected households. Researchers work in the field to provide on-site technical assistance and guidance and check data quality. With the advent and easy availability of the global positioning system, mapping exercises at the household level have been carried out more recently.

Data management

Data for the DSS were initially entered as text strings, but the DSS has, since 1994, used software based on the dBase IV platform. As developed for Butajira, this program includes procedures for automatic consistency checking and has more sophisticated facilities for data management and retrieval. The indigenous calendar used in Ethiopia runs behind the international calendar by 2809 days and has 13 months in a year, and this has presented serious obstacles to using proprietary packages for longitudinal data.

Data are currently entered in Butajira, which allows any inconsistent questionnaires to be immediately sent back to the field. This is a significant improvement over earlier practice, which was to centralize data operations in Addis Ababa.

The site manipulates and analyzes data with dBase, Epi-Info, and the Cohort program, developed by Umeå University, which does person–year-based analyses of events in dynamic cohorts. National and international publications and scientific conferences have been the main routes to disseminate this information. Community feedback meetings have been held periodically.

Butajira DSS basic outputs

Demographic indicators

The study population increased from a baseline of 28 616 in 1987 to 37 323 at the beginning of 1997, suggesting a mean yearly growth rate of 2.7%. The explanation for the major difference in population growth between areas has been urbanization, along with a marked excess of births over deaths and net migration into the urban area.

Figure 8.2. Population pyramid for person–years observed at the Butajira DSS site, 1995–99.

The population profile is typical for sub-Saharan Africa: 4.3% of person–years occur in the first year of life, 14.4% in the next 4 years, 29.9% in the 5–14 age group, 48.6% in the 15–64 age group, and only 2.8% in the ≥65 age group (Figure 8.2). The age-dependency ratio is thus 106%. The male–female ratio is 94%.

During 10 years of surveillance, 5143 deaths and 15 667 births were registered in the area, from a total of 336 074 person–years of follow-up. Thus, based on the observed total number of deaths in this study base, the crude mortality rate is 15.3 per 1000 person–years. A total of 71 004 person–years has been observed among women 15–44 years old, representing 2367 reproductive lifetimes and hence an overall fertility of 6.6 births/woman. The maternal mortality ratio has been estimated using several methods and is believed to be around 600 per 100 000 live births (Berhane et al. 2001).

Deaths among children <5 years old represent 48% of all mortality. Half of these deaths occurred during the first year of life, and 53% before 2 months. From the age-specific mortality rates we can estimate the cumulative mortality throughout life. Thus, among live births, an estimated 4.2% die during the first 2 months of life, 8.0% before 1 year, 16.6% before 5 years, 36% before 15 years, and 56% before 65 years. Substantial variations have occurred between areas with regard to under-five mortality, with rates ranging from 80 per 1000 person–years in the urban area to 219 per 1000 person–years in the lowlands. From the age-specific mortality rates, we estimate a current life expectancy at birth of 50.8 years — 49.3 years for males and 52.3 years for females.

Table 8.1 shows the age- and sex-specific all-cause mortality at the Butajira DSS site.

Table 8.1. Age- and sex-specific mortality at the Butajira DSS site, Ethiopia, 1995–99.

Note: CBR, crude birth rate (actual number of births per 1000 population); CDR, crude death rate (actual number of deaths per 1000 population); CRNI, crude rate of natural increase (CBR minus CDR per 100; does not take into account migration); _nD_x, observed deaths between ages x and x +n; _nPY_x, observed person–years between ages x and x +n.

Acknowledgments

Our heart-felt gratitude goes, first of all, to the people of the Butajira site, who generously shared their personal information and experiences, and to the entire field staff, who diligently collected the data. We thank the health and administrative authorities in Butajira, Gurage Zone, and the Regional Health Bureau in Awassa for their facilitation of the fieldwork. We are very grateful to the Ethiopian Science and Technology Commission and the Swedish Agency for Research Cooperation with Developing Countries for generously funding the program since its establishment. Research, technical, and administrative staff at the Faculty of Medicine, Addis Ababa University, Ethiopia, and the Division of Epidemiology, Department of Public Health and Clinical Medicine, Umeå University, Sweden, are acknowledged for their all-round facilitation.

Chapter 9
DAR ES SALAAM DSS, TANZANIA

Robert Mswia, David Whiting, Gregory Kabadi, Honorati Masanja, and Philip Setel¹

Site description

Physical geography of the Dar es Salaam DSA

The Dar es Salaam region, on the east coast of Tanzania, includes the municipalities of Ilala, Temeke, and Kinondoni (which constitute the city of Dar es Salaam) and a few outlying areas (Figure 9.1). It borders on the Indian Ocean to the east and, on all other sides, the coast region. In 1988, the estimated population of Dar es Salaam, according to a national census, was 1 360 865. But the city grew rapidly during the 1990s, and the current population of Dar es Salaam is estimated at 3 million. The area participating in surveillance covered eight “branches” in two municipal areas of Dar es Salaam: Temeke and Ilala.

Dar es Salaam is at sea level, and the DSS site lies between latitudes 6.82° and 6.89°S and longitudes 39.24° and 39.30°E. The climate is typically tropical, with hot weather throughout the year (range, about 26°–35°C) and two rainy seasons: short rains in November–December and long rains in March–May.

Figure 9.1. Location of the Dar es Salaam DSS site, Tanzania (monitored population, 70 000).

Population characteristics of the Dar es Salaam DSA

Because socioeconomic status is important in the study of mortality differentials, three areas of the city — Ilala, Mtoni, and Keko — were chosen. These areas were thought best for the following reasons:

These three areas contain eight branches, with a total population of 69 304 (as of June 1999).

Originally, the Zaramo ethnic group inhabited the area that is now Dar es Salaam. During the 20th century, however, the population became a mix of many of the country’s ethnic groups. Thus, the population in the DSA is a mixture of people from all parts of Tanzania. The majority of the population in the project area are Muslims (70%), and the remainder are Christians (30%).

The major language of people participating in the DSS is Kiswahili. A large portion of people in the DSA engage in small business or manual labour (both skilled and unskilled); and a few have office jobs.

Information on the highest level of education attained by individuals has been collected in both the census and the mortality-data survey, and the proportion of girls attending school is slightly higher than that of boys for all ages up to 14 years. Thereafter, this proportion drops significantly. From age 25, women with no formal education constitute a proportion two to three times greater than that of men. From age 30, significantly more men than women have completed primary and postprimary education.

The majority of people in Dar es Salaam live in low-cost rental housing, and the mean household size is 3.8. More than 70% of households occupy only one or two rooms. About 80% of households have tap water. Use of pit latrines is extensive (90%) in the Dar es Salaam DSA. It has both paved and unpaved roads, and all areas are well served with public transportation. The area has both public and private telecommunications. Electricity is available in these areas, mostly for domestic uses, such as lighting and cooking.

The Dar es Salaam region has one national and three municipal-government hospitals. People in the surveillance area make good use of all these facilities, although none is within the DSA itself. Two of the municipal hospitals, however, are within easy access to the study community, and the DSS population also has access to government health centres and dispensaries and to a number of private hospitals and dispensaries. The private hospitals are outside of the DSA, but some private dispensaries operate within the area.

Dar es Salaam DSS procedures

Introduction to the Dar es Salaam DSS site

DSS work is carried out in Dar es Salaam primarily to provide reliable population denominators for continuous cause-specific mortality monitoring. The demographic and mortality monitoring together provide municipal authorities with information on the burden of disease, health-facility use in the period before death, and population conditions. These data are used for evidence-based planning and evaluation of health services. The monitoring is an activity of the Tanzanian Ministry of Health and the municipal health-management team, as part of the Adult Morbidity and Mortality Project, phase 2 (AMMP-2). The goal of AMMP-2 is to decrease the morbidity and mortality stemming from conditions particularly likely to cause suffering and disadvantage among Tanzania’s poor people, where these conditions are amenable to health-service interventions. To contribute to this goal, the project has aimed to strengthen evidence-based planning and development of cost-effective health services within the context of health-sector reform in project districts and in the Ministry of Health of Tanzania.

Demographic and mortality monitoring in AMMP-1 and -2 is carried out in Hai District (Kilimanjaro region) and Morogoro District (Morogoro region), as well as in Dar es Salaam (see Chapters 10 and 12). In 1992, when DSS work began, the Dar es Salaam DSA comprised seven urban branches, with a total population of 67 000. At the end of 1993, one more branch, with a population of 4500, was included in the monitoring to make a total population of 71 500. The population in the Dar es Salaam DSA has remained remarkably constant, despite considerable in- and out-migration each year. Although the initial focus was on adults, the system has been collecting data on people of all ages.

The DSS is incorporated into both national and district structures. In the Ministry of Health, the National Sentinel System assumes overall responsibility for using DSS to gather demographic and mortality data. This system also operates in the Hai, Morogoro, and Rufiji DSS sites. At the district level, surveillance work will become part of the routine systems of the district. Mortality monitoring will continue indefinitely, and DSS will continue as long as the district has no cost-effective alternative way of generating reliable population denominators.

The Dar es Salaam Public Health Service Delivery System is the primary local user of the data, and the Ministry of Health is the primary national user. Additional users of the data include

Dar es Salaam DSS data collection and processing

The initial population in the DSS approached the level that Hayes et al. (1989) suggested is best for the ascertainment of cause-specific mortality. As stated above, the Dar es Salaam areas were chosen to represent a range of urban living conditions, including variations in socioeconomic status and population density.

Field procedures

INITIAL CENSUS — An initial census was carried out in 1992, because neither vital registration nor the 1988 National Census provided an accurate basis for estimating population denominators. At first, a baseline census was taken to determine who was resident in each household under surveillance. A single form was used for each household.

REGULAR UPDATE ROUNDS — Subsequently, the population has been enumerated twice a year (May–June and October–November). In each update round, the information from the previous round is printed on new forms for each household. Each household is visited, and an adult member is interviewed. The enumerators verify and, where necessary, update existing data. When new households appear as a result of either migration into the area or splitting of existing households, they are registered on new-household forms. Key informants, such as local leaders, identify these households. Vital events (births and deaths) and migrations are recorded for each household. The following information is recorded for each individual: name, age, sex, relationship to head of household, main occupation, marital status, alcohol consumption and smoking habits, date of entry into the household, mode of entry, date of exit, mode of exit, and whether the individual’s parents are alive. Recently, questions on religion have been added. Migration tracking is limited to recording the dates of entry into and exit from the area and the district of origin or destination; successive migrations of individuals into and out of the area are not linked. Thus, although it is possible to determine who is resident at any point in time (and therefore to calculate denominators), it is impossible to calculate the total time particular individuals spend in the DSA.

The DSS employs eight community-development workers as enumerators for the census-update rounds, and three clinical officers act as verbal-autopsy (VA) supervisors. The system also has community-based key informants, who report deaths to the VA supervisor on a regular basis. Whereas the census-update rounds take place twice annually, mortality monitoring, which provides information on probable causes of death, is continuous. Probable causes of death are determined using VA techniques.

CONTINUOUS MORTALITY SURVEILLANCE — The primary objective of the AMMP approach to DSS is to provide sentinel data on the burden of disease to inform health planning and priority-setting, and thus efforts are made to determine the cause of death for each person who dies in the area under surveillance. This is achieved by interviewing the relatives and caretakers of the deceased, using a short, standard interview schedule. Different forms are used for deaths among infants <31 days old, children between 31 days and <5 years old, and all persons ≥5 years old. The forms contain a section to identify the respondent, one to identify the deceased, an open-ended history section, a checklist of previously diagnosed conditions, a checklist of symptoms and their duration, a list of health services sought in the period leading up to the death, a residential history, and a summary of any confirmatory evidence, such as medical records or a death certificate. Trained health personnel complete the form after interviewing one or more of the deceased’s relatives or caretakers. Wherever possible, the interview takes place within 6 weeks of the death.

Deaths are usually reported by community-based key informants, and in Dar es Salaam various individuals are used for this purpose. Key informants are chosen because of their awareness of events, such as deaths, in their communities. In addition, communities receive feedback in a newsletter; consequently, they perceive a benefit in participating in the surveillance system and actively report deaths to the key informants, thus making this a form of vital registration. Recently, each key informant from a village or area has been given a turubai (canvas tarpaulin) so that the bereaved families from the community can borrow it for funeral gatherings during the mourning period. This has enabled key informants to get information on a death that has occurred in his or her area and thus report it to the supervisor. The VA personnel meet the key informants on a regular basis to find out about new deaths that have occurred. They then meet with the relatives or caretaker of the deceased to verify that the death has occurred, then perform the VA.

Two physicians independently assign a cause of death. Until 1999, a modified version of ICD-10 was used. From 2000, a shorter, broader list of codes, developed by AMMP and the Ministry of Health, has been used. The diagnoses given by the two coders are compared, and discrepancies are given to a third coder. If all three coders disagree, the form is coded as “uncertain/unknown.” Wherever possible, confirmatory evidence of the cause of death is obtained. This includes in- and out-patient records, death certificates, and burial permits.

Data management

During the census, a field supervisor reviews all completed forms and returns those with errors and inconsistencies to the enumerators for correction. Those passing inspection are sent to the data centre in Dar es Salaam and entered into a computer. All census forms with errors detected during data entry are logged and returned to the field for correction. Once the corrected forms are returned to the office, they are logged back in, and the problems are resolved.

Staff are trained to enter the data into microcomputers using a data-entry system designed specifically for the project in Microsoft FoxPro. They are instructed on how the census forms should be completed so that, in addition to the computer validation programs, they, too, can detect errors and inconsistencies. The validation programs range from simple range checks to checks for inconsistencies across household

members, such as an individual identified as a “spouse” but with marital status recorded as “never married.”

Several methods are employed to ensure data quality, including checks in the field and in the data-entry process. Supervisors visit a random sample of households to verify entries on the census forms and check that the census includes all the households visited and that no nonexistent households have been included. Following each census, reinterviews are also conducted of a sample of households for each enumerator. Because of the large amount of data collected in a single census, it is impossible to double enter all data for verification; instead, a 5–10% random sample is taken, and the forms are checked against the entered data.

At the end of each interview, the interviewers give each household a newsletter designed by the municipal health-management team and produced and distributed by the project for US $0.11 per household. It contains health-education messages and simplified presentations of results from the previous round. It shows that the DSS is part of the functioning of the district health system. The newsletter is designed to help the communities and their leaders better understand the areas where they live. In 1999, 94% of households reported receiving the newsletter, and 89% of households reported reading it.

Dar es Salaam DSS basic outputs

Demographic indicators

The primary outputs of the system are estimates of cause-specific mortality for all ages. As stated, the resident population of the DSS site is about 70 000. Average household size is 3.8. Male–female ratio is 100 : 102, with an age-dependency ratio of 59%. The main age group structures of the current population are as follows: those <1 year old account for 3.1%; 1–4 years old, 10.4%; 5–14 years old, 21.9%; 15–64 years old, 63.0%; and ≥65 years old, 1.6%. Between July 1992 and June 1999, the maternal mortality ratio was 669 per 100 000 live births.

The following migration figures reflect changes of residence of people on an annual basis and do not capture short-term movements between the enumeration rounds. In 1998–99, the surveillance area had an out-migration of 17 796 people. The region of destination was obtained for 15 124 of these: most (75%) migrated to another part of Dar es Salaam; the rest, to various parts of the country, except for 172 who moved to other countries outside Tanzania. During the same years, 16 581 people migrated to households within the surveillance area. The place of origin for 13 087 (79%) was determined: 68% migrated from areas within Dar es Salaam; the remainder came from other parts of Tanzania, except for 67 who came from other countries. As can be seen from the figures above, the population in the Dar es Salaam surveillance area is very dynamic. Dar es Salaam attracts young adults, and this can be seen in the shape of the population pyramid (Figure 9.2). The excess of females becomes obvious in the 15–19-year age group, whereas for males this occurs 5 years later.

Table 9.1 shows the age- and sex-specific all-cause mortality at the Dar es Salaam DSS site.

Figure 9.2. Population pyramid for person–years observed at the Dar es Salaam DSS site, Tanzania, 1995–99.

Table 9.1. Age- and sex-specific mortality at the Dar es Salaam DSS site, Tanzania, 1995–99.

In 2000, AMMP added questions to the census round to determine more detailed fertility and migration characteristics and their effects on the population structure. Preliminary analyses of the data indicated that a considerable amount of short-term migration occurred between census rounds. In addition, the age-specific fertility rates of those who migrated to Dar es Salaam were about half those of older residents. The in-migration of young adults, with lower levels of fertility, contributed to the “bulge” in the population pyramid in the 15–49 age group. These are preliminary data, and further analyses of these data are planned.

Acknowledgments

AMMP is a project of the Tanzania Ministry of Health, funded by the Department for International Development (DFID), United Kingdom. The project is implemented in partnership with the University of Newcastle upon Tyne, United Kingdom.

This chapter is, in part, an output of a project that DFID has funded for the benefit of Tanzania and other developing countries, and the views expressed are not necessarily those of DFID.

The AMMP team includes K.G.M.M. Alberti, Richard Amaro, Yusuf Hemed, Berlina Job, Gregory Kabadi, Judith Kahama, Joel Kalula, Ayoub Kibao, John Kissima, Henry Kitange, Regina Kutaga, Mary Lewanga, Frederic Macha, Haroun Machibya, Honorati Masanja, Louisa Masayanyika, Mkamba Mashombo, Godwill Massawe, Gabriel Masuki, Ali Mhina, Veronica Mkusa, Ades Moshy, Hamisi Mponezya, Robert Mswia, Deo Mtasiwa, Ferdinand Mugusi, Samuel Ngatunga, Mkay Nguluma, Peter Nkulila, Seif Rashid, J.J. Rubona, Asha Sankole, Daudi Simba, Philip Setel, Nigel Unwin, and David Whiting.

The AMMP team would like to acknowledge the municipal health-management team from Temeke and Ilala for their continued support and collaboration. We are also grateful for the contributions and efforts of AMMP support staff: Mariana Lugemwa, Dorothy Lyimo, Rukia Mwamtemi, Getrude Peter, Charles William, Mustapha Kahise, and Juma Mfinanga. Finally, we would like to express our sincere thanks to all those who live in the project area for their patience and cooperation.

Chapter 10
HAI DSS, TANZANIA

Robert Mswia, David Whiting, Gregory Kabadi, Honorati Masanja, and Philip Setel¹

Site description

Physical geography of the Hai DSA

Hai District is on the slopes of Mount Kilimanjaro, in Kilimanjaro region, in northeast Tanzania (Figure 10.1). The district has an area of 13 000 km², spanning three ecological zones. Its lowland zone lies between 750 and 1000 m above sea level (asl), with scanty rainfall (about 325 mm a year), warm to hot temperatures, and sparse population density (about 70 people per km²). The midland zone lies between 1000 and 1600 m asl and has higher rainfall (about 1560 mm a year), moderate temperatures, and higher population density (about 150–160 per km²). The highland zone is above 1600 m asl and has heavy rainfall, cool temperatures, and mountain forests and grasslands. People do not live in this zone, but it constitutes the largest water reservoir (from rainfall and glacial runoff) and forest reserve in Kilimanjaro. Multiple springs and rivers flow from this zone and water both the midland and the lowland zones

Figure 10.1 Location of the Hai DSS site, Tanzania (monitored population, 154 000).

¹ Adult Morbidity and Mortality Project, Tanzania.

before emptying into the Pangani River basin. The district has 4 administrative divisions, 11 wards, and 61 villages. The Hai DSS site lies between latitudes 3.13° and 3.46°S and longitudes 37.11° and 37.36°E, and it covers three of the four divisions of the district.

Population characteristics of the Hai DSA

As of 1999, the Hai DSS site has had a population of about 154 000, with an average annual rate of growth of 1.7% from 1992 through 1999. It has 36 000 households (rounded to nearest thousand), and the average household size is 4.6. A household is defined as “people eating from the same cooking pot.” Many villages in the DSA are rural, and some are peri-urban. The main ethnic groups are Chagga and the Maasai. Major religious groups in the area are Christian (79%) and Muslim (20%). Indigenous languages are commonly spoken in the villages, although the national language, Kiswahili, is widely understood and spoken.

Agriculture, livestock keeping, dairy farming, commercial mining, and cottage industries are the main economic activities. At present, the district has 139 primary schools, with a total student population of 40 000. It has 13 secondary schools (both public and private), with 5000 pupils, and 5 postprimary technical schools, with 720 pupils. About 95% of school-age children attend school, and about 96% of the population is literate.

The district has 2 hospitals, 2 health centres, 39 dispensaries, and 61 village health posts. These provide curative, preventive, and health-promotion services. About 85% of children <5 years old are vaccinated against five major communicable diseases. Water is plentiful in the highland and midland zones, but it is often polluted with microbes, toxic minerals, and agricultural chemicals.

Wood is the main source of fuel. Because the population is growing rapidly, demand for fuelwood is rising sharply, and people are now encroaching on the highland forest reserve in Kilimanjaro National Park. Thirty-six of 61 villages have electricity, but the use of this source of energy is limited because of the cost. A few households use biogas (gas processed from waste products, such as cow dung and agricultural wastes).

Hai has a transportation infrastructure comprising 710 km of road and an international airport (Kilimanjaro International Airport). Most roads are unpaved, and they are often impassable for vehicles during the rainy season. Community-based data show that the main causes of death in Hai are HIV–AIDS, cancer, perinatal causes, acute febrile illness (including malaria), pneumonia, diarrheal diseases, injuries (both intentional and unintentional), malnutrition, and maternal causes.

Hai DSS procedures

Introduction to the Hai DSS site

DSS work is carried out in Hai to provide reliable population denominators for continuous monitoring of cause-specific mortality. Such demographic and mortality monitoring activities have together provided district authorities with information on the burden of disease, health-facility use in the period before death, and population conditions. These data are used for evidence-based planning and evaluation of health services. The monitoring is an activity of the Tanzanian Ministry of Health and the district health-management team, as part of the Adult Morbidity and Mortality Project, phase 2 (AMMP-2). The goal of AMMP-2 is to decrease the morbidity and mortality stemming from conditions particularly likely to cause suffering and disadvantage among Tanzania’s poor people, where these conditions are amenable to health-service interventions. To contribute to this goal, the project has aimed to strengthen evidence-based planning and development of cost-effective health services within the context of health-sector reform in project districts and in the Ministry of Health in Tanzania.

AMMP was established in 1992 to provide information for the Tanzanian Ministry of Health, regarding the policy implications of adult morbidity and mortality in the country. Demographic and mortality monitoring in both phases of AMMP has been carried out in Temeke and Ilala districts (Dar es Salaam) and Morogoro rural district (Morogoro region), as well as in Hai District (see Chapters 9, 12). In 1992, the Hai project area had 51 rural villages, with a total population of 142 000. The population has grown to a current total of 154 000. Although the initial focus was on adults, the system now collects data on people of all ages.

As stated, the demographic and mortality monitoring began in 1992. Since then there has been one enumeration round each year. The DSS is now being incorporated into both national and district structures. In the Ministry of Health, the National Sentinel System has overall responsibility for using DSS to collect demographic and mortality data. Dar es Salaam and Morogoro rural sites are also becoming part of this sentinel system, and the Rufiji DSS site contributes data on cause-specific mortality to it. Monitoring work is expected to become part of the routine systems of the district. Mortality monitoring will continue indefinitely, and DSS will continue as long as the district has no cost-effective alternative for generating reliable population denominators.

Hai DSS data collection and processing

The size of the population in the DSS (when originally established) was intended to approximate the level that Hayes et al. (1989) suggested is best for ascertaining cause-specific mortality. Within the original set of three AMMP-supported areas (Hai, Dar es Salaam, Morogoro), Hai, a fairly affluent rural area, was chosen to represent a range of rural living conditions, including variations in socioeconomic status and population density.

The Hai District health-management team and the district council (through the office of the district executive director and the district social-services committee) are primary local users of the data. The Ministry of Health is the primary national user. Additional users of the data include

Field procedures

INITIAL CENSUS AND REGULAR UPDATE ROUNDS — The initial census was carried out in 1992, starting with a baseline census to determine who was present in each household in the DSA. A single form was used for each household. Subsequently, the population has been enumerated once a year for an 8-week period beginning each July. Although the census-update rounds take place annually, mortality monitoring, which provides information on probable causes of death, is continuous throughout the year. Probable cause of death is determined using the verbal-autopsy (VA) technique. Fifty-six village members — mostly rural medical assistants, nurses, village health workers, and retired health personnel — are given a small amount of money to act as enumerators for the census-update rounds. They are also key informants who report deaths to the VA supervisory team on a regular basis. Five clinical-health officers from the district constitute the VA supervisory team. In each census round the information from the previous round is printed on new forms for each household. Each household is visited, and an adult member of the household is interviewed. The enumerators verify and, where necessary, update existing data. When new households appear as a result of either migration into the area or splitting of existing households (for example, through marriage), they are registered on new-household forms. The enumerators, with the help of 10 cell leaders (local branch leaders, who are in charge of 10 households and are supposed to know members of the branch), identify these households. Vital events (births and deaths) and migrations are recorded for each household. The following items of data are recorded for each individual during a household visit: name, age, sex, relationship to head of household, main occupation, marital status, drinking and smoking habits, date of entry into the household, mode of entry, date of exit, mode of exit, and whether the individual’s parents are alive. Recently, questions on religion and residency were added. Migration tracking is limited to recording the date of entry into and exit from the area and the district of origin or destination;

successive migrations of individuals into and out of the area are not linked. It is possible therefore to determine who is resident at any point in time (and therefore to calculate denominators), but it is impossible to calculate the total time each individual has spent in the surveillance area.

CONTINUOUS MORTALITY SURVEILLANCE — The primary objective of the AMMP DSS is to provide sentinel data on the burden of disease to inform health planning and priority-setting. Therefore, an effort is made to determine the cause of death for each person who dies in the area. This is achieved by interviewing the relatives and caretakers of the deceased, using a short, standard interview schedule. Different forms are used for deaths of infants <31 days old, of children between 31 days and <5 years old, and of all persons ≥5 years old. The forms contain a section to identify the respondent, one to identify the deceased, an open-ended history section, a checklist of previously diagnosed conditions, a checklist of symptoms and their duration, a list of health services sought in the period leading up to the death, a residential history, and a summary of any confirmatory evidence, such as medical records or a death certificate. Trained health personnel complete the form after interviewing one or more of the deceased’s relatives or caretakers. Wherever possible, the interview takes place within 6 weeks of the death. Deaths are usually reported by the community-based key informants mentioned above. These individuals have been chosen because of their awareness of events in their communities and the likelihood that they will know of any deaths that occur. In addition, communities receive feedback in a newsletter; consequently, they perceive a benefit in participating in the surveillance system and actively report deaths to the key informants. Recently, each key informant from a village or area has been given a turubai (canvas tarpaulin) so that bereaved families from the community can borrow it for funeral gatherings. This has enabled key informants to get information on deaths that occur in their area and report them to the supervisor. The personnel who perform the VAs meet with the key informants on a regular basis to find out about new deaths. They then arrange to meet with the relatives or caretakers of the deceased to verify that the death has occurred, then perform the VA.

Two physicians independently assign a cause of death. Until 1999, a modified version of ICD-10 was used. From 2000, AMMP began to use a shorter, broader list of codes. The diagnoses given by the two coders are compared, and discrepancies are given to a third coder. If all three coders disagree, the form is coded as “unknown.” Wherever possible, confirmatory evidence of the cause of death is obtained. This includes in- and out-patient records, a death certificate, or a burial permit.

Data management

Enumerators meet weekly with their supervisors during a census to assess progress and solve the various problems the enumerators encounter. In addition, during the census, a field supervisor reviews all completed forms and returns those with errors or inconsistencies to the enumerators for correction. Those that pass inspection are sent to the data centre in Dar es Salaam and entered into a computer. All census forms with errors detected during data entry are logged and returned to the field for correction. Once the corrected forms are returned to the office, they are logged back in, and the problems are resolved. Staff are trained to enter data into microcomputers using a data-entry system designed specifically for the project in Microsoft FoxPro. They are

instructed on how the census forms should be completed so that they, too, in addition to the computer validation programs, can detect errors and inconsistencies. The validation programs include simple range checks and checks for inconsistencies across household members, such as an individual identified as a “spouse” but with marital status recorded as “never married.”

Several methods are employed to ensure data quality, including checks in the field and the data-entry process. Supervisors visit a random sample of households to verify entries on the census forms and check that the census includes all the households visited but no nonexistent households. Following each census, reinterviews are also conducted of a sample of households for each enumerator. Because of the large amounts of data collected in a single census, it is impossible to double enter all data for verification; instead, a 5–10% random sample is taken and the forms are checked against the captured data.

The interviewers give each household a newsletter at the end of each interview. It is produced and distributed for US $0.08 per household and contains health-education messages and simplified presentations of results from the previous round. It demonstrates that the DSS is part of the functioning of the district health system. The newsletter is designed to help the communities and their leaders better understand the areas where they live. In 1999, 95% of households reported receiving the newsletter, and 81% of households reported reading it.

Hai DSS basic outputs

Demographic indicators

The primary outputs of the system are estimates of cause-specific mortality for all ages. The DSS has a current population of 154 000, with an annual growth rate of 1.7%. Average household size is 4.6. The male-to-female ratio is 100 : 108, with an age-dependency ratio of 91%. The main age-group structures of the current population are as follows: those <1 year old account for 2.7% of the entire population participating in the DSS; 1–4 years old, 11.1%; 5–14 years old, 27.7%; 15–64 years old, 52.4%; and ≥65 years old, 6.1% (Figure 10.2). Between July 1992 and June 1999, the maternal mortality ratio was 368 per 100 000 live births.

The figures presented below reflect change of residence on an annual basis and do not capture short-term movements between the enumeration rounds. In the year 1998–99, the DSA had an out-migration of 14 951 people. The region of destination was obtained for 12 855 of these people. The most common destination (48%) was another part of Hai, whereas 10% migrated to Dar es Salaam, the commercial centre of Tanzania. The remainder migrated to other parts of Tanzania, except for a few who migrated internationally (mostly to Kenya). Also in that year, 16 575 people migrated to households within the DSA. The place of origin for 73% of them was determined: 53% migrated from areas within Hai, and 686 (5.6% of those who gave a place of origin) came from Dar es Salaam. The remainder came from other parts of Tanzania, except for 88, who came from 10 other countries, mainly Kenya.

Table 10.1 shows the age- and sex-specific all-cause mortality at the Hai DSS site.

Figure 10.2. Population pyramid for person–years observed at the Hai DSS site, Tanzania, 1995–99.

Table 10.1. Age- and sex-specific mortality at the Hai DSS site, Tanzania, 1995–99.

Acknowledgments

This chapter is an output of a project that DFID funded for the benefit of Tanzania and other developing countries, and the views expressed are not necessarily those of DFID.

The AMMP team would like to acknowledge the district health-management team from Hai for their continued support and collaboration. We are also grateful for the contributions and efforts made by the AMMP support staff: Mariana Lugemwa, Dorothy Lyimo, Rukia Mwamtemi, Getrude Peter, Charles William, Mustapha Kahise, and Juma Mfinanga. Finally, we would like to express our sincere thanks to all those who live in the project area for their patience and cooperation.

Chapter 11
IFAKARA DSS, TANZANIA

Joanna Armstrong Schellenberg,^1,2 Oscar Mukasa,¹ Salim Abdulla,¹ Tanya Marchant,¹ Christian Lengeler,² Nassor Kikumbih,¹ Hassan Mshinda,¹ and Rose Nathan¹

Site description

Physical geography of the Ifakara DSA

The Ifakara DSS (latitudes 8°00′–8°35′S, longitudes 35°58′–36°48′E, altitude 270–1000 m above sea level) includes 25 villages of Kilombero and Ulanga districts, in the Morogoro region of southwest Tanzania, about 320 km from Dar es Salaam (Figure 11.1). The area covers 80 km ×18 km in Kilombero District and 40 km ×25 km in Ulanga District, making a total of 2400 km² of Guinea savannah in the floodplain of the Kilombero River, which divides the two districts. The Udzungwa Mountains lie to the northwest. The area has a rainy season from November to May, but rain may fall in any month of the year. Annual rainfall is 1200–1800 mm, and the annual mean temperature is 26°C.

Figure 11.1. Location of the Ifakara DSS site, Tanzania (monitored population, 60 000).

Population characteristics of the Ifakara DSA

The DSA has a population of 60 000 people, living in 12 000 scattered rural households. Its population density is 25 people/km². It has a wide mix of ethnic groups, including Wandamba, Wapogoro, Wabena, Wambunga, and Wahehe. About 40% of the population is Muslim; and 60%, Christian. Although most people speak the language of their own ethnic group, the national language, Swahili, is widely spoken.

The most common occupations are subsistence farming, fishing, and small-scale trading; and rice and maize are the predominant food crops. All villages have government primary schools. The median age at enrollment in school is 8.75 years. Literacy rates among adults are 88% for men and 69% for women. Most local houses have mud walls and thatched roofs, but up to one-third have brick walls and corrugated iron roofs. Most families have a second house known as a shamba house (farmhouse), where they stay during the planting and harvesting seasons. The most common sources of water are shallow wells, open wells, and rivers.

The site has no paved roads, and some villages are cut off for parts of the year as a result of flooding. Limited seasonal bus service runs up to three times each day between the towns of Ifakara, Mahenge, and Malinyi. The Tazara railway links the towns of Ifakara and Mlimba. The site has no telephone service, and most houses have no electricity. Catholic missionary stations in some villages are connected by a radio-call network.

The public health system comprises a network of village health workers, health posts, dispensaries, health centres, and hospitals, offering a varying quality of care. In Ifakara, the capital of Kilombero, the main hospital is a large, well-equipped mission-designated district hospital. The hospital in Mahenge, the Ulanga District capital, has more limited facilities. The mother and child health services are well developed, and vaccination coverage is high, with 74% of children in Kilombero and 63% of children in Ulanga receiving measles vaccines by the time they are 1 year old. Use of health services is also fairly high: 49% of children <5 years old who were reported sick in Kilombero in a 2-week period were taken to a health facility. Use of health facilities is slightly lower in Ulanga, with 39% of sick children being taken to a health facility.

According to health services and local people, malaria is the foremost health problem for both adults and children (Tanner et al. 1991). Malaria transmission from Plasmodium falciparum is intense and perennial, despite marked seasonality in mosquito densities, which peak with the rains. Anopheles gambiae and Anopheles funestus are the main vectors, with an estimated 200–300 infective bites/person a year occurring in rural areas close to Ifakara (Smith et al. 1993). Life-threatening malaria occurs largely in children and commonly in those <1 year old (Schellenberg et al. 1999). Anemia is extremely common: 86% of children <5 years old have some level of anemia (Hb < 11 g/dL), and 9% of children aged 6–11 months have life-threatening anemia (Hb < 5 g/dL). The largest single cause of this anemia is malaria (Menendez et al. 1997).

In 1997, median monthly household expenditure varied from US $77 to US $96, depending on the season, of which about 75% is for food.

In January–May 1999, part of the area suffered a famine, during which emergency food aid was distributed by the government, the World Food Programme, and local nongovernmental organizations. Since mid-1997 a social-marketing program for insecticide-treated mosquito nets to control malaria has been ongoing in Kilombero and Ulanga. The nets are popular, with 54% of children <5 years old using a net and

37% using an ever-treated net. However, most nets are not treated regularly: only 13% of children <5 years old sleep under a recently treated mosquito net.

Ifakara DSS procedures

Introduction to the Ifakara DSS site

The original aim of the DSS was to provide a framework to evaluate a social-marketing program for treated mosquito nets (Armstrong Schellenberg et al. 1999). The current aims are

The DSS started in 1996, with a population of 52 000. The current population (end of 1999) is 60 000. Each household is visited every 4 months, to collect information on pregnancies, births, deaths, and migrations, using the household-registration system (HRS) developed originally in Navrongo, Ghana (Binka et al. 1999). Between household visits, key informants report births and deaths as they occur. No verbal autopsy (VA) questionnaires were used until 2000, although deaths among children born in 1998 and 1999 were followed up with an open question on the events leading up to the child’s death. Educational level, roofing type, and a brief checklist of household possessions are assessed annually. Additional surveys of samples of the population have covered perceptions of malaria, its treatment, and its prevention; household expenditures; willingness to pay for treated mosquito nets; fertility; child fostering; an evaluation of a discount-voucher system for treated mosquito nets; and the effect of treated nets on child survival, malaria, and anemia.

The DSS team currently employs 39 full-time staff: 22 interviewers, 8 supervisors, 2 assistant field managers, 3 data entry clerks, a filing clerk, a driver, a data manager, and a field manager. This team works under the overall coordination of a demographer or epidemiologist. In addition, at the subvillage level, 104 key informants chosen by village leaders are paid a small allowance for every event they report. The interviewers live in the villages where they work, using bicycles for transportation. Supervisors also live in the DSA and hold weekly meetings with managers and data-management staff in Ifakara. Supervisors use motorbikes for transportation. All field staff attend monthly meetings in Ifakara, and key informants attend meetings every 4 months in the DSA.

The Ifakara DSS is a unit within the Ifakara Health Research and Development Centre, which is an independent Tanzanian trust. Its current scientific partners include the Swiss Tropical Institute (STI), the US Centers for Disease Control (CDC), the Tanzanian Ministry of Health, the International Development Research Centre, the Tanzania Essential Health Interventions Project, the Adult Morbidity and Mortality Project, and the World Health Organization (WHO). Funding for the DSS is provided by the Swiss Agency for Development and Co-operation (SDC), CDC, WHO, STI, and the Swiss National Science Foundation (SNSF).

Outputs of the DSS are disseminated locally through a community newsletter (delivered to all households every round) and through meetings with community leaders. Results are also made available to district health-management teams in printed form and through their attendance at district health-planning meetings.

Ifakara DSS data collection and processing

The area was originally selected as a rural area, including parts of two districts. It had an initial target population of about 50 000 people for the evaluation of the social-marketing program for malaria control using treated nets.

Field procedures

INITIAL CENSUS — A baseline census was carried out from September to December 1996. It noted people’s names, sex, dates of birth, and relationships within the household, made sketch maps of household locations, and recorded the rough locations of any shamba houses.

REGULAR UPDATE ROUNDS — Since January 1997 every household has been visited every 4 months by a DSS interviewer, who updates the census record by asking an adult member of the household about in- and out-migrations, pregnancies, births, and deaths. Bereavement interviews (the VAs) on all deaths were introduced in September 2000. A field supervisor carries out these interviews.

CONTINUOUS SURVEILLANCE — Village-based reporters in each kitongoji (subvillage) are a source of information on births and deaths between the visits every 4 months. They record all such events in a notebook. This information is checked and transcribed on a standard form every month by a field supervisor. Village reporters are paid a small sum of money for each event they report and for meeting with their supervisor each month.

FIELD SUPERVISION AND QUALITY CONTROL — Every week, supervisors revisit a randomly selected 10% of the households visited by DSS interviewers and repeat the interview. At the time of these interviews, supervisors do not have access to the original data but use a copy that contains a random number of deliberate errors they should detect and correct. Supervisors also carry out accompanied interviews with a convenience sample of two households for every interviewer every week. Assistant field managers and field managers also carry out spot checks on every interviewer and supervisor at least once each round. They check from a random starting point that every neighbouring household has been registered. Information from the village-based reporters is checked against that from the DSS interviewers.

Data management

All forms are brought to Ifakara, where they are logged by a filing clerk before being processed by the data clerks, who update the databases for every household visit and every event (pregnancy, birth, death, or migration) detected during household visits. Data from each week’s work is entered into the HRS database and processed before the following week’s field meeting. Checking programs are run, and any inconsistencies or

Figure 11.2. Population pyramid for person–years observed at the Ifakara DSS site, Tanzania, 1997–99.

queries are referred back to the field team for correction within 2 weeks of the original interview.

Population, mortality, and fertility data are sent to village leaders every year (more often if requested). Summary information is disseminated to each household in the DSA through a community newsletter. Feedback is given to district-level health workers at semiannual meetings. DSS results reach national and international levels through technical reports and publications.

Ifakara DSS basic outputs

Demographic indicators

Three percent of the population is <1 year old; 16%, 0–4 years old; 26%, 5–14 years old; 53%, 15–64 years old; and 4%, ≥65 years old (Figure 11.2). The age-dependency ratio is 87%, and 51% of the population is female, giving a sex ratio of 97 males for every 100 females. The total fertility is estimated at 4.8 births per woman. During 1999, the infant mortality rate was 90 per 1000 live births, and mortality in children 1–4 years old was 12.9 per 1000 per year. Average household size is 5.0, and 81% of household heads are male. The population is highly mobile, with most families moving to the shamba areas for a few weeks at a time, depending on the farming season.

Table 11.1 shows the age- and sex-specific all-cause mortality at the Ifakara DSS site.

Table 11.1. Age- and sex-specific mortality at the Ifakara DSS site, Tanzania, 1997–99.

Acknowledgments

We wish to thank the residents of the DSA; Dr Lwilla and Dr Mbena, district medical officers of Kilombero and Ulanga districts; and the field and data-room staff of the Ifakara DSS. We also thank SDC, CDC, WHO, STI, and SNSF for financial support.

Chapter 12
MOROGORO DSS, TANZANIA

Robert Mswia, Gregory Kabadi, David Whiting, Honorati Masanja, and Philip Setel¹

Site description

Physical geography of the Morogoro DSA

Morogoro District is situated in Morogoro region, about 180 km from Dar es Salaam. Morogoro has a low population density and mixed topography, which includes mountains and plains (Figure 12.1). It covers an area of 19 250 km² and has 10 administrative divisions, divided into 43 wards, each of these divided into 215 registered villages. The Morogoro rural DSS site lies between latitudes 6.60° and 7.29°S and longitudes 37.35° and 38.30°E. The surveillance area covers 61 of the 215 registered villages. These are in three divisions — Ngerengere, Kingolwira, and Mlali. The villages cover a wide area, including the lowlands and slopes of the Uluguru mountain range. The most isolated villages (Kidunda and Usungura) are close to the Selous Game Reserve, about 160 km away from district headquarters.

Figure 12.1. Location of the Morogoro DSS site, Tanzania (monitored population, 120 000).

Population characteristics of the Morogoro DSA

Currently, the surveillance area has a population of 120 000, with an annual growth rate of 3.6%. The current number of households is 31 000, and the average household size is 4.0. A household is defined as “people eating from the same cooking pot.” About 18% are single-person households. The area is generally poor, rural, and is among the 50% most deprived regions in Tanzania, according to the poverty and welfare indicators for 1999 of the Vice President’s Office. The main ethnic groups are the Luguru, Sagara, and Pogoro. The population participating in surveillance, however, comprises a wide mixture of ethnic groups. The religious groups in the area are Muslims (57%), Christians (41%), and others (2%). Indigenous languages are commonly spoken in the villages, but the national language, Kiswahili, is widely understood and spoken.

The main occupation of people of all ages in the area is farming, including 45.2% of males and 52.7% of females. The proportion of girls attending school is slightly higher than that of boys for all ages up to 14 years. The proportion of people from age 15 stating that they had no formal education is 65% for women but only 35% for men.

About 40% of the households in the Morogoro DSS use tap water (34.8% public tap, 2.7% neighbour’s tap, and 3.2% own tap); 32.3%, river or rain water; and 26.9%, wells. More than 90% use pit latrines. Less than 1% of the households in the project area use electricity or gas as the main cooking fuel. The majority (90%) use firewood, and the remainder use kerosene stoves or charcoal for cooking. Some of the villages in Morogoro surveillance area have access to electricity, especially those along the main roads and those with health facilities.

Transportation in the district is mainly by road. The roads from Dar es Salaam to Dodoma and Iringa pass through the district. Most of the other roads are unsealed and difficult to travel along during the rainy season. Morogoro District has 3 hospitals, 6 health centres, and 81 dispensaries.

The main causes of death in the area are acute febrile illness (including malaria), diarrheal diseases, HIV–AIDS, injuries (both intentional and unintentional), anemia, pulmonary TB, and malnutrition.

Morogoro DSS procedures

Introduction to the Morogoro DSS site

DSS work is carried out in Morogoro to provide reliable population denominators for continuous cause-specific mortality surveillance. The demographic and mortality monitoring provides district authorities with information on the burden of disease, health-facility use in the period before death, and population conditions. These data are used for evidence-based planning and evaluation of health services. The monitoring is an activity of the Tanzanian Ministry of Health and the district health-management team, as part of the Adult Morbidity and Mortality Project, phase 2 (AMMP-2). The goal of AMMP-2 is to decrease the morbidity and mortality stemming from conditions particularly likely to cause suffering and disadvantage among Tanzania’s poor people, where these conditions are amenable to health-service interventions. To contribute to this goal, the project aims to strengthen evidence-based planning and development of

cost-effective health services within the context of health-sector reform in project districts and the Ministry of Health in Tanzania.

Demographic and mortality surveillance in both phases of AMMP has been carried out in Hai District (Kilimanjaro region) and Dar es Salaam, as well as in Morogoro (see Chapters 9 and 10). In 1992, the DSS site in Morogoro had a population of 99 000, but it has since grown to a current total of 120 000. Whereas the initial focus was on adult health, the system now collects data on people of all ages.

Demographic and mortality surveillance began in 1992. Since then the DSS has had one enumeration round each year. The system is incorporated into both national and district structures. In the Ministry of Health, the National Sentinel System assumes overall responsibility for ascertaining a national picture of the burden of disease, using the DSS to gather demographic and mortality data. The Hai and Dar es Salaam sites are also becoming a part of this sentinel system, and for the time being the Rufiji site contributes cause-specific mortality data. Surveillance work will become part of the routine systems of the district. Mortality surveillance will continue indefinitely, and the DSS will continue as long as the site is without a cost-effective alternative for gathering reliable population denominators.

Morogoro DSS data collection and processing

The size of the population in the DSS approached the level that Hayes et al. (1989) suggested is best for the ascertainment of cause-specific mortality. Within the original set of three AMMP-supported areas, Morogoro was chosen to represent poor rural living conditions and low population density.

Field procedures

INITIAL CENSUS — Because the 1988 national census did not provide an accurate basis for estimating population denominators, an initial census round was carried out in 1992. The baseline census was taken to determine who was present in each household under surveillance. Since then the population has been enumerated once a year for an 8-week period beginning each August. Whereas the census-update rounds take place annually, mortality surveillance to provide information on probable causes of death is continuous. Probable cause of death is determined using the verbal-autopsy (VA) technique. About 86 villagers, most of them village health workers and primary-school teachers, are paid a small remuneration for acting as enumerators for the census-update rounds and as key informants to report deaths to the VA supervisory team. Four clinical-health officers from the district constitute the VA supervisory team.

REGULAR UPDATE ROUNDS — In subsequent census rounds, information from the previous round is printed on new forms for each household. Each household is visited, and an adult member of the household is interviewed. The enumerators verify and, where necessary, update existing data. When new households appear as a result of either migration into the area or splitting of existing households (for example, through marriage), they are registered on new-household forms. Vital events (births and deaths) and migrations are recorded for each household. The following items of data are recorded for each individual during a household visit: name, age, sex, relationship to head of household, main occupation, marital status, alcohol consumption and smoking

habits, date of entry into the household, mode of entry, date of exit, mode of exit, and whether his or her parents are alive. Recently, questions on religion have been added to increase knowledge of the social characteristics of the population. Migration tracking is limited to recording the date of entry into and exit from the area and the district of origin or destination; successive migrations of individuals into and out of the area are not linked. It is thus possible to determine who is resident at any point in time (and therefore to calculate denominators) but not to calculate the total time each individual has spent in the surveillance area.

CONTINUOUS MORTALITY SURVEILLANCE — The primary objective of the AMMP approach to DSS is to inform health planning and priority-setting by providing sentinel data on the burden of disease. Therefore, an effort is made to determine the cause of death for each resident. This is achieved by interviewing the relatives and caretakers of deceased people, using a short, standard interview schedule. Different forms are used for deaths of infants <31 days old, of children between 31 days and <5 years old, and of all persons ≥5 years old. The forms contain a section to identify the respondent, another to identify the deceased, an open-ended history section, a checklist of previously diagnosed conditions, a checklist of symptoms and their duration, a list of health services sought in the period leading to the death, a residential history, and a summary of any confirmatory evidence, such as medical records or a death certificate. The form is completed by trained health personnel, who interview one or more relatives or caretakers. Wherever possible, the interview takes place within 6 weeks of the date of death. Deaths are usually reported by community-based key informants. The key informants are chosen because of their awareness of events in their community; that is, they are likely to hear of deaths that occur. In addition, the communities receive feedback in a newsletter; they consequently perceive a benefit from taking part in the surveillance system and actively report deaths to the key informants. The personnel who perform the VAs meet with the key informants on a regular basis to find out about new deaths. They then arrange to meet with the relatives or caretakers of the deceased to verify the death and then perform the VA. Two physicians independently assign a cause of death. Until 1999, a modified version of ICD-10 was used. From 2000, a shorter, broader list of codes, developed by AMMP, has been used. The diagnoses given by the two coders are compared, and discrepancies are given to a third coder. If all three coders disagree, the cause of death is coded as “unknown.” As noted, confirmatory evidence of the cause of death is obtained, whenever possible.

Data management

During the census a field supervisor reviews all completed forms, and those with errors or inconsistencies are returned to the enumerators for correction. Those that pass inspection are sent to the data centre in Dar es Salaam and entered into a computer. All census forms with errors detected during data entry are logged and returned to the field for correction. Once the corrected forms are returned to the office, they are logged back in, and the problems are resolved.

Staff are trained to enter data into microcomputers using a data-entry system designed specifically for the project in Microsoft FoxPro. They are instructed on the correct completion of the census forms so that they, too, in addition to the computer validation programs, can detect errors or inconsistencies. The validation programs include simple range checks and checks for inconsistencies across household members,

such as an individual identified as a “spouse” but with a marital status recorded as “never married.”

Several methods are employed to ensure data quality, including checks in the field and data-entry processes. Supervisors visit a random sample of households to verify entries on the census forms to check that all households visited have been included in the census and that no nonexistent households have been included. Following each census, reinterviews are conducted of a sample of households for each enumerator. Because of the large amount of data collected during a single census, it is impossible to double enter all data for verification; instead, a 5–10% random sample is taken, and the forms are checked against the captured data.

Each household is given a newsletter at the end of each interview. This newsletter is produced and distributed at a cost of about US $0.10 per household and contains health-education messages and simplified presentations of results from the previous round. The newsletter demonstrates that the DSS is part of the functioning of the district-health system and is designed to help the communities and their leaders better understand the areas where they live. In 1998, 94% of households reported receiving the newsletter, and 65% reported reading it.

Morogoro DSS basic outputs

Demographic indicators

The primary outputs of the system are estimates of cause-specific mortality for all ages. The DSS shows that the current population of the surveillance area is 120 000 and has an annual growth rate of 3.6%. The proportions of the population in the various age groups are as follows: <1 year old, 2.6%; 1–4 years old, 9.7%; 5–14 years old, 26.4%; 15–64 years old, 56.3%; and ≥65 years old, 5.0%. The ratio of males to females is 100 : 103, and the age-dependency ratio is 78.6%. The infant mortality rate is 99.7 per 1000 live births, and the under-five mortality rate is 39.6 per 1000. The maternal mortality rate is 1183 per 100 000 live births for the period between July 1992 and June 1999. The average household size is 4.0 with a headship of 73% males and 27% females.

The shape of the population pyramid (Figure 12.2) in the two rural Tanzanian districts, based on AMMP census techniques, shows a narrowing in the base over time; that is, the proportion of the population <5 years old is less than expected, assuming “typical” developing-country conditions of high-fertility and a growing population. This narrowing effect may be due to real factors or to the under-enumeration of infants and young children in the annual census round, or to both. Some possible contributing factors would be rapidly declining fertility (as a result of both HIV–AIDS and so-called secular trends) and higher child mortality (perhaps partly a result of the HIV–AIDS pandemic). In addition, the narrowing effect is most pronounced in villages along the major east–west highway, suggesting that part of the explanation may be the mobility of young people in these areas. If this aspect of the population structure is an artifact of under-ascertainment of infants, our estimates of infant mortality will be too high. If need be, this could be corrected using indirect methods.

Table 12.1 shows the age- and sex-specific all-cause mortality at the Morogoro DSS site.

Figure 12.2. Population pyramid of person–years observed in the Morogoro DSS site, Tanzania, 1995–99.

Table 12.1. Age- and sex-specific mortality at the Morogoro DSS site, Tanzania, 1995–99.

Migration depends on the time used to define migration events. The following figures reflect changes of residence on an annual basis and do not capture short-term movements between the enumeration rounds. During 1998–99, 10 896 people migrated out of the surveillance area. The region of destination was obtained for 7887 of these: most (65%) migrated to another part of Morogoro, and 15% migrated to Dar es Salaam, the commercial centre of Tanzania. The rest migrated to various parts of the country, except for 13 people who migrated to other countries. During the same years, 15 585 people migrated to households within the surveillance area. The place of origin for 11 298 (72.4%) was determined: a similar proportion (67%) migrated from areas within Morogoro, and 809 (7.2% of those who gave a place of origin) came from Dar es Salaam; the rest came from various other parts of Tanzania, and just 4 came from other countries.

Acknowledgments

This chapter is, in part, an output of a project that DFID has funded for the benefit of Tanzania and other developing countries, and the views expressed are not necessarily those of DFID.

The AMMP team would like to acknowledge the district health-management team from Morogoro for its continued support and collaboration. We are also grateful for the contributions and efforts of the AMMP support staff: Mariana Lugemwa, Dorothy Lyimo, Rukia Mwamtemi, Getrude Peter, Charles William, Mustapha Kahise, and Juma Mfinanga. Finally, we would like to express our sincere thanks to all those who live in the project area for their patience and cooperation.

Chapter 13
RUFIJI DSS, TANZANIA

Eleuther Mwageni, Devota Momburi, Zaharani Juma, Mohamed Irema, Honorati Masanja, and the Tanzania Essential Health Interventions Project and Adult Morbidity and Mortality Project teams

Site description

Physical geography of the Rufiji DSA

The Rufiji DSA extends between latitudes 7.47° and 8.03°S and longitudes 38.62° and 39.17°E. The Rufiji DSS is in Rufiji District, Tanzania, about 178 km south of Dar es Salaam (Figure 13.1). Rufiji is one of the six districts of the coast region, the others being Bagamoyo, Kibaha, Kisarawe, Mafia, and Mkuranga. Rufiji, in the south, has 6 divisions, with 19 wards, divided into 94 registered villages and 385 hamlets. The district covers an area of about 14 500 km². The Rufiji DSS operates in 6 contiguous wards and 31 villages (about 60 km long × 30 km wide) and covers an area of 1813 km².

Figure 13.1. Location of the Rufiji DSS site, Tanzania (monitored population, 85 000).

Rufiji has a mean altitude of <500 m above sea level. Its vegetation is mainly tropical forest and grassland. The district has hot weather throughout the year and two rainy seasons: short rains (October–December) and long rains (February–May). The average annual precipitation in the district is 800–1000 mm. A prominent feature of the district is the Rufiji River, with its large floodplain and delta, the most extensive in the country. Mangrove forests flank the tributaries of the delta. The river, from which the district takes its name, divides the district geographically into roughly equal halves. The district is also a gateway to Selous Game Reserve, which has a variety of wild animals, such as zebras, buffalo, hartebeest, monkeys, lions, hyenas, warthogs, and elephants.

Population characteristics of the Rufiji DSA

Rufiji has a population of about 182 000, of which 85 000 (about 47% of the district) is under surveillance. The population densities for the district and for the surveillance area are 12.5/km² and 46/km², respectively. The mean household size for the whole district is about 5.0 (TBS 1994). The district is largely rural, but the population is clustered around Utete (district headquarters), Ikwiriri, Kibiti, and Bungu townships (see “Rufiji DSS basic outputs,” below, for DSS-generated demographics).

Rufiji District is home to several ethnic groups. The largest of these is the Ndengereko, who, according to oral tradition, are the original inhabitants of the area; other groups include the Matumbi, Nyagatwa (concentrated in the delta area), Ngindo, Pogoro, and Makonde. The majority of the people are Muslims, with a few Christians and followers of indigenous religions. In addition to local languages, Kiswahili is widely spoken; English is not commonly used in the area.

The majority of the people in Rufiji are subsistence farmers. Farming areas are often set some distance from the family home to take advantage of periodically flooded alluvial soils. With temporary houses located on farmland, this means that some households are often split geographically for up to 4 months of the year. Major crops include cassava, maize, rice, millet, sesame, coconut, and cashew nuts. Fruit such as mangoes, oranges, pineapples, papaya, and jackfruit are also grown. Some residents are involved in fishing or small-scale commercial activities, such as selling wood products (for example, timber, furniture, and carvings).

Each village has at least one primary school (with standard grades 1–7). The district has four secondary schools (three government and one private), of which two are located in the DSA. A Folk Development College — a postprimary polytechnic — is located in Ikwiriri township. According to the 1988 population census (TBS 1994), more males (66%) than females (34%)are literate in the district.

Most villages in the surveillance area have a central place for shops and a market. The dwellings are simple, comprising a mixture of huts with walls made of mud and wooden poles, with thatched or corrugated roofs, as well as conventional brick houses in the townships. In the Rufiji floodplain, dungus (traditional shelters on stilts built to deal with flooding) are a common feature. Tap-water supply is very limited, and the majority of people rely on communal boreholes or use natural-spring or river water for domestic purposes — a few use harvested rainwater. The DSA’s main transportation route is the north–south Dar es Salaam – Lindi and Mtwara trunk road, half of which is paved; and the other, unsealed. Unpaved feeder roads and tracks link most of the villages to this trunk road. Telephone facilities in the district are located in the

three townships. The district is not connected to the national electric grid, but Ikwiriri township has 24-hour diesel-generated electric power from the national electric-supply company. Other places that have electricity depend on private generators.

The district has 55 health facilities: 2 hospitals (1 government and 1 mission), 5 government health centres, 44 government dispensaries, and 4 nongovernment dispensaries. A private dispensary based at Kibiti offers the services of a mobile clinic in some parts of the district. Over-the-counter drugs are available from many private shops and kiosks in the villages. Many people also obtain services from traditional healers, including traditional birth attendants. Malaria and waterborne diseases, such as cholera and diarrhea, are the major health problems in the area, according to both the health services and local people. Major causes of mortality include acute febrile illnesses (including malaria), acute lower-respiratory infections, tuberculosis, AIDS, and perinatal illnesses. Immunization coverage ranges from 85% for BCG (tuberculosis) to 66% for measles in children 12–23 months old. About 89% of the population lives within 5 km of a formal health facility. All villages and health facilities in the district have been positioned using GPS and mapped in a GIS database of district health resources.

Rufiji DSS procedures

Introduction to the Rufiji DSS site

The objectives of the Rufiji DSS are to provide sentinel data for health policy and planning and to monitor the impact of health reforms. Data and experiences from the Rufiji DSS are being assessed for their use in assisting district health-management teams, policymakers, and planners to make more appropriate resource allocations to improve the health situation in the district and the country as a whole. With the Adult Morbidity and Mortality Project (AMMP), experiences from the Rufiji DSS are also informing the development of methods for the National Sentinel System, which monitors the burden of disease in Tanzania. The Rufiji DSS commenced field operations in November 1998.

The Rufiji project employs the DSS to collect health-status and demographic data. The DSS approach involves a continuous surveillance of households and members within households in cycles or intervals, known in the Rufiji DSS as “rounds,” of 4 months each. Members (or residents) of the Rufiji DSS are individuals who have resided in the surveillance area for a period of the previous 4 months. The Rufiji DSS collects information on demographic, household, socioeconomic, and environmental characteristics of the population. Verbal autopsies (VAs) conducted on all Rufiji DSS-registered deaths, using specific standard questionnaires, determine cause of death. The VA instruments and coding procedures used in the Rufiji DSS are identical to those used by AMMP.

The Rufiji DSS has a team of 52 people, who are entirely district based. The staff, headed by a station manager, is organized into three groups: field (field manager, 7 enumerator supervisors, 3 VA supervisors, 4 migration supervisors, and 25 enumerators); data (data manager, data assistant, filing clerk, and 3 data-entry clerks); and support (accountant, secretary, driver and mechanic, cleaner, and security guard). Most fieldworkers are deployed throughout the DSA, whereas the data and support teams are based in the field station in Ikwiriri township, south of the DSA.

The Rufiji DSS also has access to about 118 key informants. These are community leaders, whose responsibilities are to assist the field staff in reporting births or deaths in their respective areas and sometimes in finding prospective households for inclusion in the DSS.

The project is comanaged by the Tanzania Essential Health Intervention Project (TEHIP) (funded by the International Development Research Centre, Canada) and AMMP (funded by the Department for International Development, United Kingdom). Both TEHIP and AMMP are projects under the auspices of the Tanzania Ministry of Health. AMMP is implemented in partnership with the University of Newcastle upon Tyne (United Kingdom).

The local district health-management team, the Tanzania Ministry of Health, and national and international collaborative research-and-development projects are the main consumers of Rufiji DSS data.

Rufiji DSS data collection and processing

Field procedures

MAPPING — The Rufiji DSS employed a nonrandomized, purposive technique in selecting the wards under surveillance. It covers the total population in the six contiguous wards of Bungu, Ikwiriri, Kibiti, Mchukwi, Mgomba, and Umwe and operates exclusively to the north of the Rufiji River, which flows along a roughly west–east axis through the district. This side of the river is home to the majority of the population and is more easily accessible throughout the year, whereas communities south of the river, as well as those in the delta, may be inaccessible for varying periods during the long rains. The Rufiji DSS targeted an initial population of 70 000, which was set to provide mortality data similar to those from other DSS sites in the AMMP (see Hayes et al. 1989 and Chapters 9, 10, and 12). Given an average household size of 4–5, it was estimated that the DSS would need to include 14 000–17 500 households. All villages have been positioned using GPS. Mapping of households is planned.

INITIAL CENSUS — The Rufiji DSS data collection began with enumerators conducting an initial census in the sampled area to establish the baseline population. This population forms the foundation for establishing a longitudinal DSS and provides background data on the population. The census data are obtained using standard questionnaires with both closed- and open-ended questions. The enumerators collect data on household (household head, relation to household head), demographic (age, sex), socioeconomic (education, occupation), and environmental (source of drinking water and sanitation facility) conditions. For purposes of identity, each registered household and person is given a unique number to distinguish the household within its village and the individual within his or her household. The unique number for each individual is known as the “permanent ID” and comprises IDs of the village and household and the number for the individual within the household.

REGULAR UPDATE ROUNDS — Longitudinal data collection of demographic, household, socioeconomic, and environmental characteristics is maintained through subsequent update rounds. These rounds take 4 months to complete; the day after one round finishes the next round begins, and households are visited in sequence. Update rounds

are undertaken to maintain accurate denominators for estimation of age, sex, and cause-specific death rates. In their periodic visits, enumerators register new people found in the households. These include unregistered individuals who could have been missed during the initial census. During the rounds the enumerators verify the status of each household and individual, using the household-registration books (HRBs), and, if necessary, change their records. The enumerators make all alterations in the respective HRBs, in conjunction with filling in a changes form.

CONTINUOUS SURVEILLANCE — The Rufiji DSS involves the continuous recording of vital events within households and among members over time. These events, recorded by enumerators using specific event forms, include births, deaths, pregnancies, pregnancy outcomes, marital-status changes, and migrations (in and out of the surveillance area). In addition, lay key informants assist the enumerators by independently recording births and deaths in their hamlets.

VA interviews on all DSS-registered deaths are conducted by VA supervisors, using specific standard questionnaires for deaths of infants <31 days old; children between 31 days and <5 years old; and all persons ≥5 years old. The interviews are held with one of the adult relatives of the deceased (preferably a caretaker) well informed of the sequence of events leading up to the death. VA supervisors conduct interviews within 2 months of the report of a death and use any available documents, such as a death certificate or prescriptions, to obtain confirmatory evidence about the cause of death from the last health facility the deceased visited before dying. Such evidence, however, is often unavailable. The completed questionnaires are then coded independently by two physicians, according to a list of causes of death, based on the 10th revision of the International Classification of Diseases. A third physician is asked to independently code the cause of death in the case of discordant results. Where there are three discordant codes, the cause is registered as “unknown.”

SUPERVISION AND QUALITY ASSURANCE — The field manager supervises all field operations and spends about 60% of the time supervising field activities and the remainder in the field station’s office. On completion of interviews or household visits, field supervisors randomly select and revisit 3–5% of the households interviewed or updated by the enumerators, for quality control. Errors noted are communicated immediately to enumerators or brought up for discussion during the regular bimonthly field-staff meetings.

Data management

The Rufiji DSS data-collection process uses a variety of forms. These forms include baseline census, event, changes, HRBs, and VA questionnaires. A reliable mechanism is in place to ensure smooth production and flow of these forms between the field and the DSS data centre. The DSS filing clerk is responsible for ensuring production and distribution of the forms to the field staff. On completion of the data collection, the supervisors or the filing clerk take the forms to the Rufiji DSS data centre, where the forms are registered by the filing clerk before data entry.

Data management of the Rufiji DSS uses standard, public-domain household-registration system (HRS) computer software, with built-in reporting and checking routines (Indome et al. 1995). The HRS is capable of maintaining a consistent record

of vital events occurring among people in a fixed geographic area and generating up-to-date registration books for field use. Once enumerators have completed their interviews, the data are taken to the Rufiji DSS data centre for entry, and the data entered are then printed in loose sheets or forms, known as HRBs. The filing clerk systematically arranges the HRBs by household and hamlet to facilitate the fieldwork and household interview. The HRB is printed so that it can be used in three rounds of interviews. Likewise, completed VA forms are double entered in the DSS data centre, the differences are reconciled, and then the forms are dispatched for physician coding and returned to the DSS data centre for final processing.

The software for data entry has a built-in series of logical checks and menu-driven procedures to maintain the consistency of the event data with data in the database. For example, the HRS will disallow data entry of a pregnancy of a male resident. To optimize quality, field activities are performed in conjunction with data operations. Completed forms from the field are taken to the DSS data centre for data entry. Errors noted during data entry are verified, reported to the field supervisors for diagnosis, and then corrected both in the field and at the DSS data centre.

The HRS software is also used for data analysis. The software can compute basic demographic rates, such as fertility, mortality, in- and out-migration, and person–year denominators. If all the field and data protocols are followed, fully edited and cleaned data should result at the end of each 4-month round. The data can be used to describe characteristics of the population — such as age, sex, marital and parental relations, and household headship — and the dynamics of birth, death, migration, and nuptiality. The addition of the mortality surveillance using the VA allows the generation of cause-specific mortality rates and other measures of disease burden (such as years of life lost) for all ages and both sexes. The findings obtained are presented to the community in simple tables or graphics through biannual newsletters issued to every household in the surveillance area. In addition, TEHIP reprocesses the findings into intervention-addressable shares of the burden of disease, before they are given, in graphical format, to the Rufiji District health-management teams and the Ministry of Health.

Rufiji DSS basic outputs

Demographic indicators

The Rufiji DSA now has a population of about 85 000 and an annual population growth rate of 2.3%. This means the population in the DSA will take about three decades to double. The age and sex composition of the area is presented in the population pyramid in Figure 13.2. The pyramid reveals a broad base that tapers toward the older ages, indicating that the population is young. The population structure is as follows: <1 year old, 2.7%; 0–4 years old, 16%; 5–14 years old, 30%; 15–64 years old, 46%; and ≥65 years old, 8%. The male–female ratio is 92.7 : 100. The DSA has more females (52%) than males (48%). The age-dependency ratio is 110. The total fertility rate is 6.2 children per woman 15–49 years old. The infant-mortality rate is 102.1 per 1000 live births. The under-five mortality ratio is 133 per 1000 live births. The under-five mortality rate is 32.7 per 1000 children. Average household size is 4.8. Males are more likely to be heads of households (73%) and educated (57%) than females (27% and 43%, respectively), and 26% of the population has migrated out of, or into, the DSA.

Figure 13.2. Population pyramid for person–years observed at the Rufiji DSS site, Tanzania, 1999.

Table 13.1. Age- and sex-specific mortality at the Rufiji DSS site, Tanzania, 1999.

Table 13.2. Age-specific fertility rate at the Rufiji DSS site, Tanzania, 1999.

Note: GFR, general fertility rate; TFR, total fertility rate.

Out-migration exceeds in-migration in the DSA. The propensity to migrate into the DSA is higher among females (57%) than males (43%). These percentages are the same for out-migration.

All-causes mortality data for the Rufiji DSS are summarized in Table 13.1. The table reveals that mortality in the DSA is fairly high. The mortality pattern is U shaped, indicating high death rates among children (<5 years old) and adults ≥65 years old. Variations occur in the mortality-age profile between men and women, with women 20–44 years old having higher probabilities of dying than men.

The age-specific fertility rates (ASFRs) and total fertility rate of the DSA are presented in Table 13.2. The ASFRs show a regular feature, with a childbearing peak occurring among women 20–24 years old and fertility levels declining thereafter. ASFR distributions can be classified into three broad groups: early-peak type (20–24 age group), late-peak type (25–29 age group), and broad-peak type (where ASFRs in the 20–24 and 25–29 age groups differ slightly) (Kpedekpo 1982). One notes that fertility levels in the Rufiji DSS are of the early-peak type. This indicates that in the DSA women marry or begin childbearing early in life.

Acknowledgments

This work is supported in part by a grant from the International Development Research Centre (IDRC), Canada, through TEHIP, in collaboration with the Tanzanian Ministry of Health. It is also an output of AMMP. AMMP is a project of the Tanzanian Ministry of Health, funded by the Department for International Development (DFID), United Kingdom, and implemented in partnership with the University of Newcastle upon Tyne, United Kingdom. The views expressed are not necessarily those of the Ministry of Health, IDRC, or DFID.

The Rufiji DSS team includes Ali N. Mangara, Amina S. Mtumbuka, Amiri B. Msati, Antonia M. Shayo, Asha Juma Mzoa, Athumani M. Mwinyihija, Baraka R. Bashir, Cecilia R. Makwaia, Denis Navakongwe, Devota B. Momburi, Eleuther Mwageni, Ephrem Mapunda, Fikiri M. Mtandatu, Fredrick A. Swilla, Grace A. Massawe, Hamisi A. Milandu, Hamisi Sodangu, Hashim M. Kalungo, Hermenegilda D. Mtena, Jafari A.

Mpwapwa, Jane I. Masumai, Julieth L. Kulanga, Kahema I. Nassoro, Kulwa L. Francis, Liberati M. Kahumba, Makala M. Mbura, Manitu M. Malekano, Maua H. Msango, Mohamed Y. Kitambulio, Moshi B. Kitingi, Muhidin B. Mlanzi, Mwajuma N. Mkundi, Mwanate A. Dyandumbo, Mzuzuri Mrisho, Nivone Kikaho, Nuhu A. Kihambwe, Omari S. Matimbwa, Omari S. Mkumba, Omari S. Mnete, Peter S. Ndali, Priscilla F. Mlay, Ramadhani Makutika, Said H. Putta, Sharifa O. Sobo, Sihaba S. Ngabunzwa, Subilaga A. Mwaisela, Tabley N. Tangale, Tumu Nindi, Uwesu Mohamed, Wabishi M. Nyangalilo, Yahya K. Mkilindi, and Zaharan Juma.

The Rufiji DSS team wishes to acknowledge the financial, technical, management, and administrative support of TEHIP (Don de Savigny, Harun Kasale, Robert Kilala, Victor Lihendeko, Conrad Mbuya, Godfrey Munna, Graham Reid, and Elimamba Tenga) and AMMP (Yusuf Hemed, Regina Kutaga, Honorati Masanja, Hamisi Mponezya, Robert Mswia, Ferdinand Mugusi, Philip Setel, and David Whiting). We also wish to express our gratitude to INDEPTH and the Navrongo Health Research Centre, Ministry of Health, Ghana, and in particular, Fred Binka, Felix Kondayire, Pierre Ngom, and Peter Wontuo for their technical exchange visits and support in establishing the Rufiji DSS. Finally, we are grateful for the continuing collaboration with the Rufiji District council, the Rufiji District health-management team, the district medical officer, Dr Saidi Mkikima, and the entire population in the Rufiji DSA for their continued support and collaboration.

Chapter 14
GWEMBE DSS, ZAMBIA

Site description

The Gwembe Tonga Research Project (GTRP) was initiated in 1956 by Elizabeth Colson and Thayer Scudder to study the impact of resettlement associated with the creation of Lake Kariba. Initially, seven villages were chosen as intensive study sites, but as the study progressed the number was reduced to four. Two of the four villages, Sinafala and Siameja, moved relatively short distances and were relocated near or a few kilometres away from Lake Kariba. The other two villages, Mazulu and Musulumba, had to move about 160 km downstream to a site below the dam.

Physical geography of the Gwembe DSA

The primary portion of the Gwembe Tonga DSS study area is located between latitudes 16° and 18°S and longitudes 26° and 29°E in the Southern Province of Zambia (Figure 14.1). The study area encompasses four study villages (and many other villages not in the study) each covering an area of several square kilometres. The villages are scattered for 300 km along the length of the Gwembe Valley, a relatively low-lying

Figure 14.1. Location of the Gwembe Tonga DSS site, Zambia (monitored population, 15 000).

semi-arid area with an average elevation of about 400–500 m above sea level. The Gwembe Valley contains the Zambezi River, and, after 1958, Lake Kariba, which resulted from the construction of the Kariba Dam. Rainfall occurs between October and March and is variable. Small droughts and yearly periods of hunger during the rainy season are more common than not. Temperatures range from near 0°C at night during the cold season (June–August) to 40°C and above during the hot, wet season (November–March). Between August and November, it is hot and dry.

In addition to the rural villages in the Gwembe Valley, the Gwembe Tonga DSS follows migrants and emigrants to the urban areas of Lusaka and Ndola and also to a rural frontier settlement area called Chikanta, located on the plateau several hundred kilometres northwest of the Gwembe Valley study area.

Population characteristics of the Gwembe DSA

Over the more than 40 years since the study began, 15 000 people have been observed, and 10 000 of them were still alive in 1995, the last date, at the time of writing, for which analysis was carried out. The population density varies considerably from high-density villages and urban settings to fairly low-density rural areas.

The main ethnic group inhabiting the study area is the Gwembe Tonga, who practice various indigenous religions, various forms of Christianity, and combinations of the two. Most of the study population is engaged in subsistence and small-scale cash-crop agriculture, but a minority obtain jobs in the rural areas. Many young people migrate to urban areas to seek wage labour, and there is some circular local migration to access seasonal wage labour available with large commercial farming enterprises. Almost all people inhabiting the study area have access to primary education and, if they can afford it, secondary education as well. Few can complete secondary education, and very few move on to tertiary education.

Access to the villages varies: two are readily accessible via tarred and short dirt roads; and two are much less accessible, via poor-quality dirt roads only. Homes are generally built of mud brick and have thatch roofing. In the rural areas, water is provided from shallow wells, the Zambezi River, Lake Kariba, and, in rare instances, boreholes. The urban water supply comes mainly from boreholes and the Kafue River. Most water is of poor quality. None of the rural villages have electricity, although some structures do.

All villages have a clinic within or nearby, but the service varies considerably because of inadequate staff and supplies. There are district hospitals of reasonable quality, although access can be difficult, depending on the season. Additionally, the cost of medical treatment at the hospitals is a burden to most villagers, especially since the International Monetary Fund’s (IMF’s) structural adjustment in 1992. Immunization programs have been fairly reliable and reasonably successful, but exact figures cannot be quoted. Perinatal clinics operate weekly or biweekly in most of the villages. The villagers themselves rate hunger as their primary health problem, followed by malaria, dysentery, and HIV.

Since the study began, in 1956, the population has faced forced relocation, several measles and cholera epidemics, the war for Zimbabwean independence (waged in part where the population lives), the economic downturn that began in the 1970s, severe droughts in the early 1980s and mid-1990s, the IMF’s structural-adjustment programs, and now the HIV–AIDS epidemic, which began having an impact during the early 1990s.

Gwembe DSS procedures

Introduction to the Gwembe DSS site

The Gwembe Tonga Research Project was designed to document the way of life of the Gwembe Tonga before they were forcibly relocated to make space for Lake Kariba in 1956 and to document their adaptation to the new situation after they were relocated in 1958. The study was designed and initiated by anthropologists Elizabeth Colson and Thayer Scudder and was largely conceived as an investigation into social change and adaptation. Although the core focus has always been on social and socioeconomic issues, significant components have been added in the areas of nutrition, growth, development processes, and demography.

Gwembe DSS data collection and processing

Field procedures

INITIAL CENSUS — The initial census was conducted in 1956, and it enumerated the entire population of the four villages remaining in the study: Mazulu, Musulumba, Siameja,¹ and Sinafala.

REGULAR UPDATE ROUNDS — Until 1995, data were collected at roughly 3-year intervals by complete enumerations that updated information on the original population, all of its descendents, and those who had married someone descending from the original population. As a result of the anthropologists’ methodology — recording accurate genealogies for their work — the study population consists of all the original inhabitants of the four villages plus all their direct descendents and those who have married into their families. Hence, this is a genealogically defined sample predominantly. In addition to the genealogically defined group, the study also includes a relatively small number of people who have moved permanently into the geographical and social boundaries of the village but are not directly related to anyone in the original enumeration.

CONTINUOUS SURVEILLANCE — Starting during the 1970s, local informants in each village have kept records of vital events between the major updates and kept daily diaries describing a range of activities in the village.² Since 1995 a more typical data-collection system has been initiated, consisting of event-specific questionnaires. From 1995 on, two interviewers in each village³ have been employed full time to record all vital, nuptial, and migration events and to administer a long questionnaire, once each year, to

¹ The initial enumeration of Siameja captured roughly half of the original population of that village, and it is that half that has been followed ever since.

² The local informants in Siameja began recording information later than those in the other three villages.

elicit a range of socioeconomic indicators. Additionally, the prices of a large range of daily consumables are recorded on a quarterly basis. These will be used to construct a local price index to correct monetary transactions for local inflation.

The current questionnaires are designed to record information on birth, death, migration, initiation of marital union, marital separation, resumption of marital union, divorce, end of marital union through death, marriage payment, annual socioeconomic interview, and quarterly price index information.

The interviewers use a genealogical list to identify individuals and locate their individual ID numbers (names are not permanent identifiers). In addition, one or two long-term employees keep daily diaries describing a range of activities in the village, with particular emphasis on the proceedings of village court cases.

Data management

In each village, a supervisor oversees the operation, and someone from the GTRP senior staff visits the villages twice a year to resupply the questionnaires, collect the completed ones, and pay the research assistants.

Until 1995, all data management was handled by the two senior anthropologists, Elizabeth Colson and Thayer Scudder. They recorded everything in ASCII text files, using an ingenious coding mechanism to relate individuals in the genealogy, and between 1992 and 1997 those files were converted to a relational database. At the same time, event-driven questionnaires, designed to work with the relational organization of the data, were introduced.

Data quality is assured through multiple reinterviews and continuous data collection, with checking and rechecking of recorded data at each subsequent interview. Data quality is measured by analyzing patterns in event counts, age reporting, and trends and comparing these patterns with those of neighbouring populations and standard models.

Data analysis relies on a collection of custom-designed relational-database tools, and the use of statistical techniques for the analysis of longitudinal data. The first work to come from the demographic analysis of the data is a basic analysis of the demographic history of the population (Clark et al. 1995).

Gwembe DSS basic outputs

Demographic indicators

Between 1957 and 1995, the Gwembe Tonga DSS recorded 82 000 person–years of exposure for males and 94 000 person–years of exposure for females. For males, 22% of those were lived by children aged 5 years and younger and 51% by children aged 15 years or younger. For females, the corresponding figures were 20% and 50%. This information is displayed graphically in Figure 14.2 as a population pyramid.

During 1991–95, mortality was fairly high. During that period, roughly 1 in 10 newborns died before reaching 1 year old, and only 8 of 10 newborns survived to their fifth birthday. For adults, 6 of 10 men or women who lived to age 20 survived to age 50. Over that same period, the two sexes combined generated a crude death rate of 25 per 1000 (24 per 1000 when standardized using the Segi [1960] age standard). This is

Figure 14.2. Population pyramid for person–years observed at the Gwembe Tonga DSS site, Zambia, 1957–95.

high in comparison with the developed world, but only moderately high compared with other developing regions. Over the duration of the study, the life expectancy at birth rose from 38 years for both men and women in 1957–61 to 52 years for men and 58 years for women in 1982–86, then fell back to 46 years for men and 50 years for women in 1992–95. The recent reduction in life expectancy at birth reflects the impact of HIV–AIDS and the deterioration of the economy and the health-care system in Zambia.

The total fertility rate peaked during the period of 1972–76 at a level of 7 children per woman and subsequently fell to a level of about 4 children per woman during the period of 1992–95. The married total fertility rate peaked at a little more than 10 children per woman in the 1972–76 period and fell to just over 6 children per woman in 1992–95. In both cases, the decline was mediated by a substantial reduction in age-specific fertility rates for women between 20 and 39 years old, with a slightly greater reduction for women between 30 and 39 years old.

Tables 14.1 and 14.2 contain age-specific mortality and fertility rates calculated over the period of 1957–95. They do not reveal the changes that have taken place over time, but they do provide good measures of the average age-specific rates during the entire period over which data have been collected.

Table 14.1. Age- and sex-specific mortality at the Gwembe Tonga DSS site, Zambia, 1957–95.

Note: _nD_x, observed deaths between ages x and x +n; _nM_x, observed mortality rate for ages x and x + n; _nPY_x, observed person–years between ages x and x +n.

Table 14.2. Female age-specific fertility rates at the

Note: Rates reflect annual hazards of giving birth. TFR, total fertility rate for women 15–49 years old.

Acknowledgments

The 15 000 people of Gwembe Tonga who have contributed their time and patience to the creation of the Gwembe Demographic Database deserve primary acknowledgment and thanks.

Elizabeth Colson and Thayer Scudder collected and did the original coding and input all the data presented here. They have been supported over the years by numerous organizations, to which we are thankful. Material presented in this chapter has been made possible through the support of the National Science Foundation of the United States, the William Penn Foundation, the Fulbright International Scholarship Program of the United States, and the National Institute on Aging of the National Institutes of Health of the United States. We are grateful to INDEPTH for making it possible to contribute these data to this monograph on mortality.

Chapter 15
MANHIÇA DSS, MOZAMBIQUE

P.L. Alonso, F. Saúte, J.J. Aponte, F.X. Gómez-Olivé, A. Nhacolo, R. Thomson, E. Macete, F. Abacassamo, P.J. Ventura, X. Bosch, C. Menéndez, and M. Dgedge¹

Site description

Physical geography of the Manhiça DSA

The Manhiça DSA is in the district of Manhiça (Maputo Province) in southern Mozambique at latitude 25°24′S and longitude 32°48′E (Figure 15.1). It lies at an average altitude of 50 m above sea level and covers an area of 100 km². The district has two distinct zones: the fertile lowlands, which comprise the floodplains of the Incomati River, are sparsely inhabited, and are subject to intensive sugar cane and fruit farming; and an escarpment of moderate height, which gives rise to a flat plateau on which virtually the entire DSA is situated. The area has two distinct seasons. The warm season is between November and April, when most of the rains fall (annual rainfall during 1998 was 1100 mm); a cool, dry season lasts for the rest of the year.

Figure 15.1. Location of the Manhiça DSS site, Mozambique (monitored population, 36 000).

Population characteristics of the Manhiça DSA

The town of Manhiça and the surrounding villages have a population of about 36 600 inhabitants, with a density of 360 inhabitants/km². The population is peri-urban and rural. People of the area are mainly Xironga and Xichangana, and their languages are often termed Ronga and Changana. The two dominant religions are Islam and Christianity. The people of the DSA are mostly subsistence farmers and workers in an agricultural cooperative that grows sugarcane, bananas, and rice. Workers also operate a large sugarcane-processing factory. An increasing number of small traders are establishing shops and businesses along the busy road that transects the district from north to south. There are 10 primary schools in the study area (6768 students and 85 teachers) and 1 secondary school (1492 students and 32 teachers). The rate of illiteracy is higher among females, at 47%, than among males, at 24%. Whereas 66% of men and 49% of women have primary education, only 9% of men and 4% of women have secondary education; and less than 1% of both men and women have gone beyond their secondary education.

Villages in this area typically comprise a loose conglomeration of compounds separated by garden plots and grazing land. Houses are simple, with walls typically made of cane, with thatched or corrugated roofs. In towns, houses are often grouped into family compounds and surrounded by grass fences. Towns grew substantially during the civil war in the 1980s as displaced people looked for refuge. After the end of the war, few inhabitants returned to their original homes, and displaced settlements have now been integrated into towns. Water comes mainly from community wells, although some households have their own wells. Some areas have community-run pumps. Both wells and pumps are supervised and chlorinated regularly by the District Water and Sanitation Department. The Maputo–Beira road and the Maputo–Xai railroad cross the area from north to south. With the exception of the (small) centre of the town of Manhiça, which has an erratic public-electricity service, the rest of the area relies on more traditional systems for lighting.

Centro de Investigação em Saúde de Manhiça (CISM, Manhiça health research centre) is in the centre of the study area. This 80-bed health facility includes a busy outpatient clinic, a maternity and child-care unit with an expanded immunization program and nutritional services, and a 24-hour emergency room. A smaller, 10-bed health centre is located 6 km south of the village of Manhiça. Malaria, acute respiratory infection, and malnutrition remain the most important causes of illness and death in children <5 years old.

Mozambique is recovering from a long period of war, including the independence wars against the Portuguese colonial power and the more recent civil armed conflict. The country still ranks as one of the poorest in the world, with an estimated per capita income of less than US $300. Although mild flooding of the alluvial plains of the rivers that cross southern Mozambique is not uncommon, the devastation caused by the large-scale floods of February 2000 had not been experienced for the last 30 years.

Manhiça DSS procedures

Introduction to the Manhiça DSS site

The overall objective of the Manhiça DSS site is to create a demographic platform to contribute to the research infrastructure of CISM. Its specific objectives are

The first census was carried out in the second half of 1996, registering a total of 33 500 inhabitants in the area. Currently, the total population under surveillance is around 36 600. The DSS was set up in the area immediately after the first enumeration and was based on the household-registration system (HRS), with some modifications. Update rounds are conducted every 4 months. During these rounds, every household is visited, and all vital events and changes of residency are recorded. Vital events include all births and deaths of the registered resident population in the study area. A resident is defined as any person who lives in the study area and expects to stay for at least the next 3 months. Should a resident leave the study area for 3 months or more, he or she is regarded as a migrant.

A number of field surveys have been carried out to define the epidemiology of malaria. These have included both cross-sectional surveys of children, adults, and pregnant women and cohort studies. The DSS site also has the natural catchment population of the Manhiça District hospital. Since late 1996, it has had a 24-hour hospital monitoring system that identifies all children attending the hospital from the study area and characterizes the morbidity patterns of this rural population.

The DSS operates under the direction of an epidemiologist and a junior demographer. A team of two supervisors and eight fieldworkers assists them. Researchers from CISM and others from the Ministry of Health and the School of Medicine at Eduardo Mondlane University are the main users of this facility.

Manhiça DSS data collection and processing

The selection of the site was made in early 1995. A suitable place to establish a peripheral research centre to investigate priority health issues of rural populations with access to a district hospital was sought. A balance between the rural settings and the logistic and supply needs of a sophisticated research centre had to be achieved. The town of Manhiça and its surrounding population, only 80 km from Maputo on good roads, was the optimum choice. Finally, the available data indicated that malaria was hyperendemic in the area, and therefore the site had potential for studies of malaria.

Field procedures

MAPPING — Airphotos of the area, available from the National Cartographic Institute, were digitized by the Catalan Cartographic Institute. The main geographic landmarks, including the Incomati River, the national road, and the railroad, were georeferenced. All households of the area were systematically numerated, and their position was determined using a GPS with differential correction. These data were then downloaded on the digitized photographs. The limits of the neighbourhoods were designed on the map, using the numbering of the households already positioned.

INITIAL CENSUS — The initial census was carried out from August to October 1996. After meetings with the community leaders, the census team was scheduled to visit the zones included in the DSS. For each zone, the chief of zone indicated which houses belonged to that zone. All households were numerated and then mapped using GPS, and every single person received a permanent ID number. For the households, information about the type of construction, number of constructions, and availability of a kitchen and toilet in the household was collected. Information collected on individuals included date of birth, parents, marital status, relationship with the chief of the household, and education level.

REGULAR UPDATE ROUNDS — Update rounds, where the fieldworkers visit all houses, are carried out every 4 months. In addition, the supervisors visit the chiefs of the zones every 2 weeks to collect information on vital events. Informal visits to other community key informants are done while doing the fieldwork.

Out- and in-migrations are registered, as well as destinations and origins. When the migration occurs within the study area the person receives a localization number related to the new house where he or she is going to live. A new in-migrant receives an ID number. As the ID number is permanent, a former resident of the area who reinmigrates receives a localization number.

CONTINUOUS SURVEILLANCE — The two field supervisors carry out daily visits at both CISM and Maragra’s Maternity, where all deliveries of the last 24 hours are registered. The baby and its mother are then visited at home, weekly until the baby is 1 month old. Two supervisors with motorbikes, supported by a large network of key community informants, identify all vital events in the study area and maintain a pregnancy register every week. The supervisors, with the objective of collecting demographic information, carry out regular fortnightly visits to the chief of the zones. Every 6 months, specially trained medical students of Eduardo Mondlane University in Maputo carry out verbal autopsies (VAs) of all deaths occurring among children <15 years old.

SUPERVISION AND QUALITY CONTROL — The DSS undertakes two types of supervision: field and computer supervision. The field supervision comprises random visits to households already visited by the fieldworker during the previous 24 hours to check the information collected by the fieldworker. Computer quality control consists of weekly comparisons between data recorded by the fieldworker in the census book and the information available on the computer. Moreover, once a year, a direct comparison is made of all information entered into the computer and information recorded by the fieldworkers in the census book.

Data management

Paper processing is used for all information collected. Information is registered in the fieldworkers’ household registration book and on precoded questionnaires, which are then processed at the computer centre. The researchers carry out weekly spot checks of the forms to discover any incompleteness of information and inconsistencies. The forms are then transferred to the data-management unit, where they are recorded and issued with a unique ID serial number. Once a questionnaire is entered into the computer, it is then sequenced and stored according to the type of demographic event.

Five working stations with an uninterruptible power supply are linked to a server with Windows NT 4.0 and Windows 95 environment. Specific software written in Visual FoxPro 5.0 is used for data entry and cleaning. This software is based on the HRS. It has a built-in series of logical checks and procedures to maintain the consistency and referential integrity of the databases. Standardized data-management procedures include systematic double entry of all forms by two data clerks. Inconsistencies are listed and corrected, based on the form information. Once the two entries are made consistent, the first entry is copied to a different folder accessible to researchers as “read only.” The main server contains a mirror disk to produce continuous backup. Moreover, a weekly backup is made onto a CD. All databases are transferred to STATA software for analyses.

Data quality control is assured through weekly checks using Visual FoxPro. These checks produce lists of inconsistencies, which are then corrected in the field by supervisors.

Demographic rates, such as fertility, mortality, and in- and out-migration, are calculated at the end of each round. The fieldworkers use VA to estimate specific mortality rates among individuals <15 years old. Merged with the hospital monitoring system, the DSS allows estimations of community-based morbidity rates. This information is accessible to the Ministry of Health.

Manhiça DSS basic outputs

Demographic indicators

The Manhiça population in mid-1999 reached 34 526: 4% are <1 year old; 13%, 0–4 years old; 26%, 5–14 years old; 51%, 15–64 years old; and 5%, ≥65 years old (Figure 15.2). Table 15.1 shows age- and sex-specific mortality at the DSS site. The infant-mortality rate is 78.5 per 1000 live births, and the under-five mortality rate is 130 per 1000. The age-dependency ratio is 0.87. The sex ratio is 83, and the total fertility rate is 5 (Table 15.2). Average household size is 4.0. Female-headed households account for 35%; and male, 65%. Primary school was attended by 76.4% of males and 53% of females ≥15 years old.

Figure 15.2. Population pyramid for person–years observed at the Manhiça DSS site, Mozambique, 1998–99.

Table 15.1. Age- and sex-specific mortality at the Manhiça DSS site, Mozambique, 1998–99.

Table 15.2. Age-specific fertility rates at the Manhiça DSS, Mozambique, 1998.

Note: TFR, total fertility rate.

Acknowledgments

The Spanish Agency for International Cooperation funds the running costs of CISM. During 1999, various studies were funded by 13 sources, including the World Health Organization, United Nations Children’s Fund, INDEPTH, Hospital Clinic of Barcelona (University of Barcelona, Spain), the Spanish Ministry of Health, and Eduardo Mondlane University, Mozambique. CISM is the first peripheral research centre of the Mozambican Ministry of Health. It is being developed under the premises of the cooperation program between Mozambique and Spain. A collaborative agreement, further developed between the Ministry of Health, the Maputo School of Medicine at Eduardo Mondlane University, and the Hospital Clinic of Barcelona, ensures the running of the centre.

Chapter 16
AGINCOURT DSS, SOUTH AFRICA

Mark Collinson,¹ Obed Mokoena,¹ Niko Mgiba,¹ Kathleen Kahn,¹ Stephen Tollman,¹ Michel Garenne,² Kobus Herbst,³ Elizabeth Malomane,⁴ and Sheona Shackleton¹

Site description

Physical geography of the Agincourt DSA

The Agincourt DSS is situated about 500 km northeast of Johannesburg in the Agincourt subdistrict of the Bushbuckridge region, Northern Province (Figure 16.1). Until 1994, the site was in a “homeland,” or bantustan area. The site extends between latitudes 24°50′ and 24°56′S and longitudes 31°08′ and 31°25′E. The altitude is 400–600 m above sea level. The field site, with 21 village communities, covers 390 km² and measures 38 km×16 km at its widest points.

The geoecological zone is semi-arid savanna, better suited to game farming and low-density cattle farming than to crop cultivation. Low average rainfall is common, with mean rainfall ranging from 700 to 550 mm in the western and eastern parts of the site. In addition, a high variability in interseasonal rainfall patterns renders the

Figure 16.1. Location of the Agincourt DSS site, South Africa (monitored population, 66 800).

area vulnerable to drought. Some 80% of the rain falls during the summer months of November–March. The area is affected by drought on average every 3.5 years, and even when rainfall is normal it is insufficient to fully supply the domestic-water or irrigation needs of the area. The area experiences hot summer and mild winter months. The temperature range is 12°–40°C in summer and 5°–27°C in winter.

Population characteristics of the Agincourt DSA

The total DSS population is 66 840, with 10 500 households and a population density of 172 persons/km². The male–female ratio for the total population is 0.929. The setting is rural in terms of distance from urban centres and lack of infrastructure. The main ethnic group is Shangaan, although Mozambicans, originally refugees, constitute more than a quarter (29%) of the total population. Both groups are Shangaan-speaking, and the Mozambicans are culturally affiliated with the South African host population. The area has mainstream Christian churches and independent African churches, and many people hold a mixture of indigenous and Christian beliefs.

Unemployment is estimated at 40–50%. Formal-sector employment involves migrant men who work in the mines, in the manufacturing and service industries of larger towns, and on nearby game and commercial farms and timber plantations. Women make up an increasing proportion of the migrant-labour population. Another source of local employment is the public sector. Informal-sector activities are widespread, and they include food- and fruit-vending. Pensions are an important source of income for many families. Female-headed households constitute 32% of all households (Tollman et al. 1995).

Almost all villages have at least one primary school, and 14 of the 21 villages have a secondary school. More than 40% of adults 25–59 years old have received no formal schooling. Six percent have completed secondary school, and only 3% have proceeded to some form of postsecondary education. Of those 15–24 years old, almost all have attended primary school, but only 46% have made the transition to secondary school. Although 85% of those 10–14 years old were enrolled in primary school, age of enrollment is frequently delayed. Adult female literacy (56%) is somewhat lower than adult male literacy (62%) (Tollman et al. 1995).

Various types of housing are found, ranging from traditional mud huts to brick dwellings with tin or tiled roofs. Stands are generally too small to support subsistence agriculture, and crops grown merely supplement the family diet. Water is pumped via purification plants, in some cases only to the main reservoirs in the villages. In other cases, water is harvested from wells dug mainly in riverbeds, and in rare cases from boreholes. Women or children collect water manually, usually in 25-L drums, and transport it either by wheelbarrow or on the head. Water shortage poses a serious problem in most villages. Levels of household sanitation are poor, and pit toilets of varying effectiveness are the norm. All roads are unpaved. Public transportation is limited to privately owned minibus taxis. Although seriously lacking, electricity and telephone services are benefiting from recent development initiatives.

The DSS site has a health centre and five satellite clinics, all staffed by nurses. A restricted number of drugs are dispensed from each of these primary-care facilities, and the health centre has a small laboratory, able to perform a limited number of diagnostic tests. An ambulance is based at the health centre. All services are free, and they include child health, family planning, antenatal care, delivery and postpartum care, minor ailments, and chronic-disease treatments. Although waiting times are long, most of these services are underused. A contributing factor in this is poor drug supply. Referrals are to two district hospitals, each about 25 km from the health centre. The main health problems revealed by verbal-autopsy (VA) analysis are diarrhea, kwashiorkor, and AIDS in children <5 years old; accidents, violence, and AIDS in the 15–49 age group; and chronic degenerative diseases, mainly cardiac, cerebrovascular, liver, and malignant diseases, among those ≥50 years old (Kahn et al. 1999; Tollman et al. 1999; Garenne, Tollman, Kahn, and Gear 2000). Seasonal malaria is evident. A high rate of adolescent fertility occurs in the midst of escalating HIV sero-prevalence (Garenne, Tollman, and Kahn 2000).

Forced relocation of communities under the apartheid regime (in the 1940–60 period) and the formation of ethnically divided homelands in the 1970s had a significant impact on the social, economic, and demographic profile of the population. The government of the time sought to massively exploit the labour in rural South Africa while simultaneously restricting development. Homeland governments were given hegemonic powers, with dubious results. Although poorly organized and managed, homeland health services rested on a network of mission hospitals and clinics and were thus less fragmented than their urban counterparts. This led to their having a more comfortable fit with South Africa’s new decentralized, district-based health system. As a result, densely settled rural villages with cash-based economies and males 20–59 years old largely absent from the permanent population became the pattern of settlement. Recent changes in government have affected movement patterns. With more freedom of movement people tend to move to rural towns. Along paved roads through these rural areas, the rural towns are becoming development nodes.

Agincourt DSS procedures

Introduction to the Agincourt DSS site

The current primary objective relates directly to objective 3, namely, to provide a research infrastructure and longitudinal database for a range of community-based studies on the burden of disease, health-systems interventions, and social–household–community dynamics, to inform decentralized health and social policy.

The Agincourt baseline census was conducted in 1992. The original population monitored was 57 509 individuals in 8896 households. By 1999 this had increased to 66 840 individuals in 10 500 households. VAs and maternity histories were introduced in 1993. A partnership between the Agincourt DSS, the study communities, and the local health services was established and is carefully nurtured (Tollman et al. 1995).

The Agincourt DSS data are updated every 12 months. Residents are defined as either “permanent” (resident in the study site for ≥6 months in the preceding year) or “migrant” (resident in the study area for <6 months but nevertheless regarding the Agincourt area as “home”). A VA is conducted in the vernacular on all deaths by a trained lay fieldworker and assessed by medical practitioners (Kahn et al. 2000). The software system contains a relational database, constructed in Microsoft Access 2000. The main demographic, health, and socioeconomic variables measured routinely by the DSS include births, deaths, in- and out-migrations, household relationships, resident status, refugee status, education, and antenatal and delivery health-seeking practices. In the DSS update of 1999, information on chronic cough was collected for a study on active case-finding for tuberculosis. In 2000, information on labour-force participation in the formal and informal sectors was collected; for 2001, a module describing the burden of disabilities was planned.

The Agincourt DSS is the foundation for the Agincourt Health and Population Programme (AHPP), a research initiative of the University of Witwatersrand. It is housed within the Health Systems Development Unit of the Faculty of Health Sciences. AHPP has strong ties to the Northern Province Department of Health and the Bushbuckridge regional and district health services. The core management team comprises the AHPP leader, senior researcher, field-research manager, and site manager. The DSS field team comprises 4 supervisors, 20 fieldworkers, 1 VA supervisor, and 4 VA fieldworkers, all employed on contract for the duration of data collection. In 2000, a part-time data-form checker was employed. The data-capture team comprises a supervisor and two data typists.

Work is under way within AHPP to address a portfolio of clinical, public-health, population, and social challenges, including

Community feedback and dialogue are integral to the Agincourt research process. Information from the DSS and related research initiatives is communicated to the study communities through printed “village fact sheets” and through ad hoc community meetings. This facilitates community involvement in local health action and related development activities. Information is regularly discussed with district and regional health-service managers and senior officials of the Northern Province and national Department of Health.

Agincourt DSS data collection and processing

Several factors influenced the choice of the Agincourt field site, in particular

Field procedures

MAPPING — Hand-drawn maps of each village were made for the initial census in 1992. These included roads, dwellings, and other reference landmarks, such as railway lines, power lines, shops, churches, and soccer fields. Since then village maps have been updated each year through both specific fieldwork exercises and routine correction and updating during census fieldwork. The maps make it possible for any member of the team to return to a particular household without risk of confusion.

INITIAL CENSUS AND REGULAR UPDATE ROUNDS — Six census rounds have been completed to date (1992 baseline, 1993–94, 1995, 1997, 1999, 2000). Rounds are conducted in the dry season, that is, July–November. A fieldworker interviews the best respondent available at the time of the visit. Individual information is checked for every household member. All events that have occurred since the previous census are recorded, and any status observations are updated. Where possible, questions are directed to particular household members; for example, maternity-history or pregnancy-outcome questions are directed to the woman involved. If appropriate respondents are unavailable, the fieldworkers undertake revisits, usually during evenings and on weekends, with a limit placed at two revisits per household. From the second round onwards, a VA has been conducted on every death to determine the most probable cause. VAs are conducted concurrently with the census, but with a separate team of fieldworkers who are dedicated to VA interviews only.

In 1999, with the aim of increasing the speed of data collection, the field team was expanded to 4 supervisors, 20 fieldworkers, 1 VA supervisor, and 4 VA fieldworkers. The team operates out of five field offices, provided by clinics or community members at no charge.

SUPERVISION AND QUALITY CONTROL — To ensure data quality, supervised visits and random duplicate visits are conducted. For supervised visits, the supervisor goes into the field with the fieldworker and observes a number of interviews. After each interview, constructive feedback is given to the fieldworker, with the aim of improving interview technique. Random duplicate visits are conducted by the supervisor on 2% of the population. After a careful explanation is given, the entire interview is conducted again; differences between the first and second interviews are identified; and possible reasons for these are determined. From these data, quality can be assessed, and error rates can be computed.

Furthermore, form-checking occurs in a structured system at four levels of the field organization. The checks become more detailed as the form progresses through the system. An error is returned to the fieldworker for correction, and, where necessary, a revisit is done. Supervisors keep track of forms, using printed checklists.

Data management

Existing details of each household are printed on the census forms. The fieldworker checks this information, and, in addition, status fields are updated for each household member. Separate event forms are used for pregnancy outcomes, deaths, migrations, and maternity histories and are only completed if one of these events occurred in the intercensal period. Death forms are completed in duplicate, so that one copy can be passed on to the VA team. The forms from each interview are stapled together in a predetermined order. Supervision-checklist forms allow for the monitoring of data collection from each dwelling.

When a form has left the field and passed all quality checks it is captured on a software system. Currently, data are captured using simultaneous data entry on three computers connected to a network, writing to a database on a server. The software system is a “relational” database, currently held in Microsoft Access 2000. A custom-made data-entry program has been developed that sits on top of the Access database and, by mirroring the format of the data forms, provides an easy-to-use interface between the user and the database. The database consists of related tables that store different aspects of the data. The main table is the “Individual” table, which stores key information on all individuals encountered. The “Residence” table provides information on individual residence episodes, indicating how and when a person entered and exited a particular location in the field site. A “Memberships” table records information on how and when an individual entered and exited a particular household (that is, a social grouping, defined as people “eating from the same pot”). The system contains tables for each event category — “Births,” “Deaths,” “Migrations,” and “Maternity histories” — and an “Observations” table records information about each interview. In addition, a range of status-observation tables record information about individuals updated at various frequencies during census rounds. These include “Residence status” (updated in rounds 1–6), “Education status” (rounds 1 and 4), “Cough status” (round 5), and “Labour status” (round 6).

The system incorporates built-in validation checks. Implausible data (such as a date of death occurring before a date of birth) are prevented from entering the database. When these errors occur, the form is put to one side, reviewed by the data manager, and, when necessary, returned to the team supervisor for resolution. Unusual, though possible, data (such as a delivery by a woman >50 years old) are also flagged by the system and reviewed by the data manager. As data are entered, computer checks are done to look for invalid codes, missing values, inconsistencies within and between records, incorrect spellings of place names, and duplicate entries. A useful data-quality check after a census is a comparison of the villages of origin of “internal” in-migrants with destinations of internal out-migrants, as well as a review of demographic trends.

Basic analyses are done to produce village fact sheets, community-feedback information, sampling frames, and denominator information. Further data-cleaning and demographic analyses are conducted to produce reliable population data. A monograph of the “baseline” findings was produced in 1994, and all findings of specific scientific or policy interest are published in local and international peer-reviewed journals. Presentations are made to policymakers at subdistrict, district, regional, provincial, and national levels.

Agincourt DSS basic outputs

Demographic indicators

The total (de jure) population was 66 840 in 1999. Of these, the permanent population (resident in the site for more than 6 months of the preceding year) was 56 566. The sex ratio (male–female) in the total population was 0.929, falling to 0.712 in the permanent population 15–49 years old. The age structure in the total population at the end of 1999 was as follows: 2.3% were <1 year old; 12.0%, 0–4 years old; 27.6%, 5–14 years old; 55.9%, 15–64 years old; and 4.5%, ≥65 years old (Figure 16.2). The total fertility rate (TFR) was 2.72 in 1999 (Table 16.1). The proportion of female-headed households was 32%. The age-dependency ratio was 0.79. The infant mortality rate was 43.0 per 1000 live births among males and 45.1 per 1000 live births among girls. Mean household size was 6.4, and the adult literacy rate⁵ in females was 56% and in males 62%.

Migration monitoring is conducted by recording data on individuals moving into or out of a household between census rounds (Table 16.2). Moves are classified as internal if they have their origin and destination within the DSS site villages; otherwise they are classified as external moves. The place of origin or destination, date of the move, and reason for its occurrence are captured for each move.

⁵ Computed as the percentage of persons ≥15 years old with at least 4 years of formal schooling.

Figure 16.2. Population pyramid for person–years observed at the Agincourt DSS site, South Africa, 1995–99.

Table 16.1. Age-specific fertility rates at the Agincourt DSS site, South Africa, by 2-year periods, 1992–99.

Note: ASFR, age-specific fertility rate; PYs, person–years; TFR, total fertility rate.

Demographic trends

A profound decline has occurred in TFR, from around 6 births per woman in 1970–74 to 2.72 births per woman in 1999 (Garenne, Tollman, and Kahn 2000). Mortality had been declining for some time, until about 1993. Since 1994 an increase in mortality in three age groups has been documented: young adults 20–49 years old (both sexes), children 0–4 years old (both sexes), and older adult women 50–64 years old (male mortality continues to decline in this age group) (Table 16.3). Migration trends show a net population loss, as a result of an excess of external out-migration over external in-migration (about 1% of the population per year). The main focus for departures has been nearby towns, in particular Mkhuhlu. This move toward Mkhuhlu was particularly strong in the 1994–95 period (Collinson et al. 2000).

Table 16.2. Out- and in-migration rates by age at the Agincourt DSS site, South Africa, 1992-99.

Table 16.3. Age- and sex-specific mortality at the Agincourt DSS site, South Africa, 1995–99.

Acknowledgments

The Agincourt team is indebted to the communities of Bushbuckridge and to the Northern Province Department of Health for their partnership, support, and ongoing contributions to these efforts. We also acknowledge, with pleasure, the encouragement and support of the Wellcome Trust, United Kindom, Andrew W. Mellon Foundation, United States, University of Witwatersrand, South Africa, Anglo American Chairman’s Fund, South Africa, the European Union, and the Henry J. Kaiser Family Foundation, United States.

Chapter 17
DIKGALE DSS, SOUTH AFRICA

Site description

Physical geography of the Dikgale DSA

The Dikgale DSS site is located in the central region of Mankweng District, Northern Province, South Africa, about 40 km from Pietersburg, the capital of the Northern Province, and 15 km from the University of the North (Figure 17.1). The site covers an area of 71 km² and is 6 km

Figure 17.1. Location of the Dikgale DSS site, South Africa (monitored population, 7900).

Population characteristics of the Dikgale DSA

The total population comprises 7956 people, with a population density of 116 inhabitants/km². The site is peri-urban, and the main ethnic group is Pedi. Most of the inhabitants belong to the Moria Zionist Church, which has a combination of Christian and indigenous beliefs, while others belong to the Lutheran or Anglican churches. The language spoken by all inhabitants of the site is Northern Sotho.

A large proportion of adults are migrant workers, whereas others work as farm labourers on neighbouring farms or as domestic workers in nearby towns. Many are pensioners. The unemployment rate in the area is high.

There are four primary schools and three secondary schools in the DSA. In all schools, the classrooms are overcrowded, and few educational amenities are available. Most children attend primary school, and the adult literacy rate is 79.8 and 73.6% in males and females, respectively.

Dwelling units consist of a mixture of shacks, traditional mud huts, and conventional brick houses. A few households have water taps in their yards, but most must fetch water from taps situated at strategic points in the villages. Most households have a pit latrine in their yards, but they have no organized waste disposal. Infrastructure in the villages is poor, and none of the roads is tarred. A bus service is available mornings and evenings during weekdays.

Free health care is given at a primary health-care clinic in the field site to children <6 years old, pregnant women, and the elderly. The service provided by the clinic includes family planning, antenatal care, growth monitoring and immunization in children, and management of patients with chronic diseases. Mankweng Hospital, situated 15–20 km from the field site, serves as a referral hospital.

Both infectious and noninfectious diseases are prevalent in the area. According to records kept at the clinic, the main health problems in children are respiratory and gastrointestinal diseases. Undernutrition is common, and the growth of a large proportion of children is stunted. From a survey undertaken in the DSS site, the health problems in adults include type-2 diabetes, hypertension, iron overload, and obesity.

Dikgale DSS procedures

Introduction to the Dikgale DSS site

The broad aim of the Dikgale DSS is to provide information to improve the health of the people of Northern Province and to assist the local government in developing an effective health-care policy. As no accurate data were available on the prevalence of diseases in rural and peri-urban areas of the Northern Province, the initial objective of the DSS was to establish a field site to assess the incidence and prevalence of diseases.

Community leaders in Dikgale were approached regarding the possibility of conducting research on health status, and their cooperation was obtained. The site was subsequently established, and the first census was undertaken from August to November 1995. At that time, the population was 8001, but it decreased to 7956 by 1998. Every year an update is undertaken. A distinction is made between the total population in the study area, which comprises all who regard their home as being in the area, and the permanent population, which consists of those resident in the area for ≥6 months in the year preceding the census update.

Demographic variables measured routinely include births, migrations, and deaths. Maternity histories are also conducted. Several special surveys have also been undertaken to determine the prevalence of specific diseases and disorders, such as iron and vitamin-A status in preschool children, vision defects, and the prevalence of noninfectious diseases in the adult population.

An office for the use of the fieldworkers is located in the field site. The coordinator of the Dikgale DSS is a staff member of the University of the North. Data collected at the DSS site are regularly forwarded to the Department of Health, Northern Province.

Dikgale DSS data collection and processing

The Dikgale DSS site was chosen because of its proximity to the University of the North and because of the presence at the site of a primary health-care clinic.

Field procedures

MAPPING — The fieldworkers constructed a sketch map of each village, with all roads and landmarks, such as schools and shops, indicated. They gave each household a number.

INITIAL CENSUS — The initial census was undertaken during August–November 1995. Fieldworkers visited each household and recorded the name, age, and education of each household member.

REGULAR UPDATE ROUNDS — Updates occur annually. During the update, fieldworkers visit each household with a printout of the census form. Any changes that have occurred since the last visit are recorded on the form.

A full maternity history is taken from each woman in the household who has had a child. Particulars pertaining to date of birth, gender, and live or stillbirths are entered on the maternity-history form. All births that occurred since the last visit are recorded, along with birth weight, site of delivery, and use of contraceptives. Death forms are completed for each death. Migration inquiries include information on origin or destination and reasons for the change of locality.

SUPERVISION AND QUALITY CONTROL — The field site is visited regularly, and quality checks are carried out by coordinators on a random sample of 2.5% of households. Any problems encountered are discussed and solved.

Data management

The forms that have been completed by the fieldworkers are checked manually before being processed. All data are entered into a custom-designed Access database program. The program contains checks to limit entry errors. A series of validation routines are run; corrections are made with reference to the raw data; and return visits to the field are made, where indicated.

Data analysis is done in Microsoft Excel. Reports are produced on a regular basis and forwarded to the community and local authorities.

Figure 17.2. Population pyramid for person–years observed at the Dikgale DSS site, South Africa, 1995–99.

Dikgale DSS basic outputs

Demographic indicators

The population in 1998 was 7956, and the proportion of the population <1 year old was 1.6%; 1–4 years old, 11.25%; 5–14 years old, 25.68%; 15–64 years old, 57.33%; and ≥65 years, 5.74% (Figure 17.2). The age-dependency ratio is 0.74; the sex ratio, 0.96; and the infant mortality rate, 38.9 per 1000 live births. The average household size is 6.33, and the household headship is 58% male and 42% female. The percentage of people aged ≥15 years who are literate is 79.8% for males and 73.6% for females.

Migration surveys are undertaken every year and indicate a complex pattern of migration. Most movement takes place either within the same village or to another village at the DSS site or to a neighbouring village. Very few move to urban areas. The largest proportion of subjects either leaving or in-migrating occurs in the age group of 0–24 years. Few subjects >40 years old move away from their homes.

Table 17.1 shows the age- and sex-specific all-case mortality at the Dikgale DSS site.

Table 17.1. Age- and sex-specific mortality at the Dikgale DSS site, South Africa, 1995–99.

Acknowledgments

The project was funded by the Norwegian Universities Committee for Development Research and Education.

The Dikgale DSS team includes the following staff: M. Mogashoa, E. Makhura, S. Makgato, and S. Mokokoane. Colleagues include S. Tollman, Department of Community Health, University of Witwatersrand, South Africa, M. Garenne, Centre français sur la population et le développement, Paris, France, K. Herbst, Department of Community Health, Medical University of South Africa, South Africa, and J. Sundby, Department of Medical Anthropology, Oslo University, Norway.

Chapter 18
HLABISA DSS, SOUTH AFRICA

Geoff Solarsh, Justus Benzler, Vicky Hosegood, Frank Tanser, and Annemie Vanneste¹

Site description

Physical geography of the Hlabisa DSA

Hlabisa health district is located in northern KwaZulu-Natal, South Africa, and covers an area of 1430 km² (Figure 18.1). Altitude ranges from 20 to 500 m above sea level. The terrain is flat to undulating, to mountainous, and vegetation ranges from sparse grassland to thick forest. The DSS site is located between latitudes 28.18° and 28.47°S and longitudes 31.97° and 32.38°E.

Figure 18.1. Location of the Hlabisa DSS site, South Africa (monitored population, 85 000).

¹ Africa Centre Demographic Information System, Africa Centre for Population Studies and Reproductive Health (University of Natal, University of Durban-Westville, and Medical Research Council of South Africa), South Africa.

Population characteristics of the Hlabisa DSA

The Hlabisa District is one of many official rural magisterial districts in KwaZulu-Natal. Before the political transition in 1994, this district formed part of the KwaZulu homeland and was coterminous with the Hlabisa health ward, an integrated and semiautonomous unit of the homeland health system. It was built around a community, or nonspecialist, hospital, from which comprehensive primary health-care services were administered and supervised through a radiating network of fixed and mobile clinics. After the political transition, the KwaZulu homeland was dismantled, and all services were reorganized into a three-tier national health system, with national-, provincial-, and district-level responsibilities. Although the districts are in a state of transition, the Hlabisa health subdistrict is largely coterminous with the previous Hlabisa health ward and still functions as a semiautonomous and integrated health system at the district level. This site, therefore, constitutes a functional unit of the national health system and provides a representative district health-system model for the implementation of national health policies and programs.

The resident population of 210 000 is Zulu speaking and predominantly rural, although the district has pockets of urban and peri-urban populations in the southeast, near the market town of Mtubatuba. It is made up of four tribal areas, each under the local authority of a tribal chief, or inkosi. Although residents of the tribal area owe allegiance to, and fall under the local jurisdiction of, the chief, they are not necessarily members of his traditional clan or tribe. No data are yet available on religion, but the majority of people are considered Christians.

Unlike in many other parts of Africa, where homesteads are clustered in clearly identifiable villages, rural populations in KwaZulu-Natal live in scattered multigenerational homesteads of varying size (1–100 people). The area is characterized by large variations in population density (0–6500 people per km²). There is substantial circulatory migration between the district and commercial and industrial centres, at varying distances from the Hlabisa District, and, to a lesser extent, between the district and remote rural areas in the hinterland. This is largely driven by the need for employment and educational opportunities. Wide differentials also appear in living standards, literacy rates, and access to electricity and clean water, although overall social and environmental conditions are substantially better than in many other countries in sub-Saharan Africa. Annual per capita income is US $1730, the literacy rate is 69%, and life expectancy at the beginning of the AIDS epidemic averaged 63 years. South African populations show substantial demographic and epidemiological variability, reflecting regional and ethnic differences and, in the final analysis, vast differentials in social and economic conditions.

The health infrastructure in the Hlabisa District is typical of many other rural health districts in KwaZulu-Natal and, to a lesser extent, elsewhere in South Africa. The central fixed health facility in the district is a community hospital, run by generalist medical practitioners and nurses. It provides a wide range of curative and emergency services, including surgical and obstetrical care and the usual range of primary health-care services offered at fixed clinics. Scattered throughout the district are 12 fixed nurse-run clinics, providing routine prenatal, natal, and postnatal care; family planning, preventive child-health services (including immunizations); treatment for TB, sexually transmitted diseases, and noncommunicable diseases, such as diabetes and hypertension; and treatment of a wide range of minor complaints. All conditions

considered to exceed the capacity or skill of the resident nurses are referred to the hospital. These clinics are supervised from the hospital and are visited biweekly by a medical doctor from the hospital.

In those parts of the district not covered by fixed clinics, mobile health services are provided on a 2–4-week basis at defined points. The level and range of services offered are similar to those offered at the fixed clinics, but the mobile clinics are unable to offer deliveries or any other forms of treatment or care requiring short-term stay or admission. Community health workers cover about half of the homesteads in the district and are largely responsible for nutritional and general health promotion in these households, supervised home care, and, where necessary, referral to clinic or hospital.

Clinics are generally well used, with about 95% of pregnant women attending an antenatal clinic at least once during their pregnancies and up to 80% of children achieving full primary vaccination, that is, up to and including a measles vaccination. Substantial use is also made of the medical services offered by private practitioners in Mtubatuba and at private clinics farther afield in the towns of Empangeni and Richard’s Bay.

Hlabisa DSS procedures

Introduction to the Hlabisa DSS site

In 1991, MRC established a research station at Hlabisa hospital as a rural research unit of the national Centre for Epidemiological Research in South Africa. The presence of this productive MRC unit was probably the most important factor in the selection of the Hlabisa District for the Africa Centre for Population Studies and Reproductive Health.

The Africa Centre for Population Studies and Reproductive Health was established in April 1997 and moved into the Hlabisa District in November 1997. Since then residential and office infrastructures have been established, the entire health district has been mapped, and the DSS has been set up in the southeastern section of the district. Other projects have also been set up during this period, including studies to determine the effect of exclusive breastfeeding on mother-to-child transmission (MTCT) of HIV and another to determine the effect of male labour migration on HIV infection of nonmigrant partners.

The demographic surveillance area is based in the tribal area of Mpukunyoni, the most populous and least mountainous part of the Hlabisa District, and includes the township of Kwamsane. It is about 435 km² in area and is sharply demarcated by the hard boundaries of large perennial rivers, nature reserves, forestry areas, and commercial farmland on all but its northern boundary. As a result, the area is a fairly discrete and well-circumscribed geographical unit, allowing clear definition of the survey population. Although we do not yet have the data to show it, we fully expect the surveillance population to be representative of rural populations in KwaZulu-Natal and, to a lesser extent, of rural black populations elsewhere in the country.

The total population of the Hlabisa District was estimated at 210 000 in the last national census (1996). The DSS population is now expected to include 70 000 residents and an additional 15 000 nonresidents who are considered part of the 11 000 registered households in the DSA. This number falls within the recommended range for DSSs and is a sufficient sample size to generate most key demographic rates besides maternal mortality. We cater to nonresident members in the DSS, because of the large numbers of migrants who regularly return home, contribute significantly to the financial resources of households, and, because they often return home to give birth or to die, will contribute to both mortality and fertility estimates.

Formal data collection began in 2000, and, at the time of writing, the first annual data set is not yet complete.

The DSS in the Hlabisa site is a foundation project of the Africa Centre for Population Studies and Reproductive Health, an international research centre based in the Hlabisa health district. The Africa Centre was established in 1997 as a consortium between two universities — the University of Natal and the University of Durban-Westville — and the Medical Research Council (MRC) of South Africa. It has been funded through a generous 5-year grant from the Wellcome Trust in the United Kingdom.

The DSS, specifically, will generate demographic, social, and economic trends in a population going through the health transition and concurrently experiencing the complex effects of a rampant HIV epidemic. It will also provide a platform and framework for a wide range of smaller research projects concerned with household and family dynamics, microeconomic policy and programs, and burden-of-disease assessments, including HIV and noncommunicable diseases. These data will be of extreme importance to provincial and national health planners in South Africa. They will also be very important for anyone interested in modelling the impact of the HIV epidemic in rural African populations throughout the subcontinent. The DSS will be intimately integrated into a mother-and-child cohort of 2000 mother–child pairs and therefore uniquely positioned to evaluate interventions to reduce MTCT of HIV and other interventions in early childhood. In relation to the three participating institutions the DSS also provides a very important source of data for the construction of community-based curricula for population studies, reproductive health, and child health relevant to undergraduate and postgraduate trainees. A strong commitment to community-based and service-oriented education at these institutions creates many

opportunities for these linkages, and some initial steps have already been taken toward the development of these programs.

Hlabisa DSS data collection and processing

Field procedures

MAPPING — The entire DSS data set is georeferenced and is linked to programs with cartographic functions and the capacity for complex spatial analysis. This GIS, a key component of the larger demographic-information system, comprises a series of geographical layers of the district — including magisterial and nature-reserve boundaries, roads, and rivers — and covers about 500 facilities and 26 000 homesteads, 10 000 of which are under continuous demographic surveillance. Extensive use was made of differential GPS units (accuracy of <2 m) in setting up the DSS. Fourteen fieldworkers and 3 supervisors were trained in the use of differential GPS. The fieldworkers were divided into four teams (three mapping teams of four members each and one backup team of two members). Each mapping team was assigned a supervisor and given a portion of the district to map and a set of maps covering the district. The maps, based on recent aerial photographs, contained the estimated positions of all homesteads and facilities in the district. The supervisors were responsible for coordinating the movements of the fieldworkers.

Data dictionaries, which restricted data entry to a unique block of 5000 numbers, were uploaded to each fieldworker’s hand-held GPS unit. The data dictionaries allowed the fieldworkers to collect all attribute data, using the GPS units — thereby obviating the use of all forms. Tags bearing the unique ID numbers of the homesteads were affixed and information about the homestead was collected only if a senior resident of the homestead was present and in a position to provide permission for the homestead to be mapped. The backup team was responsible for visiting absentee homesteads and collecting the associated attribute data and affixing tags. Once the fieldworkers returned from the field, the data were downloaded to computers. Differential correction occurred the next day.

The area is socially and physically heterogeneous, and the population is dispersed throughout the DSS site. This presented numerous problems for the equitable distribution of work among fieldworkers. A fuzzy-logic model within a GIS was therefore used to ergonomically define 48 fieldworker areas and thereby equitably distribute the workload. The GIS procedure for doing this was described in Tanser (2000).

Future GIS research will focus on the geographic analysis of the demographic and health data collected by the DSS down to homestead level (once these data are available). In particular, it is hoped that HIV status will be geographically linked to homestead data so that researchers can explore patterns of HIV in the DSS site and possible models to explain the distribution.

INITIAL CENSUS — The Africa Centre for Population Studies and Reproductive Health, only recently established, commenced its first round of visits to all homesteads in the DSS site in February 2000. The population within the circumscribed DSS site is the study sample, and data collection covers an observation period starting 1 January 2000. In the first round, a complete census of all homesteads (places of residence), households (social groups), and individuals has been conducted. Baseline data collection has included descriptive characteristics of homesteads and households, demographic

attributes of all individuals, and detailed pregnancy and reproductive histories from all women of reproductive age.

REGULAR UPDATE ROUNDS — Continuous registration and updates of all new and existing homesteads, households, and individuals occurs during each round of data collection. All homesteads in the DSA are visited every 4 months to register new individuals and households, update demographic variables on registered individuals and households, and record all births, deaths, and migrations. Reproductive-health questionnaires are completed on all women of reproductive age (15–49 years) who were not registered in previous rounds or were not previously eligible for this baseline questionnaire. In subsequent update rounds, all demographic events and changes in the status of all homesteads, households, and individuals in the surveillance area are recorded. Additional modules dealing with subjects such as household socioeconomic status, health-related conditions, HIV sero-prevalence, or disability may be added on to routine updates as the need arises. Monitoring will continue for a minimum of 5 years, but it will almost certainly be extended if the research program is productive.

Baseline surveys on all women of reproductive age will be conducted throughout the duration of the project. Household socioeconomic questionnaires are expected to be administered annually at all homesteads in the DSS site. Other surveys, such as disability and HIV sero-prevalence surveys, are likely to be conducted in the future as the need arises. At present, no biological samples or direct measurements are taken as part of DSS activities.

CONTINUOUS SURVEILLANCE — Births and deaths are recorded and related data are regularly updated as above. Deaths are reported as part of the regular notification of vital events. Every death notification triggers a separate visit by a verbal-autopsy (VA) nurse, who administers a standard three-part VA questionnaire to determine the likely cause of death.

SUPERVISION AND QUALITY ASSURANCE — Supervisors check all questionnaires when they come out of the field, and serial subsamples are further checked by managers at various levels. Supervisors make weekly quality-control visits to at least 5% of homesteads visited by fieldworkers in the preceding period, to validate interview findings. During the first 2–4 weeks of each round, supervisors accompany each of their fieldworkers on weekly supervised visits to take any opportunity to support and correct interviewing-technique difficulties or misconceptions as early as possible. Additional, unannounced spot checks are made in the case of weak or unreliable fieldworkers.

Data management

Data are also collected at homestead and household levels from a hierarchy of key informants, with highest priority given to those with the greatest knowledge of other household members. Individual data are collected preferentially from the individual concerned, but in his or her absence they are collected from the best-informed key informant. In the case of reproductive-health questionnaires, all information is collected directly from the woman herself. All questionnaires are entered into a large relational Access database, using a customized front end (programed in Delphi 5) specifically developed for the Africa Centre for Population Studies and Reproductive Health.

Figure 18.2. Population pyramid of person–years (both resident and nonresident) observed at the Hlabisa DSS, South Africa, 2000.

Data are single punched in two shifts of six data capturers each and subjected, in the data-entry program, to a series of checks of validity and consistency. All questionnaires with evident errors or omissions are returned to the field for correction, and those that are free of errors are archived after data capture. All errors that cannot be corrected by supervisors are returned to fieldworkers for revisits and corrections.

The core data set will be routinely analyzed and form the subject of regular annual reports, which will be widely disseminated in hard copy and on the Internet. The core data set will be exclusively available to the Africa Centre’s investigators and their collaborators for a limited period (still to be defined) but will thereafter be made available in anonymized form to a larger audience for secondary analysis. Substudies built on the DSS core data will be encouraged, be the property of the investigators, and be made available to the scientific community through peer-reviewed literature. A final policy on data-sharing and dissemination has not yet been drafted, but the above principles are likely to apply.

Hlabisa DSS basic outputs

Demographic indicators

Actual DSS monitoring began in 2000, and the first annual data set was not yet complete at the time of writing. Hence, no indicators can be provided for the moment. A population period for 2000 is shown in Figure 18.2.

Acknowledgments

The Africa Centre for Population Studies and Reproductive Health was established through a large core grant from the Wellcome Trust in the United Kingdom for a minimum period of 5 years. Since then additional funding has been obtained from the National Institutes of Health in the United States. Additional funders — such as the Centre for International Migration, a German governmental aid organization supporting human capacity development in developing countries — have made important contributions to funding for the work of expatriate scientists.

The DSS team would like to specially acknowledge the support we have received throughout the DSS-development process from Dr Robert Howells, the program leader, and from Dr Wendy Ewart, the program officer responsible for the Population Studies Programme at the Wellcome Trust. Wendy’s unflagging optimism and belief in the abilities of this team went a long way to sustaining us through the many demanding challenges of setting up a DSS while concurrently establishing a new research field site. We believe the support we received went far beyond anything we could have expected from the representatives of a large international funder.

An important formative influence in the development of the Hlabisa DSS has come from the DSS model provided by the Navrongo Health Research Centre. We are aware of the important contributions to this model provided by Fred Binka and Jim Phillips and wish to thank them for hosting us at Navrongo and for sharing their experience so freely. Important elements of the system design were also influenced by the Nouna DSS in Burkina Faso. We are grateful to Michel Garenne for much support for the theoretical background.

We are also grateful to Steve Tollman and the team at Agincourt in South Africa who have supported us through many different phases of this project and have generously shared important ideas and resources from the Agincourt team.

A team of social, medical, and population scientists from the participating institutions assisted with the pilot studies that laid down the definitions and concepts on which the DSS has been based. In particular, we wish to thank Eleanor Preston-Whyte, Tessa Marcus, Mark Lurie, and Abby Harrison for their assistance.

Special mention must be made of the continuous and critical nurturing role of Dudu Biyela, the community liaison manager of the Africa Centre. Dudu helped us to take our first halting steps into the Hlabisa community and watched over us through every twist and turn in our relationship with it. The position of “mother” of the Hlabisa DSS is not reflected in our staffing structure or as a budgetary line item, but a few of us know to whom this role belongs and how deservedly it has been earned.

Finally and most importantly, we wish to recognize the important services and essential contributions of all members of the 80-strong DSS team responsible for the diverse tasks and functions that make a system of this kind run efficiently and enable high-quality data to be generated.

Chapter 19
NOUNA DSS, BURKINA FASO

Site description

Physical geography of the Nouna DSA

The DSS site of the Centre de recherche en santé de Nouna (CRSN, Nouna Health Research Center) is in the Nouna health district in northwest Burkina Faso, 300 km from the capital, Ouagadougou (Figure 19.1). The Nouna area is a dry orchard savannah, populated almost exclusively with subsistence farmers of various ethnic groups. The area has a sub-Saharan climate, with a mean annual rainfall of 796 mm (range 483–1083 mm) over the past five decades.

PART III. INDEPTH DSS SITE PROFILES
Introduction	129
Chapter 8. Butajira DSS, Ethiopia	135
Chapter 9. Dar es Salaam DSS, Tanzania	143
Chapter 10. Hai DSS, Tanzania	151
Chapter 11. Ifakara DSS, Tanzania	159
Chapter 12. Morogoro DSS, Tanzania	165
Chapter 13. Rufiji DSS, Tanzania	173
Chapter 14. Gwembe DSS, Zambia	183
Chapter 15. Manhiça DSS, Mozambique	189
Chapter 16. Agincourt DSS, South Africa	197
Chapter 17. Dikgale DSS, South Africa	207
Chapter 18. Hlabisa DSS, South Africa	213
Chapter 19. Nouna DSS, Burkina Faso	221
Chapter 20. Oubritenga DSS, Burkina Faso	227
Chapter 21. Farafenni DSS, The Gambia	235
Chapter 22. Navrongo DSS, Ghana	247
Chapter 23. Bandim DSS, Guinea-Bissau	257
Chapter 24. Bandafassi DSS, Senegal	263
Chapter 25. Mlomp DSS, Senegal	271
Chapter 26. Niakhar DSS, Senegal	279
Chapter 27. Matlab DSS, Bangladesh	287
Chapter 28. ORP DSS, Bangladesh	297
Chapter 29. FilaBavi DSS, Viet Nam	305

Appendix 1. Working Examples of DSS Forms
Example 1. DSS Baseline Form (RufijiDSS)	312
Example 2. Household Registration Book (HRB) (Rufiji DSS)	313
Example 3. Pregnancy Outcome / Birth Form (Rufiji DSS)	315
Example 4. Death Registration Form (Navrongo DSS)	316
Example 5. Marital Status Form (Butajira DSS)	316
Example 6. VA Form: Deaths of Children from Day 31 to 5 Years (Morogoro DSS)	318
Example 7. In-migration Form (Navrongo DSS)	320
Example 8. Out-migration Form (Navrongo DSS)	321
Appendix 2. Acronyms and Abbreviations	323
Appendix 3. Glossary	327
Appendix 4. Bibliography	333

Additional rounds of data collection	29
Geographic information systems	29
Conclusion	30
Chapter 4. Processing DSS Data
Introduction	31
Background	32
The INDEPTH concept of a data core	33
The reference data model	35
The role of the reference data model in maintaining data integrity	39
Extending the core	40
Conclusion	41
Chapter 5. Assessing the Quality of DSS Data
Introduction	43
Assessing data quality in the field	43
Assessing data quality at the data centre	44
Conclusion	47
PART II. MORTALITY AT INDEPTH SITES
Chapter 6. Comparing Mortality Patterns at INDEPTH Sites
Abstract	51
Introduction	51
Age-specific mortality rates and life tables	52
Crude death rate	53
Child mortality	57
Adult mortality	59
Discussion	61
Annex: Lifetables	63
Chapter 7. INDEPTH Mortality Patterns for Africa
Abstract	83
Mortality models and Africa	83
Principal-components analysis	87
Principal components of INDEPTH mortality data	89
INDEPTH mortality patterns	96
Demonstration of the HIV–AIDS model life-table system	111
Conclusion	114
Annex: AIDS-decremented model life tables	115

POPULATION AND HEALTH IN DEVELOPING COUNTRIES

CONTENTS

FOREWORD

PREFACE

ACKNOWLEDGMENTS

INTRODUCTION

PART IDSS CONCEPTS AND METHODS

Chapter 1CORE CONCEPTS OF DSS

Introduction

Demographic surveillance systems

Demographic surveillance area

Longitudinality

Visits

Unique identifiers

Primary DSS subjects

Residential units

Households

Individuals

Eligibility

Residency and membership

Core DSS events

Births and fertility

Deaths and mortality

Migrations and mobility

Episodes

Other events

Nuptiality and conjugal relationships

Construction and disintegration of residential units

Events occurring in households

Chapter 2DSS-GENERATED MORTALITY RATES AND MEASURES

Introduction

Rates and ratios

Crude death rate

Age-specific death rate and ratio

Infant mortality

Under-five mortality

Maternal mortality rate and ratio

Standardization

Age-standardized death rate

Confidence intervals for rates

Chapter 3DSS METHODS OF DATA COLLECTION

Introduction

Establishing the monitored population

Planning for data collection

Initial census

Update rounds

Recording demographic events

Monitoring mortality

Tracking migrants

Additional rounds of data collection

Geographic information systems

Conclusion

Chapter 4PROCESSING DSS DATA

Introduction

Background

The INDEPTH concept of a data core

Tailoring the core system

The reference data model

The role of the reference data model in maintaining data integrity

Extending the core

Conclusion

Chapter 5ASSESSING THE QUALITY OF DSS DATA

Introduction

Assessing data quality in the field

Supervised visits

Form checking

Duplicate visits

Assessing data quality at the data centre

General procedures

Statistical techniques

Matching of records

Population pyramid

Alternative techniques

Age preference

Sex ratios

Age ratios

Comparison with population models

Conclusion

PART IIMORTALITY AT INDEPTH SITES

Chapter 6COMPARING MORTALITY PATTERNS AT INDEPTH SITES

PART I
DSS CONCEPTS AND METHODS

Chapter 1
CORE CONCEPTS OF DSS

Chapter 2
DSS-GENERATED MORTALITY RATES AND MEASURES

Chapter 3
DSS METHODS OF DATA COLLECTION

Chapter 4
PROCESSING DSS DATA

Chapter 5
ASSESSING THE QUALITY OF DSS DATA

PART II
MORTALITY AT INDEPTH SITES

Chapter 6
COMPARING MORTALITY PATTERNS AT INDEPTH SITES

Chapter 7
INDEPTH MORTALITY PATTERNS FOR AFRICA

PART III
INDEPTH DSS SITE PROFILES

Chapter 8
BUTAJIRA DSS, ETHIOPIA

Chapter 9
DAR ES SALAAM DSS, TANZANIA

Chapter 10
HAI DSS, TANZANIA