statistical annex explanatory notes The tables in this statistical annex present information on population health in WHO Member States and regions for the year 2003 (Annex Tables 1, 2a and 2b) under-five and neonatal causes of deaths for 2000-2003(Annex Tables 3 and 4), selected national health accounts aggregates for 1998-2002(Annex Tables 5 and 6), and selected indicators related to reproductive, maternal and newborn health(Annex Tables 7 and 8). These notes provide an overview of concepts, methods and data sources, together with references to more detailed documentation. It is hoped that careful scrutiny and use of the results will lead to progressively better measurement of core indicators of population health and health system financing The theme of The World Health Report 2005 is maternal and child individuals is a prerequisite to economic development. In health. The latest estimates of under-five mortality and causes of death order to monitor progress in achieving the MDGs as are now available, so special consideration is given both to estimates as major childhood health initiatives, a reliable inform and to the empirical basis of under-five mortality and causes of death. tion base is critical Annex Table 3 on the estimated number and distribution of deaths by It is essential for the United Nations to disseminate cause focuses on the deaths of children under the age of five years. identical estimates on the MDGs, including under-five For the first time, the estimated numbers of deaths for neonates by mortality, in order to enhance proper use of these figures cause are being published (as Annex Table 4). Consequently, the table in policy planning or in programme monitoring and eval- on estimated deaths by cause, sex and mortality stratum that appeared uation. There is thus an urgent need to develop a system in earlier World Health Reports is not being published here through which the United Nations speaks with a single Of the eight major goals set at the United Nations Millennium Sum- voice and produces estimates that agree. Four special- mit in 2000, six relate directly to the health and well-being of women ized agencies- WHO, the United Nations Children s Fund and children. These Millennium Development Goals(MDGs)reflect a (UNICEF), the United Nations Population Division, and thorough recognition by govemments that improving the well-being of the World Bank -organized a meeting on child mortality
149 The theme of The World Health Report 2005 is maternal and child health. The latest estimates of under-five mortality and causes of death are now available, so special consideration is given both to estimates and to the empirical basis of under-five mortality and causes of death. Annex Table 3 on the estimated number and distribution of deaths by cause focuses on the deaths of children under the age of five years. For the first time, the estimated numbers of deaths for neonates by cause are being published (as Annex Table 4). Consequently, the table on estimated deaths by cause, sex and mortality stratum that appeared in earlier World Health Reports is not being published here. Of the eight major goals set at the United Nations Millennium Summit in 2000, six relate directly to the health and well-being of women and children. These Millennium Development Goals (MDGs) reflect a thorough recognition by governments that improving the well-being of individuals is a prerequisite to economic development. In order to monitor progress in achieving the MDGs as well as major childhood health initiatives, a reliable information base is critical. It is essential for the United Nations to disseminate identical estimates on the MDGs, including under-five mortality, in order to enhance proper use of these figures in policy planning or in programme monitoring and evaluation. There is thus an urgent need to develop a system through which the United Nations speaks with a single voice and produces estimates that agree. Four specialized agencies – WHO, the United Nations Children’s Fund (UNICEF), the United Nations Population Division, and the World Bank – organized a meeting on child mortality statistical annex explanatory notes The tables in this statistical annex present information on population health in WHO Member States and regions for the year 2003 (Annex Tables 1, 2a and 2b), under-five and neonatal causes of deaths for 2000–2003 (Annex Tables 3 and 4), selected national health accounts aggregates for 1998–2002 (Annex Tables 5 and 6), and selected indicators related to reproductive, maternal and newborn health (Annex Tables 7 and 8). These notes provide an overview of concepts, methods and data sources, together with references to more detailed documentation. It is hoped that careful scrutiny and use of the results will lead to progressively better measurement of core indicators of population health and health system financing
150 The World Health Report 2005 (infant and under-five mortality rates) in May 2004 Meeting participants agreed on the following actions to further explore their joint activities to improve the estimation process on a regular basis: creation of a common database; discussion on the issues of the currently used methods and ways for improvement; and more focus on country capacity building and training to improve data availability and quality Accordingly, WHO and UNICEF produced a consistent set of under-five mortality rates by country for the period 1990-2003, which was used as the basis for estimation shown in Annex Tables 1 and 2a. It should be emphasized that such estimates may not be directly derived from reported data. Annex Table 2b summarizes the empirical basis for the estimation of under-five mortality by age group WHO is the primary organization to provide estimates on cause-specific mortality A major problem has been the lack of accurate cause-specific mortality data from developing countries, especially those with higher levels of mortality. In collaboration with its regional offices, WHO headquarters collects cause-of-death data from its 192 Member States. An established agreement between headquarters and the regional offices ensures that there is no duplication of work at the country level to report data to WHO. The WHO Regional Offices for the Americas, Europe and the Eastem Mediter ranean deploy simultaneous efforts to ensure that data are received in a regular and timely manner. Data from the African Region are virtually non-existent and account for the major difficulties in assessing the level of cause-specific mortality in that area The data submitted by Member States then become part of WHO,s unique historical database on causes of death(WHO Mortality Database) which contains data as far back as 1950(1). During 2000-2003 some 100 Member States provided vital regis- tration data to WHO and captured approximately 18 million deaths. It should be noted, however, that more than two thirds of deaths in the world are not being reported These data gaps need to be filled both by stepping up efforts to work with countries and initiatives to obtain more recent mortality data and by collaborating with partners to promote better tools and investment in data collection and analysis. There is also a need for better harmonization of cause-specific mortality estimates within WHO, with other organizations in the United Nations system and with academic institutions. In 2001, WHO established the Child Health Epidemiology Reference Group(CHERG to help improve estimates of cause-specific mortality in childhood. This group of in dependent technical experts has developed and applied rigorous standards for the development of estimates related to the major causes of childhood deaths, and worked closely with WHO and UNICEF to incorporate their results into broader WHO child health estimates at global, regional and when possible country level. Further detail on CHERG methods and products is available elsewhere(). The results of WHO collabo- ration with the CHERG and UNICeF are presented in Annex Tables 3 and 4. These estimates have been reviewed, agreed upon and supported by the WHO De partments of Child and Adolescent Health and Development(CAH)and Measurement and Health Information Systems( MHI), the UNICEF Division of Policy and Planning (DPP)and an independent group of external experts. Initial WHO estimates and tech- nical explanations were sent to Member States for comment. Comments or data pro- vided in response were discussed with them and incorporated where possible. The estimates published here should, however, still be interpreted as the best estimates of WHO rather than the official viewpoint of Member States
150 The World Health Report 2005 (infant and under-five mortality rates) in May 2004. Meeting participants agreed on the following actions to further explore their joint activities to improve the estimation process on a regular basis: creation of a common database; discussion on the issues of the currently used methods and ways for improvement; and more focus on country capacity building and training to improve data availability and quality. Accordingly, WHO and UNICEF produced a consistent set of under-five mortality rates by country for the period 1990–2003, which was used as the basis for estimation shown in Annex Tables 1 and 2a. It should be emphasized that such estimates may not be directly derived from reported data. Annex Table 2b summarizes the empirical basis for the estimation of under-five mortality by age group. WHO is the primary organization to provide estimates on cause-specific mortality. A major problem has been the lack of accurate cause-specific mortality data from developing countries, especially those with higher levels of mortality. In collaboration with its regional offices, WHO headquarters collects cause-of-death data from its 192 Member States. An established agreement between headquarters and the regional offices ensures that there is no duplication of work at the country level to report data to WHO. The WHO Regional Offices for the Americas, Europe and the Eastern Mediterranean deploy simultaneous efforts to ensure that data are received in a regular and timely manner. Data from the African Region are virtually non-existent and account for the major difficulties in assessing the level of cause-specific mortality in that area. The data submitted by Member States then become part of WHO’s unique historical database on causes of death (WHO Mortality Database) which contains data as far back as 1950 (1). During 2000–2003 some 100 Member States provided vital registration data to WHO and captured approximately 18 million deaths. It should be noted, however, that more than two thirds of deaths in the world are not being reported. These data gaps need to be filled both by stepping up efforts to work with countries and initiatives to obtain more recent mortality data and by collaborating with partners to promote better tools and investment in data collection and analysis. There is also a need for better harmonization of cause-specific mortality estimates within WHO, with other organizations in the United Nations system and with academic institutions. In 2001, WHO established the Child Health Epidemiology Reference Group (CHERG) to help improve estimates of cause-specific mortality in childhood. This group of independent technical experts has developed and applied rigorous standards for the development of estimates related to the major causes of childhood deaths, and worked closely with WHO and UNICEF to incorporate their results into broader WHO child health estimates at global, regional and when possible country level. Further detail on CHERG methods and products is available elsewhere (2). The results of WHO collaboration with the CHERG and UNICEF are presented in Annex Tables 3 and 4. These estimates have been reviewed, agreed upon and supported by the WHO Departments of Child and Adolescent Health and Development (CAH) and Measurement and Health Information Systems (MHI), the UNICEF Division of Policy and Planning (DPP) and an independent group of external experts. Initial WHO estimates and technical explanations were sent to Member States for comment. Comments or data provided in response were discussed with them and incorporated where possible. The estimates published here should, however, still be interpreted as the best estimates of WHO rather than the official viewpoint of Member States
statistical annex explanatory notes 151 ANNEX TABLE 1 All estimates of population size and structure for 2003 are based on the demographic assessments prepared by the United Nations Population Division (3). These estimates efer to the de facto population, and not the de jure population in each Member State The annual growth rate, the dependency ratio, the percentage of population aged 60 years and more, and the total fertility rate are obtained from the same United Nations Population Division database To assess overall levels of health achievement, it is crucial to develop the best pos- sible assessment of the life table for each country. Life tables have been developed for all 192 Member States for 2003 starting with a systematic review of all available evidence from surveys, censuses, sample registration systems, population laborato ries and vital registration on levels and trends in under-five and adult mortality rates This review benefited greatly from a collaborative assessment of under-five mortality levels for 2003 by WHO and UNICEF. WHO uses a standard method to estimate and project life tables for all Member States using comparable data. This may lead to minor differences compared with official life tables prepared by Member States Life expectancy at birth, the probability of dying before five years of age(under-five mortality rate)and the probability of dying between 15 and 60 years of age(adult mortality rate) derive from life tables that WHO has estimated for each Member State. Procedures used to estimate the 2003 life table differed for Member States depend ing on the data availability to assess child and adult mortality. Because of increasing heterogeneity of patterns of adult and child mortality, WHO has developed a model life table system of two-parameter logit life tables, and with additional age-specific parameters to correct for systematic biases in the application of a two-parameter system, based on about 1800 life tables from vital registration judged to be of good quality (4). This system of model life tables has been used extensively in the develop ment of life tables for those Member States without adequate vital registration and in projecting life tables to 2003 when the most recent data available are from earlier years. Estimates for 2003 have been revised to take into account new data received since publication of The World Health Report 2004 for many Member States and may not be entirely comparable with those published in the previous reports. The methods ised to construct life tables are summarized below and a full detailed overview has en published (4, 5) For Member States with vital registration and sample vital registration systems, de- mographic techniques( Preston-Coale method, Brass Growth-Balance method, Gen eralized Growth-Balance method and Bennett-Horiuchi method) were first applied to assess the level of completeness of recorded mortality data in the population above years of age and then those mortality rates were adjusted accordingly (6 .Where vital registration data for 2003 were available, these were used directly to construct the life table. For other countries where the system provided a time series of annual life tables, the parameters(5, 6o)were projected using a weighted regression model giving more weight to recent years(using an exponential weighting scheme such that the weight for each year twas 25% less than that for year t+1). For countries with a total population of less than 750 000 or where the root mean square error from the regression was greater than or equal to 0.011, a shorter-term trend was estimated by applying a weighting factor with 50%annual exponential decay. Projected values of the two life table parameters were then applied to a modified logit life table model, using the most recent national data as the standard, which allows the capture of the most recent age pattern, to predict the full life table for 2003
statistical annex explanatory notes 151 ANNEX TABLE 1 All estimates of population size and structure for 2003 are based on the demographic assessments prepared by the United Nations Population Division (3). These estimates refer to the de facto population, and not the de jure population in each Member State. The annual growth rate, the dependency ratio, the percentage of population aged 60 years and more, and the total fertility rate are obtained from the same United Nations Population Division database. To assess overall levels of health achievement, it is crucial to develop the best possible assessment of the life table for each country. Life tables have been developed for all 192 Member States for 2003 starting with a systematic review of all available evidence from surveys, censuses, sample registration systems, population laboratories and vital registration on levels and trends in under-five and adult mortality rates. This review benefited greatly from a collaborative assessment of under-five mortality levels for 2003 by WHO and UNICEF. WHO uses a standard method to estimate and project life tables for all Member States using comparable data. This may lead to minor differences compared with official life tables prepared by Member States. Life expectancy at birth, the probability of dying before five years of age (under-five mortality rate) and the probability of dying between 15 and 60 years of age (adult mortality rate) derive from life tables that WHO has estimated for each Member State. Procedures used to estimate the 2003 life table differed for Member States depending on the data availability to assess child and adult mortality. Because of increasing heterogeneity of patterns of adult and child mortality, WHO has developed a model life table system of two-parameter logit life tables, and with additional age-specific parameters to correct for systematic biases in the application of a two-parameter system, based on about 1800 life tables from vital registration judged to be of good quality (4). This system of model life tables has been used extensively in the development of life tables for those Member States without adequate vital registration and in projecting life tables to 2003 when the most recent data available are from earlier years. Estimates for 2003 have been revised to take into account new data received since publication of The World Health Report 2004 for many Member States and may not be entirely comparable with those published in the previous reports. The methods used to construct life tables are summarized below and a full detailed overview has been published (4, 5). For Member States with vital registration and sample vital registration systems, demographic techniques (Preston–Coale method, Brass Growth–Balance method, Generalized Growth–Balance method and Bennett–Horiuchi method) were first applied to assess the level of completeness of recorded mortality data in the population above five years of age and then those mortality rates were adjusted accordingly (6). Where vital registration data for 2003 were available, these were used directly to construct the life table. For other countries where the system provided a time series of annual life tables, the parameters (I 5, I 60 ) were projected using a weighted regression model giving more weight to recent years (using an exponential weighting scheme such that the weight for each year t was 25% less than that for year t+1). For countries with a total population of less than 750 000 or where the root mean square error from the regression was greater than or equal to 0.011, a shorter-term trend was estimated by applying a weighting factor with 50% annual exponential decay. Projected values of the two life table parameters were then applied to a modified logit life table model, using the most recent national data as the standard, which allows the capture of the most recent age pattern, to predict the full life table for 2003
152 The World Health Report 2005 For all Member States, other data available for child mortality such as surveys and censuses, were assessed and adjusted to estimate the probable trend over the past few decades in order to predict the child mortality in 2003. A standard approach to predicting child mortality was employed to obtain the estimates for 2003(see An nex Table 2a for more details)(7). Those estimates are, on the one hand, used to replace the under-five mortality rate in life tables of the countries that have a vital registration or sample vital registration system, but with incomplete registration of numbers of deaths under the age of five years. On the other hand, for countries with- out exploitable vital registration systems, which are mainly those with high mortality the predicted under-five mortality rates are used as one of the inputs to the modi fied logit system. Adult mortality rates were derived from either surveys or censuses where available; otherwise the most likely corresponding level of adult mortality was estimated based on regression models of child versus adult mortality as observed in the set of approximately 1800 life tables. These estimated child and adult mortality rates were then applied to a global standard, defined as the average of all the life tables, using the modified logit model to derive the estimates for 2003 It should be noted that the logit model life table system using the global standard does not capture high HIV/AIDS epidemic patterns, because the observed underlying life tables do not come from countries with the epidemic. Similarly, war deaths are not captured because vital registration systems often break down in periods of war 8). For these reasons, for affected countries, mortality without deaths attributable to HIV/AIDS and war was estimated and separate estimates of deaths caused by HIV/AIDS and war in 2003 were added The main results in Annex Table 1 are reported with uncertainty intervals in order to communicate to the user the plausible range of estimates for each country on each measure For the countries with vital registration data projected using time se- ries regression models on the parameters of the logit life table system, uncertainty around the regression coefficients has been accounted for by taking 1000 draws of the parameters using the regression estimates and variance covariance matrix of the estimators. For each of the draws a new life table was calculated In cases where ad ditional sources of information provided plausible ranges around under-five and adult mortality rates the 1000 draws were constrained such that each life table produced estimates within these specified ranges. The range of 1000 life tables produced by these multiple draws reflects some of the uncertainty around the projected trends in mortality, notably the imprecise quantification of systematic changes in the logit parameters over the time period captured in available vital registration data For Member States where complete death registrations were available for the year 2003 and projections were not used, the life table uncertainty reflects the event count uncertainty, approximated by the Poisson distribution, in the estimated age-specific death rates arising from the observation of a finite number of deaths in a fixed time For countries that did not have time series data on mortality by age and sex, the following steps were undertaken. First, point estimates and ranges around under-five and adult mortality rates for males and females were developed on a country-by country basis (5). In the modified logit life table system described (4), values on these two parameters may be used to identify a range of different life tables in relation to a global standard life table. Using the Monte Carlo simulation methods, 1000 random fe tables were generated by drawing samples from normal distributions around these inputs with variances defined according to ranges of uncertainty. In countries where
152 The World Health Report 2005 For all Member States, other data available for child mortality, such as surveys and censuses, were assessed and adjusted to estimate the probable trend over the past few decades in order to predict the child mortality in 2003. A standard approach to predicting child mortality was employed to obtain the estimates for 2003 (see Annex Table 2a for more details) (7). Those estimates are, on the one hand, used to replace the under-five mortality rate in life tables of the countries that have a vital registration or sample vital registration system, but with incomplete registration of numbers of deaths under the age of five years. On the other hand, for countries without exploitable vital registration systems, which are mainly those with high mortality, the predicted under-five mortality rates are used as one of the inputs to the modi- fied logit system. Adult mortality rates were derived from either surveys or censuses where available; otherwise the most likely corresponding level of adult mortality was estimated based on regression models of child versus adult mortality as observed in the set of approximately 1800 life tables. These estimated child and adult mortality rates were then applied to a global standard, defined as the average of all the life tables, using the modified logit model to derive the estimates for 2003. It should be noted that the logit model life table system using the global standard does not capture high HIV/AIDS epidemic patterns, because the observed underlying life tables do not come from countries with the epidemic. Similarly, war deaths are not captured because vital registration systems often break down in periods of war (8). For these reasons, for affected countries, mortality without deaths attributable to HIV/AIDS and war was estimated and separate estimates of deaths caused by HIV/AIDS and war in 2003 were added. The main results in Annex Table 1 are reported with uncertainty intervals in order to communicate to the user the plausible range of estimates for each country on each measure. For the countries with vital registration data projected using time series regression models on the parameters of the logit life table system, uncertainty around the regression coefficients has been accounted for by taking 1000 draws of the parameters using the regression estimates and variance covariance matrix of the estimators. For each of the draws, a new life table was calculated. In cases where additional sources of information provided plausible ranges around under-five and adult mortality rates the 1000 draws were constrained such that each life table produced estimates within these specified ranges. The range of 1000 life tables produced by these multiple draws reflects some of the uncertainty around the projected trends in mortality, notably the imprecise quantification of systematic changes in the logit parameters over the time period captured in available vital registration data. For Member States where complete death registrations were available for the year 2003 and projections were not used, the life table uncertainty reflects the event count uncertainty, approximated by the Poisson distribution, in the estimated age-specific death rates arising from the observation of a finite number of deaths in a fixed time interval of one year. For countries that did not have time series data on mortality by age and sex, the following steps were undertaken. First, point estimates and ranges around under-five and adult mortality rates for males and females were developed on a country-bycountry basis (5). In the modified logit life table system described (4), values on these two parameters may be used to identify a range of different life tables in relation to a global standard life table. Using the Monte Carlo simulation methods, 1000 random life tables were generated by drawing samples from normal distributions around these inputs with variances defined according to ranges of uncertainty. In countries where
statistical annex explanatory notes 153 uncertainty around under-five and adult mortality rates was considerable because of a aucity of survey or surveillance information, wide distributions were sampled but the results were constrained based on estimates of the maximum and minimum plausible values for the point estimate For 55 countries, mainly in sub-Saharan Africa, estimates of life tables were made by constructing counterfactual life tables excluding the mortality impact of the Hiv/ AIDS epidemic and then combining these life tables with exogenous estimates of the excess mortality rates attributable to HIvIAIDS. The estimates were based on back-calculation models developed as part of collaborative efforts between WHO and the Joint United Nations Programme on HIV/AIDS (UNAIDS)to derive country-level epidemiological estimates for HIV/AIDS. In countries with substantial numbers of deaths, estimates of their uncertainty range were also incorporated into the life table uncertainty analysis. ANNEX TABLE 2A Estimates of child mortality are regularly published by various international organiza- tions, including WHO. Footnotes are used to explain the underlying methodology and sometimes include information on the availability of empirical data that underlie the estimates More frequently, however, the reader of the tables is not informed about the source of information. In the current set of tables WHo has made a first attempt to share a brief summary of the underlying empirical information. This should allow the reader to obtain an idea of how much the estimate is based on real data versus assumptions. At this point the tables do not include an assessment of the quality of the data. The estimation process does take the quality of the empirical data into account In the context of the Millennium Development Goals(MDGs), particular attention is paid to the measurement of progress towards reaching Goal 4, to reduce by two thirds the mortality rate among children under five between 1990 and 2015.At coun try level this implies government commitment not only to implement initiatives to improve child health but also to set up a reliable system to monitor such progress Such a system, if implemented, should be able to provide the number of deaths of children under five years of age by sex, age and cause. However, countries with high levels of child mortality are those where there is very little information or none at all especially on trends Annex Table 2a presents the sources and results of information on under-five mortal- ity rates during the last 25 years which are available at WHO. All efforts were made to ensure completeness and accuracy of the information presented, but the table does not intend to be exhaustive. Data collection efforts are summarized for three periods 1980-1989, 1990-1999 and 2000-2003. Only data collected in the most recent pe riod provide new information on the trend in child mortality in the new millennium. In all other cases, the estimates for the MDGs are drawn entirely from projections based on trends derived from empirical data points prior to the year 2000 There are four primary sources of empirical under-five mortality data: vital reg istration(VR), sample registration system(SRS), surveys and censuses. The vital registration or sample registration system provides numbers of deaths by age and sex obtained by direct observation and reporting of individual deaths. These are pro- spectively collected data. In the case of a survey or a census, the empirical data are based on retrospective data. Interviews with mostly the mother or caregiver or head of household provide information on the survival history of children in the household
statistical annex explanatory notes 153 uncertainty around under-five and adult mortality rates was considerable because of a paucity of survey or surveillance information, wide distributions were sampled but the results were constrained based on estimates of the maximum and minimum plausible values for the point estimates. For 55 countries, mainly in sub-Saharan Africa, estimates of life tables were made by constructing counterfactual life tables excluding the mortality impact of the HIV/ AIDS epidemic and then combining these life tables with exogenous estimates of the excess mortality rates attributable to HIV/AIDS. The estimates were based on back-calculation models developed as part of collaborative efforts between WHO and the Joint United Nations Programme on HIV/AIDS (UNAIDS) to derive country-level epidemiological estimates for HIV/AIDS. In countries with substantial numbers of war deaths, estimates of their uncertainty range were also incorporated into the life table uncertainty analysis. ANNEX TABLE 2A Estimates of child mortality are regularly published by various international organizations, including WHO. Footnotes are used to explain the underlying methodology and sometimes include information on the availability of empirical data that underlie the estimates. More frequently, however, the reader of the tables is not informed about the source of information. In the current set of tables WHO has made a first attempt to share a brief summary of the underlying empirical information. This should allow the reader to obtain an idea of how much the estimate is based on real data versus assumptions. At this point the tables do not include an assessment of the quality of the data. The estimation process does take the quality of the empirical data into account. In the context of the Millennium Development Goals (MDGs), particular attention is paid to the measurement of progress towards reaching Goal 4, “to reduce by two thirds the mortality rate among children under five between 1990 and 2015”. At country level this implies government commitment not only to implement initiatives to improve child health but also to set up a reliable system to monitor such progress. Such a system, if implemented, should be able to provide the number of deaths of children under five years of age by sex, age and cause. However, countries with high levels of child mortality are those where there is very little information or none at all, especially on trends. Annex Table 2a presents the sources and results of information on under-five mortality rates during the last 25 years which are available at WHO. All efforts were made to ensure completeness and accuracy of the information presented, but the table does not intend to be exhaustive. Data collection efforts are summarized for three periods: 1980–1989, 1990–1999 and 2000–2003. Only data collected in the most recent period provide new information on the trend in child mortality in the new millennium. In all other cases, the estimates for the MDGs are drawn entirely from projections based on trends derived from empirical data points prior to the year 2000. There are four primary sources of empirical under-five mortality data: vital registration (VR), sample registration system (SRS), surveys and censuses. The vital registration or sample registration system provides numbers of deaths by age and sex obtained by direct observation and reporting of individual deaths. These are prospectively collected data. In the case of a survey or a census, the empirical data are based on retrospective data. Interviews with mostly the mother or caregiver or head of household provide information on the survival history of children in the household
154 The World Health Report 2005 This may be through gathering mortality information for a specific period prior to the census or survey interview, through a birth history or through questions on children ever born and children still alive("indirect" Brass questions)( 9) The sources of information as listed in the annex table 2a were used to derive the estimated trends and projections of rates for under-five-year-olds for the year 2003 shown in both Annex Tables 1 and 2a. A standard approach to predicting the most recent child mortality was employed to ensure comparability between countries and may lead to minor differences compared with official statistics prepared by Member States(7) For each country, estimates of under-five mortality rate are derived from weighted least squares regression of under-five mortality rate on their reference dates Explanatory variables include date, as well as those that capture rates of change of under-five mortality across periods of time. The weights assigned to each data point eflect its quality or consistency with all other data points. In other cases, additional sources were used as inputs in the standard regression model Vital registration can be considered as the gold standard for the collection of mortality data, as it allows the registration of deaths by age and sex. Vital registration systems with high levels of completeness are commonplace in developed countries. Although several developing countries are improving their vital registration systems, in many other countries-especially countries with high levels of mortality- such a system is non-existent. Another source of mortality data is the sample vital registration system which assesses vital events at the national level from information collected in sample areas. These two sources, in principle, provide data on a regular yearly basis The column"VR/SRS" in Annex Table 2a-vital registration/sample registration system shows the number of years of data from either system available at WHO. In the absence of a prospective data collection system in a country, household surveys will provide direct or indirect estimates of the level of under-five mortality, primarily using birth history questionnaires in which mothers are asked to provide information about their children, those still living as well as those who did not survive. Similarly, census questionnaires may include a module on mortality, which may refer to recent deaths in the household or use "indirect" Brass questions to estimate child mortality. It should be noted that one single survey or census can generate more than one estimate of under-five mortality for different periods of time. However, the Survey/Census column of Annex Table 2a shows the number of the surveys or censuses available at WHO. Furthermore, when a survey was carried over from one year to the next, only the starting year was taken into account It is worth noting the efforts of WHO regional offices in collecting vital registration data from Member States. International agencies such as the United Nations and UNICEF also maintain historical databases on under-five mortality rates, which have been generously shared and incorporated in our analyses. Other sources of informa tion include data from national censuses or surveys, or from specialist surveys such as the Demographic and Health Survey( DHS)undertaken by ORC Macro and the Multiple Indicator Cluster Survey(MICS)conducted by UNICEF. Finally, national statis tical documents such as statistical yearbooks, reports from specialized agencies and periodical paper findings were also incorporated into the database ANNEX TABLE 2B Whereas Annex Table 2a presents the estimates on under-five mortality rates, An nex Table 2b presents an empirical basis of detailed age-specific mortality rates directly obtained from the most readily available sources on the subject, namely
154 The World Health Report 2005 This may be through gathering mortality information for a specific period prior to the census or survey interview, through a birth history or through questions on children ever born and children still alive (“indirect” Brass questions) (9). The sources of information as listed in the Annex Table 2a were used to derive the estimated trends and projections of rates for under-five-year-olds for the year 2003 shown in both Annex Tables 1 and 2a. A standard approach to predicting the most recent child mortality was employed to ensure comparability between countries and may lead to minor differences compared with official statistics prepared by Member States (7). For each country, estimates of under-five mortality rate are derived from weighted least squares regression of under-five mortality rate on their reference dates. Explanatory variables include date, as well as those that capture rates of change of under-five mortality across periods of time. The weights assigned to each data point reflect its quality or consistency with all other data points. In other cases, additional sources were used as inputs in the standard regression model. Vital registration can be considered as the gold standard for the collection of mortality data, as it allows the registration of deaths by age and sex. Vital registration systems with high levels of completeness are commonplace in developed countries. Although several developing countries are improving their vital registration systems, in many other countries – especially countries with high levels of mortality – such a system is non-existent. Another source of mortality data is the sample vital registration system which assesses vital events at the national level from information collected in sample areas. These two sources, in principle, provide data on a regular yearly basis. The column “VR/SRS” in Annex Table 2a – vital registration/sample registration system shows the number of years of data from either system available at WHO. In the absence of a prospective data collection system in a country, household surveys will provide direct or indirect estimates of the level of under-five mortality, primarily using birth history questionnaires in which mothers are asked to provide information about their children, those still living as well as those who did not survive. Similarly, census questionnaires may include a module on mortality, which may refer to recent deaths in the household or use “indirect” Brass questions to estimate child mortality. It should be noted that one single survey or census can generate more than one estimate of under-five mortality for different periods of time. However, the “Survey/Census” column of Annex Table 2a shows the number of the surveys or censuses available at WHO. Furthermore, when a survey was carried over from one year to the next, only the starting year was taken into account. It is worth noting the efforts of WHO regional offices in collecting vital registration data from Member States. International agencies such as the United Nations and UNICEF also maintain historical databases on under-five mortality rates, which have been generously shared and incorporated in our analyses. Other sources of information include data from national censuses or surveys, or from specialist surveys such as the Demographic and Health Survey (DHS) undertaken by ORC Macro and the Multiple Indicator Cluster Survey (MICS) conducted by UNICEF. Finally, national statistical documents such as statistical yearbooks, reports from specialized agencies and periodical paper findings were also incorporated into the database. ANNEX TABLE 2B Whereas Annex Table 2a presents the estimates on under-five mortality rates, Annex Table 2b presents an empirical basis of detailed age-specific mortality rates directly obtained from the most readily available sources on the subject, namely
statistical annex explanatory notes 155 Demographic and Health Survey(DHS)and vital registra- tion(VR). In addition to the familiar breakdown of infants Intervala under the age of one year into neonatal(0-27 days) and 0. Under-five 0-4 years postneonatal(28 days-11 months)periods(10), the latter age 1 0-11 months group was further divided into two intervals, 28 days-5 months 1.1 0-27 days and 6-11 months. Similarly, the child period between the first 1.2 Postneonatal 28 days-11 months and fifth birthday was divided into 12-23 and 24-59 months. 1.2.1 Early postneonatal 28 days-5 months The table here summarizes the definitions of the age break- 1.2.2 Late postneonatal 6-11 months down Child The mortality rates presented in Annex Table 2b are expressed 2.1 12-23 months as the probability of dying during each period, for those who 2.2 Early childhood 24-59 months ave survived until the beginning of that period. Therefore the he upper limit of the interval refers to completed days, months or years totals are not equivalent to the sum of the rates of the compo- nent age groups From DHS raw data sets, UNICEF collaborated in re-analysing them to compute detailed age-specific death rates, following the DHS approach, using synthetic cohort probabilities of death(11). In order to obtain sufficient robustness in the estimates these represent the period of five years prior to the surveys. No adjustments have been made for reporting issues such as heaping in these calculations VR data reported by Member States (1)are the other source where age-specific mortality can be computed, although the current under-one mortality age split that WHO requests does not allow detail within the postneonatal mortality rate. Thus, only neonatal and postneonatal mortality rates are presented in Annex Table 2b. For these two rates, we applied the following formula based on live births(12) Neonatal mortality rate= neonatal deaths/ live births Postneonatal mortality rate=postneonatal deaths /(live births-neonatal deaths) For the other age groups, we applied a standard formula from the abridged lifetable n,M where .1+n(-A)M q, is the probability of dying between exact ages x and X+n; n is the interval of the age group expressed in years; x is the exact age at the beginning of the age group MM is the age-specific death rate of the age group between x and X+n; and n a is the fraction of last age interval of life In this table we relied as much as possible on empirical data; for the denominators (live births and population of age-specific death rates) national data were given priori- ty, otherwise the estimates from the United Nations Population Division were used (3) Comparisons across countries should be made with great caution as the results are not directly comparable since the method of calculation varies depending on sources and there are different degrees of completeness of vital registration data submitted by Member State Those DHS and vR data that can be supplemented by other sources of information would serve as the basis of the analysis between the age groups, by country or by region. This insight into the level of mortality would possibly lead to identification of some cause-specific pattern for a better understanding of the epidemiological transi tion within childhood mortality
statistical annex explanatory notes 155 Demographic and Health Survey (DHS) and vital registration (VR). In addition to the familiar breakdown of infants under the age of one year into neonatal (0–27 days) and postneonatal (28 days–11 months) periods (10), the latter age group was further divided into two intervals, 28 days–5 months and 6–11 months. Similarly, the child period between the first and fifth birthday was divided into 12–23 and 24–59 months. The table here summarizes the definitions of the age breakdown. The mortality rates presented in Annex Table 2b are expressed as the probability of dying during each period, for those who have survived until the beginning of that period. Therefore the totals are not equivalent to the sum of the rates of the component age groups. From DHS raw data sets, UNICEF collaborated in re-analysing them to compute detailed age-specific death rates, following the DHS approach, using synthetic cohort probabilities of death (11). In order to obtain sufficient robustness in the estimates, these represent the period of five years prior to the surveys. No adjustments have been made for reporting issues such as heaping in these calculations. VR data reported by Member States (1) are the other source where age-specific mortality can be computed, although the current under-one mortality age split that WHO requests does not allow detail within the postneonatal mortality rate. Thus, only neonatal and postneonatal mortality rates are presented in Annex Table 2b. For these two rates, we applied the following formula based on live births (12): Neonatal mortality rate = neonatal deaths / live births Postneonatal mortality rate = postneonatal deaths / (live births – neonatal deaths) For the other age groups, we applied a standard formula from the abridged lifetable: where nqx is the probability of dying between exact ages x and x+n; n is the interval of the age group expressed in years; x is the exact age at the beginning of the age group; nMx is the age-specific death rate of the age group between x and x+n; and nax is the fraction of last age interval of life. In this table we relied as much as possible on empirical data; for the denominators (live births and population of age-specific death rates) national data were given priority, otherwise the estimates from the United Nations Population Division were used (3). Comparisons across countries should be made with great caution as the results are not directly comparable since the method of calculation varies depending on sources and there are different degrees of completeness of vital registration data submitted by Member States. Those DHS and VR data that can be supplemented by other sources of information would serve as the basis of the analysis between the age groups, by country or by region. This insight into the level of mortality would possibly lead to identification of some cause-specific pattern for a better understanding of the epidemiological transition within childhood mortality. Definition Intervala 0. Under-five 0–4 years 1. Infant 0–11 months 1.1 Neonatal 0–27 days 1.2 Postneonatal 28 days–11 months 1.2.1 Early postneonatal 28 days–5 months 1.2.2 Late postneonatal 6–11 months 2. Child 1–4 years 2.1 Toddler 12–23 months 2.2 Early childhood 24–59 months a The upper limit of the interval refers to completed days, months or years. nnMx nqx = 1 + n(1–nax ) nMx
156 The World Health Report 2005 ANNEX TABLE 3 Before estimating the number of deaths for individual causes, the first step is to obtain an estimated number of deaths from all causes combined which will constitute an envelope"to make sure that the sum of all cause-specific mortality does not exceed the estimated number of deaths in each country. The envelope itself is derived from the mortality rates from abridged life tables (4, 5) and applying them to the popula tion estimates obtained from the United Nations Population Division (3). The current mortality envelope was based on the joint work by WHO and UNICEF for the period Countries with a sound vital registration system(vR)with a relatively high coverage ould capture the representative pattern of causes of death at the national level. In addition to the levels of coverage, it is important to analyse carefully the quality of the coding pratices which should follow the rules of the International Statistical Clas ification of Diseases and Related Health Problems(ICD)(6, 8, 10 In some countries, improper completion of death certificates or systematic biases in diagnosis are quite For 72 countries where the VR coverage is over 85%, WHO considers VR as the gold standard and uses the pattern directly derived from VR, after adjusting for the ill-defined categories (e. g ICD-9 Chapter XVI, ICD-10 Chapter XVll; unspecified car- diovascular diseases; cancers of unknown sites; unspecified external causes) and checking cause-specific trends for the most recent years available. When estimating death rates for very small countries whereby a small change in the number of deaths substantially affects the overall cause-of-death pattern, an average of the last three years of data from their VR is used to avoid spurious trends In the absence of a Data and methods used for estimating under-five causes of death omplete vital reports/models ge-specific CHERG Incomplete vital mple registration studies, verbal autopsies, WHO programme estimates DHS MICS and mortality UNAIDS other surveys UN estimates timates mortality patterns All-cause mortality nd sex Country level age, specific mortality estimates Global and regi level age, sex and cause-
156 The World Health Report 2005 ANNEX TABLE 3 Before estimating the number of deaths for individual causes, the first step is to obtain an estimated number of deaths from all causes combined, which will constitute an “envelope” to make sure that the sum of all cause-specific mortality does not exceed the estimated number of deaths in each country. The envelope itself is derived from the mortality rates from abridged life tables (4, 5) and applying them to the population estimates obtained from the United Nations Population Division (3). The current mortality envelope was based on the joint work by WHO and UNICEF for the period 1990–2003. Countries with a sound vital registration system (VR) with a relatively high coverage would capture the representative pattern of causes of death at the national level. In addition to the levels of coverage, it is important to analyse carefully the quality of the coding pratices which should follow the rules of the International Statistical Classification of Diseases and Related Health Problems (ICD) (6, 8, 10). In some countries, improper completion of death certificates or systematic biases in diagnosis are quite frequent. For 72 countries where the VR coverage is over 85%, WHO considers VR as the gold standard and uses the pattern directly derived from VR, after adjusting for the ill-defined categories (e.g. ICD-9 Chapter XVI, ICD-10 Chapter XVIII; unspecified cardiovascular diseases; cancers of unknown sites; unspecified external causes) and checking cause-specific trends for the most recent years available. When estimating death rates for very small countries whereby a small change in the number of deaths substantially affects the overall cause-of-death pattern, an average of the last three years of data from their VR is used to avoid spurious trends. In the absence of a Global and regional level age, sex and causespecific mortality estimates Censuses Data and methods used for estimating under-five causes of death DHS, MICS and other surveys UN estimates Age-specific mortality rates Complete vital registration Incomplete vital registration Sample registration system UN population estimates All-cause mortality envelope by age and sex Life tables Independent studies / reports / models CHERG Epidemiological data from studies, verbal autopsies, WHO programme estimates UNAIDS Cause-specific mortality patterns Country level age, sex and causespecific mortality estimates Under-5 and adult mortality rates
statistical annex explanatory notes 157 complete VR system for obtaining cause-of-death information, sample registration systems are now implemented in a few countries such as china and India to obtain representative cause-of-death patterns (8) In many countries, however, VR systems are only operating in specific areas(se lected provinces or urban/rural areas) and there are virtually none in the majority of countries with high child mortality Estimates on cause-of-death patterns should be based on both limited sets of available data and extensive use of models Since areas not covered by the VR system are often rural and marginalized regions with a lower socioeconomic status than the covered ones, mortality patterns in both areas are likely to be different. a statistical model to make such an inference has been developed (13), based on the historical VR data for selected countries since 1950 that register at least 95% of all deaths. Although a few developing countries are included most countries reporting complete VR data to WHO are from developed regions and the countries included are mostly in the WHO European Region and the Region of the This model assumes that the broader cause-of-death pattern in high-mortality coun- tries would follow the historical health transitions previously observed in the current high-income and middle-income countries in the absence of major epidemics, natural disasters and war. Conditional on the values for all-cause mortality and income per capita, the model predicts the cause-of-death pattern for the three broader cause categories: communicable diseases; noncommunicable diseases; and external causes (injuries). This model was applied for assigning the under-five mortality envelope to the three broader causes in many high-mortality countries where no reliable informa- tion on cause-of-death patterns is available. Information drawn from neighbouring countries within the same region was also used to check the plausibility of model outputs(8). Once the allocation of the all-cause under-five mortality envelope into the three broader causes is done, the final step is to obtain the distribution of deaths from individual diseases or external causes within each of the three broad groups. For communicable diseases, from which the majority of children under five years of age die, estimates on specific diseases from the Child Health Epidemiology Reference Group(CHERG)(2), WHO technical programmes and UNAIDS are taken into account when making final estimates. The results of this joint work were then incorporated into the all-cause under-five mortality envelope, including deaths from remaining com- municable and noncommunicable diseases, and injuries representing 10% and 3% of global deaths, respectively. Because 2000 was the baseline year for the calculation of the estimates of the majority of the cause distribution, except for HIV/AIDS which is updated annually, cause-of-death distribution for 2000 was applied to the average under-five mortality envelope for 2000-2003 to obtain the average annual number of deaths from each cause he recent WHO work on neonatal mortality provided a sub-envelope of deaths dur ing the neonatal period out of the total under-five mortality envelope(14). Deaths attributable to HIV/AIDS were allocated based on annual mortality estimates pro luced by UNAIDS and WHO (15). For pneumonia, diarrhoea, malaria, and measles the CHERG estimates derived from single-cause models(16-18), as well as estimates from WHO technical programmes(19)and other published literature, were then trian gulated with the results of the multi-cause proportional mortality model, which takes into account the major causes of death simultaneously (20), to produce the new set of cause-specific mortality proportions
statistical annex explanatory notes 157 complete VR system for obtaining cause-of-death information, sample registration systems are now implemented in a few countries such as China and India to obtain representative cause-of-death patterns (8). In many countries, however, VR systems are only operating in specific areas (selected provinces or urban/rural areas) and there are virtually none in the majority of countries with high child mortality. Estimates on cause-of-death patterns should be based on both limited sets of available data and extensive use of models. Since areas not covered by the VR system are often rural and marginalized regions with a lower socioeconomic status than the covered ones, mortality patterns in both areas are likely to be different. A statistical model to make such an inference has been developed (13), based on the historical VR data for selected countries since 1950 that register at least 95% of all deaths. Although a few developing countries are included, most countries reporting complete VR data to WHO are from developed regions and the countries included are mostly in the WHO European Region and the Region of the Americas. This model assumes that the broader cause-of-death pattern in high-mortality countries would follow the historical health transitions previously observed in the current high-income and middle-income countries in the absence of major epidemics, natural disasters and war. Conditional on the values for all-cause mortality and income per capita, the model predicts the cause-of-death pattern for the three broader cause categories: communicable diseases; noncommunicable diseases; and external causes (injuries). This model was applied for assigning the under-five mortality envelope to the three broader causes in many high-mortality countries where no reliable information on cause-of-death patterns is available. Information drawn from neighbouring countries within the same region was also used to check the plausibility of model outputs (8). Once the allocation of the all-cause under-five mortality envelope into the three broader causes is done, the final step is to obtain the distribution of deaths from individual diseases or external causes within each of the three broad groups. For communicable diseases, from which the majority of children under five years of age die, estimates on specific diseases from the Child Health Epidemiology Reference Group (CHERG) (2), WHO technical programmes and UNAIDS are taken into account when making final estimates. The results of this joint work were then incorporated into the all-cause under-five mortality envelope, including deaths from remaining communicable and noncommunicable diseases, and injuries representing 10% and 3% of global deaths, respectively. Because 2000 was the baseline year for the calculation of the estimates of the majority of the cause distribution, except for HIV/AIDS which is updated annually, cause-of-death distribution for 2000 was applied to the average under-five mortality envelope for 2000–2003 to obtain the average annual number of deaths from each cause. The recent WHO work on neonatal mortality provided a sub-envelope of deaths during the neonatal period out of the total under-five mortality envelope (14). Deaths attributable to HIV/AIDS were allocated based on annual mortality estimates produced by UNAIDS and WHO (15). For pneumonia, diarrhoea, malaria, and measles, the CHERG estimates derived from single-cause models (16–18), as well as estimates from WHO technical programmes (19) and other published literature, were then triangulated with the results of the multi-cause proportional mortality model, which takes into account the major causes of death simultaneously (20), to produce the new set of cause-specific mortality proportions