UN Population Division, Department of Economic and Social Affairs,
with support from the UN Population Fund (UNFPA)
************************************************************************* This document has been prepared by the Secretariat of the United Nations Inter-Agency Task Force on the Implementation of the ICPD Programme of Action. For further information please contact the United Nations Population Fund, Task Force on ICPD Implementation, 220 East 42nd Street, New York, NY 10017 USA, or send E-mail to: firstname.lastname@example.org ************************************************************************** GUIDELINES ON A COMMON APPROACH TO NATIONAL CAPACITY BUILDING IN TRACKING CHILD AND MATERNAL MORTALITY FOR THE UN RESIDENT COORDINATOR SYSTEM INTRODUCTION 1. At the request of the UNDP Administrator, on behalf of the Secretary General of the UN, a first meeting of the Inter-agency task force on the implementation of the ICPD programme of action was convened. This meeting, held on December 13, 1994, at UN headquar-ters in New York, was attended by representatives of 12 UN agencies and organizations. It agreed to establish four working groups on: child and maternal mortality data; basic education and gender disparities; social policy-related issues; and women's empowerment. 2. These guidelines are the main outcome of the Working Group on a Common Approach to National Capacity Building in Tracking Child and Maternal Mortality which met at UNICEF Headquarters in New York on May 4, 1995. They are intended to provide a succinct and readable summary of the relevance of child and maternal mortality to human development, how the indicators are measured and who in the UN system can provide what specific kinds of assistance at the country level to governments, and more broadly civil society, in their efforts to assess infant, child and maternal mortality as they act to effect improvements. A list of key references for both child and maternal mortality is attached. I. CHILD MORTALITY Child mortality - an ongoing concern 3. Under-five mortality and its major component, infant mortality, have been used as measures of childrenūs well-being for many years. However, it was the International Conference on Primary Health Care held in Alma Ata in 1978 which first considered how child mortality could be reduced world-wide by a systematic development of a primary health care system. 4. The number of under-five deaths are huge, 12 or more million annually. But this number only tells part of the problem. These 12 million represent over 700 million years of productive life lost annually. And by far the majority of these lives being lost could be saved. The following table lists the causes of under-five deaths for developing countries. It shows that over 70% of these deaths are caused by diseases for which practical, low cost inter- ventions exist - involving immunization, ORT use, antibiotics and the like. ------------------------------------------------------------------- Table 1: Under-five deaths 1993, developing world Cause % of total ARI (mostly pneumonia) 25 Diarrhoea alone 23 Malaria alone 6 ARI-measles 5 Neonatal tetanus 5 Tuberculosis 2 ARI-pertussis 2 Measles alone 2 Diarrhoea - measles 2 Pertussis alone 1 Total 73 Source: WHO, The World health report 1995, Geneva, 1995 5. In this context, it is not surprising that child mortality measures are of key relevance in assessing progress in overall national development as well as progress for children. Both U5MR and IMR measure an end result of the development process rather than an input such as school enrolment ratio, per capita calorie availabili-ty, or the numbers of doctors per thousand population - all of which are a means to an end. 6. Furthermore, child mortality is known to be the result of a wide variety of inputs: the nutritional health and the health knowledge of mothers; the level of immunization and ORT use; access to maternal and child health services (including prenatal care); income and food availability in the family; the availability of clean water and safe sanitation; and the overall safety of the child's environment. International conferences 7. Specific mention of action to be taken on child mortality can be found in paragraph 8.16 of the Report of the international conference on population and development (UN ref. A/CONF.171/13, 18 October 1994). In particular this paragraph includes the following. Countries should strive to reduce their infant and under- five mortality rates by one third, or to 50 and 70 per 1000 live births, respectively, whichever is less, by the year 2000, with appropriate adaptation to the particular situa- tion of each country. By 2005, countries with inter-mediate mortality levels should aim to achieve an infant mortality rate below 50 deaths per 1000 and an under-five mortality rate below 60 deaths per 1000 births. By 2015, all countries should aim to achieve an infant mortality rate below 35 per 1000 live births and an under-five mortality rate below 45 per 1000. Countries that achieve these levels earlier should strive to lower them further. 8. In the context of UN sponsored international conferences, these same goals for the year 2000 were first mentioned in the World Summit for Children, which was held in New York in 1990. Subsequently, the United Nations Conference on Environment and Development, held in Rio de Janeiro in June 1992 repeated these goals, as did the World Summit for Social Development in Copenhagen in March 1995. Thus the target of reducing child mortality has both broad-based and long-term support. Definitions 9. The generally accepted definitions for under-five and infant mortality rates come from demography, are cohort based and can be stated as follows. Under-five mortality rate (U5MR): The probability of dying between birth and the fifth birthday (exact age 5 years), expressed per 1000 live births. Infant mortality rate (IMR): The probability of dying between birth and the first birthday (exact age 1 year), expressed per 1000 live births. 10. The infant mortality rate is often computed as the ratio of deaths of children under one year of age occurring during a given period and births in the same period. The difference between this and the above stated cohort measure are very small. However, a similar period based estimate cannot be used for the under-five mortality rate. Data sources and estimation methods 11. In the developed countries, measures of child mortality at the national level have traditionally come from the registration of births and deaths. If births and deaths are completely recorded, and the compilation of statistics from the registration system are timely, then these are the preferred mortality estimates. 12. However, the major problem with vital registration as a data source is its quality. In many developing countries, birth registra-tion is incomplete In a still larger number of countries, the recording of child deaths is incomplete. Registration of deaths after infancy is, in general, more complete than in infancy, but the recording of the population by age in childhood is also subject to error. A further problem with vital registration systems is the frequent delay in compilation and publication. 13. Experience with programmes to improve vital registration coverage have been rather discouraging. Work in this area suggests that complete vital registration evolves over time with general administrative development and as the importance of records becomes evident to, and used by, the general population. 14. However, sample registration is proving valuable in some of the world's largest countries. A successful example is the Indian sample registration system, introduced in some states in the 1960s, and currently operating throughout the country. Bangladesh has developed a similar system which, after some uncertainty, appears to be stabilizing. China has recently embarked on a related sample system. 15. In countries where the vital registration system as a basis for child mortality estimates is of uncertain or unacceptable quality, as is the case for most developing countries, information from some type of household survey is required to validate, calibrate or substitute for vital registration estimates. A distinction can be drawn between prospective and retrospective surveys. 16. The essential characteristic of a prospective survey is that a defined population is followed over time, with the vital events occurring to the population being recorded. A typical prospective survey involves the initial recording of populations in a sample of areas. The population is then resurveyed at regular intervals, with the reported events checked against changes in household composition between rounds in order to minimize omission. With careful fieldwork, prospective surveys can provide relatively accurate estimates of child mortality. Such surveys, however, require careful fieldwork over an extended period of time to estimate trends, and because of sample size limitations, may have to be extended over several years to provide stable estimates of child mortality levels. This continuity of effort can be difficult to achieve in some developing country settings. 17. Retrospective surveys typically obtain information from mothers on the survival of their children. Such surveys provide the main source of estimates for most developing countries. 18. Response errors, which arise during data collection, are a major source of poor quality mortality data from all sources. This is of particular concern in retrospective surveys where questions require adequate specification and interviewers must be well trained and supervised. In addition to such response errors, the retrospective survey techniques are affected to a greater or lesser extent by potential selection bias, because in order for a child to be reported the mother must be a member of the study population at the time of the survey. Thus, either death or emigration of the mother can affect the reporting coverage. 19. The most extensively utilized retrospective survey techniques are: questions to women on aggregate numbers of children born and dead, often referred to as the 'Brass' questions; and questions to elicit maternity histories, where each woman is asked for the date of birth and, if applicable, the age at death of each of her live- born children. Brass questions have been used in most developing countries and are the simplest and least costly of the two techniques to apply. They have worked well in a wide variety of social contexts and of data collection vehicles, including censuses. 20. Maternity history data have provided a wealth of information on child mortality in developing countries. Complete maternity histories, such as those utilized by the Demographic and Health Surveys, are more onerous to collect than data from Brass questions, and hence have been generally limited to national household surveys where sample sizes are insufficient to provide detailed sub-national mortality estimates. Maternity history data have made a particular contribution to the exploration of differen- tials and associations in child mortality. 21. There are other retrospective survey techniques, such as asking about recent household deaths by age, or questions on the survival of a motherūs previous birth. However, for these techniques either the experiences have been mixed, or they apply more appropriately to population sub-groups - such as mothers who give birth in health centres. 22. More information on these and other survey techniques can be found in Child mortality since the 1960s, and in Approaches to the measurement of childhood mortality: a comparative review. Helping countries track child mortality 23. The activities involved in tracking child mortality at the country level can be usefully divided into three components: determine what mortality data exist; generate reasonable time series of mortality estimates; and fill data gaps. These components are further detailed in the following. Determine what mortality data exist 24. This task can generally be done by local demographers. A good example of what should be done to describe these data and to provide an updateable record is provided in Child mortality since the 1960s (see references). A typical country profile in this publication presents the key features: the available data listed and referenced, and all data graphed. These latter charts (one each for under-five mortality and infant mortality) provide a very useful visual assessment of the amount of data, the consistency of data from different sources, and the trend of mortality over time. Generate reasonable time series of mortality estimates 25. There are two important aspects to generating time series estimates (separately) of the under-five and infant mortality rates for period 1960 to 1995. The first is the ensuing result of a single set of estimates. The second is the process of obtaining the time series, which requires assessment of the existing data and discussion among experts. 26. There are often several different estimates of child mortality used by different national bodies, be they in the government, public or private sectors. Different estimates of child mortality for the same or similar time period are almost always detrimental to concerted efforts to reduce child mortality, since the lack of agreement is often associated with a lack of a coordinated effort on reducing child mortality. Different mortality estimates can also lead to different, and separate, programme strategies for mortality reduction. The aim should be to minimize differences between mortality estimates and to obtain very broad country-wide support for a single and consistent set of under-five and infant mortality estimates. 27. Arriving at a single set of under-five or infant mortality estimates is not an easy task, since there is no unique best method and the country situations can differ widely. For example, the data from seven data sets for Bolivia, shown in figure 1, display a very consistent trend and coherency over the period 1960-92. The situation in figure 2, for Papua New Guinea, shows the other extreme. The country has only two data sets, with a very low data consistency since in the period 1965-67 the 1980 census reports an under-five mortality rate of just over 100 per 1000 live births, whereas the 1970 census reports a mortality rate of around 200 for the same period. 28. In the case of Bolivia, fitting a line representing a single time series of mortality estimates from 1960 to 1995 appears feasible, and most reasonable attempts could be expected to fall within a narrow band (see the annex for further information on line fitting). Fitting such a line to the data in figure 2 for Papua New Guinea would give very questionable results since possible lines would fall within a very wide mortality range, from a line through the 1970 census data projected to 1995 with the same trend as that given by the 1970 census, to a line through the 1980 census data projected both backwards to 1960 and forward to 1995 with the same trend as the 1980 census. 29. Even in the case of Bolivia - and this is an example of a very consistent set of data - there would be individual variations if different experts attempted to produce a mortality time series. For countries with less consistent data the derivation of a single time series becomes more variable, and an explicit methodology for obtaining a consistent and repeatable time series is needed. 30. The work of Hill and Yazbeck in Trends in child mortality provides a model for generating such a time series at the country level. This important work is summarized in an annex to these guidelines. A key consideration in this model is that it be repeatable and useable by others, particularly at the country level. Hence countries can understand, adopt and implement this methodology themselves, leading to a greater awareness and commitment within each country. 31. The Hill and Yazbeck methodology aids data assessment, since the choice of regression weights explicitly assigns assessments of data source quality (see Annex). Assessment approaches which rely on comparisons between data sources can be found in Child mortality since the 1960s (pages 12 to 15). At the same time, the quality of individual sources can and should be explored. Such assessments are helped if separate quality studies have been implemented as part of the data collection process. But fairly simple data analyses, such as the calculation of male/female ratios and their comparison against known standards, can throw useful light on data limitations. Fill data gaps 32. A data ūgapū is used here to identify country situations where there are either no mortality data referenced to a year within the last five or, where there are such data, they are inconsistent or refer to a time before a catastrophic occurrence of national impact - such as civil conflict or major natural disaster. 33. Measured mortality, referenced within the last five years, is considered the minimum requirement which all countries should be able to meet. A more frequent measurement of mortality is advan- tageous in general, preferably annually, but care has to be taken to balance frequency of mortality measurement with the capacity of a country. For example, if a country is facing economic hardships and has a high child mortality level, attempting to measure mortality every year will absorb significant country resources, resources which could be better used in reducing child mortality rather than its frequent measurement. This example fits the situations in many African countries. 34. Having identified a mortality data gap exists, how it should be filled depends on the country situation. Countries with gaps to fill can be divided into two groups: those which have (or had) an adequate vital registration system, and countries without. An "adequate" vital registration system is defined here as one which covers over nearly all births and under five deaths in a country. The term "nearly all" is used deliberately; it could have been replaced with 'at least 80% of births and under-five deaths'. However, a more relevant specification is whether vital regis- tration can play the major role in tracking child mortality. Clearly a vital registration system which covers all births and under five deaths meets this specification. But so also does a system which covers enough of the births and deaths so that periodic censuses or large surveys can be used to derive an adjustment factor. This adjustment is then applied to the annual vital registration system estimates to arrive at good quality national child mortality estimates. a) Countries which have (had) an adequate vital registration system 35. Countries which had an adequate vital registration but now have data gaps, are few in number. But this situation can arise when existing systems have been run down or, as in the case of man- made or natural catastrophes, when country infrastructures have been adversely affected. Systems may not have stopped functioning, but their coverage of births or deaths may have declined, or the reporting lag between occurrence and reported estimate may have increased considerably. 36. Filling such data gaps requires a review of the vital registration system to determine what the problems are, and a support project initiated to correct them. In some situations, where there are no mortality data for the last five years, or since a catastro-phe, a survey may be required to provide more current data until the vital registration system is functioning adequately again. b) Countries without an adequate vital registration system 37. Countries without an adequate vital registration system comprise the majority of developing countries. An ultimate, long- term aim is to have complete vital registration for all countries. However, as noted earlier, experience indicates that complete vital registration evolves over time, with general administrative deve- lopment and public use. This is not to say the development of complete systems should not be supported, but they do not get built quickly. Projects for vital registration development need to recognize both the several years for which support will likely be required, as well as the implementation of household surveys to provide mortality data in the interim. 38. For those countries where adequate registration systems are sometime in the future, either retrospective or prospective surveys need to be used to fill data gaps. In general the technique of choice is the Brass questions in retrospective surveys, since these are the easiest and least costly to implement over a wide range of data collection vehicles. Where correlations of mortality with other factors are particularly sought, maternity history questions should be considered. 39. In situations where the primary source of retrospective survey data, mothers, are likely to introduce a significant selection bias, prospective surveys can be considered for filling data gaps. But such situations require careful review, balancing the country capacity to carry out such a logistically demanding survey against the degree to which other less costly and simpler techniques may suffice. 40. Any action on filling data gaps must take into account feasible data accuracy and the use to which the data are to be put. In the case of feasible data accuracy, and including both sampling and non-sampling error components, a useful general rule is that mortality measurements have an uncertainty of at least plus or minus 10% of the measurement value. For example, if an under-five mortality rate of 100 is measured, the actual rate should be interpreted as being, at best, somewhere in the range 90 to 110, and is often outside this range. Reducing the uncertainty of measurement below this 10% level is both difficult and costly. 41. Measurement is of little value if the data are not used. Hence consideration should be given to how mortality data can be used to more effect. As noted earlier in these guidelines, exploring cause of death is useful in helping to better target programme interven-tions. Additionally, it is beneficial to get users as well as producers together, not only to discuss existing data systems and additional data needs, but particularly to clarify how existing data are presently used, and how new mortality data will be used. In this the guideline should be that where a cost is incurred in measuring child mortality, this cost should produce a greater benefit in mortality reduction, and not solely result in a measurement report, however imposing and official it may look. II.MATERNAL MORTALITY Maternal mortality reduction - an overarching goal 42. Deaths of women due to pregnancy or childbirth is a major public health problem in developing countries. On average, 500,000 women die from maternity-related causes every year -approximately one maternal death every minute. 99 percent of these deaths occur in developing countries with the majority concentrated in Africa and South Asia. Although there has been a significant decline in child mortality in recent years, the gap between maternal mortality ratios in the developing and the developed countries remains wider than for any other health indicator. While the absolute number of maternal deaths may seem small in comparison with the number of infants dying, the risk of death accumulates for women with each pregnancy. For example, the life-time risk of death from pregnancy and child birth for a woman in Africa is 1 in 20 while this risk is 1 in 10,000 for a woman in northern Europe. The lack of attention that has been paid to this problem is a reflection of the lack of importance given to women's health issues in general. International conferences 43. The first time that the international health community's attention was clearly focussed on maternal deaths was in 1987, when the International Conference on Maternal Mortality was held in Nairobi, Kenya. This conference reflected a consensus that the number of maternal deaths in the developing world was too high, unnecessarily so, and could be prevented or reduced considerably. A second important moment for the Safe Motherhood Initiative was the 1990 World Summit for Children. The Summit Declaration and Plan of Action included the reduction of maternal mortality by half as one of the seven major goals to be achieved between 1990 and the year 2000. 44. Most recently, the International Conference on Population and Development (ICPD) in Cairo, and the World Summit for Social Development held in Copenhagen in March 1995, reiterated the maternal mortality reduction goal set forth in Nairobi and the World Summit for Children, and expanded it to include a further reduction in maternal mortality of 50% by the year 2015. In addition, the ICPD Programme of Action recommends that ... Countries with intermediate levels of mortality should aim to achieve by the year 2005 a maternal mortality rate below 100 per 100,000 live births and by 2015 a maternal mortality rate below 60 per 100,000 live births. Countries with the highest levels of mortality should aim to achieve by 2005 a maternal mortality rate below 125 per 100,000 live births and by 2015 a maternal mortality rate below 75 per 100,000 live births. However, all countries should reduce maternal morbidity and mortality to levels where they no longer constitute a public health problem. Disparities within countries and between geographical regions, socio-economic and ethnic groups should be narrowed... 45. Countries which have formally committed themselves to achieving the maternal mortality reduction goal are also res- ponsible for monitoring progress toward that end. To accomplish this, close collaboration among international and national agencies, governments and non-governmental organizations is essential. Indicators 46. In the context of the World Summit for Children, UNICEF, WHO, UNESCO and others have worked closely together to agree on a basic set of indicators to recommend to countries for monitoring progress toward the goals. In relation to the World Summit for Children and Health for All maternal mortality reduction goals, the two monitoring indicators agreed upon by WHO and UNICEF are the Maternal Mortality Rate (ratio)(MMR): Annual number of maternal deaths per 100,000 live births and the Annual Number of Maternal Deaths. Measurement problems 47. There are several features of maternal mortality, however, that make it technically difficult to measure. First, as compared to other commonly measured demographic events (such as births or under five deaths), it is a relatively rare event. Second, maternal deaths are often not reported, or when they are, they are not correctly classified as maternal deaths. As a result, most official measures of maternal mortality are under-estimates. 48. The relative infrequency of maternal deaths means that large populations need to be studied which makes such studies very costly. If the study population or sample is too small, the number of deaths will not be large enough to yield reliable, stable estimates. WHO has calculated that to establish a maternal mortality ratio of 300 (per 100,000 live births), correct to within 20% (95% confidence intervals) would require a sample size of 50,000 births. Of course, many more households would have to be interviewed to yield 50,000 births. 49. Maternal mortality estimates generally have wide margins of error. This presents a particular problem in measuring trends over time because, even if consecutive studies showed a decline over time, it may not be possible to rule out chance as an explanation for this finding. Figure 1 illustrates this point using data from a direct household survey. Scenario B assumes a 50% reduction in maternal mortality and Scenario C a 25% reduction. In both cases the 95% confidence limits overlap with the baseline estimate and it is therefore not possible to measure a statistically significant difference between the two estimates. In summary, measuring trends is much more difficult than generally believed, even using the new sisterhood and network methods. 50. Vital registration is usually relatively complete in most developed and a few developing countries. However, in most deve- loping countries, this is not the case. One of the reasons why many deaths in developing countries are not registered is that they do not occur in health facilities, where health personnel would be required to report them. Many deaths occur in the home or on the way to a hospital and are consequently not recorded. 51. Even in countries with relatively complete vital registration systems, misreporting of maternal deaths is a serious problem. The mis-reporting of maternal deaths means that they were reported but not properly classified as a maternal death. A maternal death is defined as the death of a woman while pregnant or within 42 days of termination of pregnancy, irrespective of the duration and the site of the pregnancy, from any cause related to or aggravated by the pregnancy or its management, but not from accidental or incidental causes. 52. Therefore, to properly report a maternal death, it is necessary to know not only that the woman died but the timing and the cause of the death as well. Few maternal deaths actually take place in obstetric wards because when a life-threatening situation arises, the patient is moved to another department and the cause of death is not certified by an obstetrician or the death certificate may not mention the obstetric cause which triggered the series of complica-tions leading to death. Even in the United States, studies have shown misreporting of between 25% and 70% of maternal deaths. Data sources 53. In addition to vital registration systems, the main sources of data on maternal mortality are household surveys, reproductive age mortality surveys (RAMOS), hospital data and community studies. The RAMOS studies are likely to produce the most reliable estimates of maternal mortality but are too costly to implement at the national level on a regular basis. While data from hospitals and health centres can be informative, they can also be misleading - this is particularly true when data come primarily from hospitals which specialize in maternal care, where mortality rates can be much higher than in the general population. On the other hand, under reporting and mis-classification can lead to gross under- estimates of maternity-related mortality, even in countries where all or most deaths are medically certified. Health systems in a large number of developing countries do not have adequate popu- lation coverage. Community studies of maternal mortality are more common in many developing countries, but these are for very limited geographical areas, and the quality varies enormously. 54. Household surveys require large sample sizes, even with the new sisterhood and network methods. While careful field work can produce good quality estimates, large surveys have often produced poor results. The sisterhood method has been developed more recently and minimizes the number of households that need to be visited in order to obtain information on a large number of women. The method asks all adult women in a household about the survival of their sisters: how many sisters they had who survived to adulthood and how many died of pregnancy-related causes. This information is then converted into a life-time risk of dying from maternal causes and maternal mortali-ty. Questions based on the sisterhood method have been successfully added to many of the Demographic and Health Surveys. However, the sisterhood method produces estimates which reflect maternal mortality levels of ten years or more in the past. Therefore, they cannot be used for monitoring progress toward the maternal mortality reduction goal during the current decade. They also do not provide information on cause of death. Nonetheless, estimates of maternal mortality derived from the sisterhood technique are valuable, particularly in places where no reliable community studies are available and/or where vital registration is inadequate. 55. Maternal mortality epidemiologic surveillance systems may be appropriate in countries where civil registration is relatively more complete and where most births take place in health facilities. The Pan American Health Organization (PAHO) has been working to develop this methodology further. However, trade-offs should be considered on the return of investment in this type of approach since the improvement of civil registration systems is a long-term undertaking and even in the best systems the measurement of maternal mortality presents specific problems and this type of surveillance does not provide information relevant to programme planners. 56. Because of the measurement problems described above, many of the national level maternal mortality estimates regularly reported by international agencies, and used by national governments, are not accurate reflections of the present situation and are of limited value in measuring trends over time. This raises a serious problem for monitoring the maternal mortality reduction goal since it is set relative to a 1990 baseline. This does not imply that all attempts to measure maternal mortality should be abandoned. However, it is important that the limitations of using these estimates for monitoring progress in maternal mortality reduction be fully recognized: they are costly to produce, may not be nationally representative or, in the case of sisterhood estimates, provide estimates which are not current. Finally, maternal mortality ratios alone do not provide the information needed for development of programme interventions or policy formulation. Model-based estimates 57. An alternative method is to base estimates of maternal mortality on a mathematical model using widely available predictor variables. At present, the WHO Maternal Health and Safe Motherhood Programme and the UNICEF Planning Office are collaboratively pursuing this option, at least for those countries which are known to have weak data or no data at all on maternal mortality. 58. Preliminary results are promising, although the predicted MMRs resulting from a mathematical model may be somewhat imprecise because of wide margins of error. Given the weakness of the existing data on maternal mortality, however, the model-based estimates will likely be an improvement. For countries which lack accurate national level estimates, the model-based estimates offer a sound alternative to investment in large-scale surveys. They provide, at minimum, an indication of the order of magnitude of the problem which can be used to stimulate action to reduce maternal mortality. 59. Work on the development of model-based estimates is continuing and final results are expected to be available by the Fall of 1995. In addition to UNICEF and WHO, UNDP and The World Bank have expressed interest in using the model-based estimates for those countries which have no reliable estimates for maternal mortality. Process indicators 60. An important alternative to monitoring the impact of programmes is to monitor the processes which are known to reduce maternal mortality. There are several distinct advantages to this approach. First, it avoids the substantial expense involved in generating maternal mortality rates, which in many cases may not be accurate, or reflect a situation ten years or more in the past. Second, process indicators can provide information essential for guiding policies and programmes. 61. In 1992, UNICEF issued a set of guidelines for monitoring progress toward maternal mortality reduction which proposed a series of process indicators [D. Maine, et al., Guidelines for Monitoring Progress in the Reduction of Maternal Mortality. (A Workin Prog-ress). UNICEF Statistics and Monitoring Section, October 1992]. These process indicators are based on the assumption that the most effective strategy for reducing maternal mortality is to increase access to prompt, adequate emergency obstetric care (EOC) and therefore are designed to measure progress toward improving access to, utilization of and the quality of EOC services. Using process indicators will help programme planners identify priority interventions and areas, as well as aspects of the programme that need strengthening. Thus, monitoring of process indicators serves a variety of purposes - not just data gathering for its own sake. 62. Following this pioneering work on indicator development, WHO convened a technical working group, in 1993, to make recommen- dations on data collection and analysis for monitoring the maternal mortality and coverage of care goals. The technical working group met at a time of growing consensus on the content of programmes for improving maternal health, growing convergence on the essential package of indicators for monitoring progress, and growing need for guidance for the collection and utilization of these indicators. The main conclusion of the Technical Working Group meeting was that ... there is a need to recognize that, at the national and subnational levels, impact indicators are much less likely to be useful for programme management than process indicators and are insufficiently accurate for monitoring purposes. From an international perspective, impact indicators are probably still needed, principally for advocacy purposes, but the uncertainty over their usefulness at lower levels and the difficulties and cost in collecting the necessary information make it difficult to attach any sense of priority to the two mortality indicators. Consumer information is needed in order to enable countries and programme managers to make a decision on this issue...(p. 26) 63. The outcome of the meeting included a series of recommen- dations on specific indicators and methodologies [See Indicators to Monitor Maternal Health Goals. Report of a Technical Working Group. Geneva, 8-12 November 1993. WHO Division of Family Health]. 64. USAID and The World Bank have also conducted in-depth reviews of indicators for monitoring and evaluation of repro- ductive health programmes and there is now a growing consensus on the use of process indicators for monitoring progress. 65. It should be noted, however, that there is relatively little experience in the use of these indicators and additional field testing (of the indicators and data collection protocols) is required. Further guidance is also needed on the use and interpretation of all these indicators. ------------------------------------------------------------------- SELECTED BIBLIOGRAPHY CHILD MORTALITY Child Mortality Estimates: UNICEF, The State of the Worldūs Children 1995, UNICEF, New York. DIESA, Mortality of children under age 5 - World estimates and projections 1950-2025, ST/ESA/SER.A/105, United Nations, New York, 1988. DESIPA, World population prospects - the 1994 revision, United Nations, New York, 1995 forthcoming. Estimation Methods: DIESA, Step-by-step guide to the estimation of child mortality, ST/ESA/SER.A/107, United Nations, New York, 1990. DIESA, Manual X. Indirect techniques for demographic estimation, ST/ESA/SER.A/81, United Nations, New York, 1983. DHS, Model A questionnaire, DHS-II Basic documentation, Macro International, Columbia, USA, 1990. K. Hill, Approaches to the measurement of child mortality: a comparative review, Population Index, Vol. 57, No. 3. P. H. David et al, Measuring childhood mortality: A guide for simple surveys, UNICEF, Amman, Jordan, 1990 Child Mortality Databases: DESD, Child mortality since the 1960s - A database for developing countries, ST/ESA/SER.A/128, United Nations, New York, 1992. K. Hill and A. Yazbeck, Trends in child mortality, 1960-90: Estimates for 83 developing countries, Background paper number 6, October 1994. Source Key for Figures 1 and 2 Figure 1: Bolivia under-5 mortality EDNi75 - Encuesta demogr fica nacional, 1975, indirect estimates CENSi76 - Census, 1976, indirect estimates EDNi80 - Encuesta demogr fica nacional, 1980, indirect estimates ENPVi88 - Encuesta nacional de poblacion y vivienda, 1988, indirect estimates ENDSd89 - Enquesta nacional de demogr fia y salud, 1989, direct estimates ENDSi89 - Encuesta nacional de demogr fia y salud, 1989, direct estimates DHSi94 - Encuesta nacional de demogr fia y salud, 1994, direct estimates Figure 2: Papua New Guinea under-5 mortality CENSi71 - Census, 1971, indirect estimates CENSi80 - Census, 1980, indirect estimates All data from Child mortality since the 1960s - A database for developing countries, except for DHSi94 data, which come from the published DHS report on Bolivia. ------------------------------------------------------------------- SELECTED BIBLIOGRAPHY MATERNAL MORTALITY General: WHO, Maternal Health and Safe Motherhood Programme, Division of Family Health. Mother-Baby Package: Implementing Safe Motherhood in Countries. Practical Guide. (WHO/FHE/MSM/94.11), 1994. Maine, D. Safe Motherhood Programs: Options and Issues. Columbia University, Center for Population and Family Health, 1990. Measurement of Maternal Mortality: Campbell, O. and W. J. Graham. Measuring Maternal Mortality and Morbidity: Levels and Trends. Maternal and Child Epidemiology Unit Publication No.2, London: London School of Tropical Medicine and Hygiene, 1990. Graham, W. J., "The Sisterhood Method for Estimating the Level of Maternal Mortality: Seven Years' Experience." The Kangaroo, December, 1994, p.82-87. Graham, W. J. and P. Airey. "Measuring Maternal Mortality: Sense and Sensitivity." Health Policy and Planning 2:323-333, 1987. Indicators for Monitoring Maternal Mortality Reduction: Maine, D. et al., Guidelines for Monitoring Progress in the Reduction of Maternal Mortality. (A Work in Progress). UNICEF Statistics and Monitoring Section, October 1992. (Update forth- coming in Fall 1995). WHO, Maternal Health and Safe Motherhood Programme. Indicators to Monitor Maternal Health Goals. Report of a Technical Working Group. Geneva, 8-12 November 1993. Bulatao, R. A. and L. B. Shrestha. Key Indicators for Reproductive Health Projects. Draft 5 June, 1995. The World Bank. Graham, W. J. and V. Filippi. Monitoring Maternal Health Goals: How Well Do the Indicators Perform? Maternal and Child Epidemiology Unit Publication No.2, London: London School of Hygiene and Tropical Medicine, 1990. International Data Sets on Maternal Mortality Ratios: WHO, Division of Family Health. Maternal Mortality. A Global Factbook. Compiled by Carla AbouZahr and Erica Royston, Geneva, 1991, and the WHO Database on Maternal Mortality. Pan American Health Organization (PAHO). Regional Plan of Action for the Reduction of Maternal Mortality in the Americas (CE111/11), 27 May 1993. The State of the World's Children Report, Human Development Report, World Development Report. Model-based Estimates of Maternal Mortality: Stanton, C. And K. Hill. Model-Based Estimates of Maternal Mortality. Report to UNICEF/WHO. July 28, 1994 Department of Population Dynamics, John Hopkins University School of Public Health. [A final report on this work is forthcoming from UNICEF and WHO]. ------------------------------------------------------------------- BRIEF SUMMARY OF TRENDS IN CHILD MORTALITY, 1960-90: ESTIMATES FOR 83 DEVELOPING COUNTRIES BY K. HILL AND A. YAZBECK A1. The report describes a methodology for trend fitting, applies it to the data for 83 countries, presents the results in text and charts country by country, and draws some general conclusions about the rates of decline of child mortality since 1960. Methodology A2. There are many ways in which a set of estimates can be obtained from a series of observations, and in which extrapolations forward or backward to any time point can be made. The simplest procedure is hand smoothing: drawing a freehand curve through a set of observations, and extending its general trend onwards to some time point for which an estimate or projection is required. Such a procedure is unlikely to be objective - different analysts would almost inevitably draw different lines, particularly for extrapola- tions beyond the latest observations. A3. Regression analysis offers a set of possible approaches: robust regression, locally-weighted least squares, weighted least squares, or ordinary least squares. Such regression techniques offer a greater degree of objectivity than hand smoothing, but still require the choice of model specification. A4. The approach adopted in the Hill and Yazbeck report is to fit a regression line to the relationship between child mortality indicators and their reference dates using weighted least squares. The basic model assumes that the rate at which child mortality changes is linear in time, that is, that child mortality changes at a constant annual percentage rate over some specific time period. The simplest model maintains a constant rate of change in child mortality over the entire period studied. The most complex model used in the report allows the rate of change of child mortality to alter every five years. The choice of model depends on the number of mortality observations by time period. A5. Weighted least squares is used because a substantial body of evidence suggests different validity weights for different types of observations. For example, it is generally thought that the quality of retrospectively reported information deteriorates with the length of time since the events reported. All estimates from vital registration or prospective surveys are given initial weights of 1.0; in the former case, the weight is justified by the typically large number of events involved and by the lack of any substantial lag between event and report; in the latter case, the high weight is justified by the lack of lag and by the accuracy enforced by the data collection methodology. A6. Estimates derived from maternity histories are assigned weights that vary with the length of time before the survey to which the estimate refers. Specifically, estimates for the five years before the survey are given a weight of 1.0, for periods five to nine years before the survey, 0.8, and periods 10 to 14 years before the survey, 0.6, and for yet longer periods, 0.4. Weights for indirect estimates based on the proportions dead of children ever born vary by age group of mother; estimates based on reports of young women are given low weight, zero for women aged 15 to 19, and 0.2 for women aged 20 to 24, because of the selection problems which affect such estimates - early childbearing is highest among the poor, who also suffer the highest child mortality rates. Estimates based on reports of women aged 25 to 29 (0.9) and 30 to 34 (1.0) get the highest weights. Then, as age increases, the weights decline slowly, on the grounds that information about events longer ago is more prone to error. A7. The observation-specific weights described in the foregoing are essentially based on the authorsū judgement and experience. However, regression techniques can be used to estimate robust weights for particular types of observation. These techniques have been applied by the authors on a subset of 13 countries with a large number of different types of observation - particularly indirect estimates based on the Brass questions and direct estimates based on birth histories. They find broad agreement between the robust regression weights and those described earlier. Applying the methodology A8. For each country, step one of the smoothing and extrapolation process fits the regression model using appropriate date variables and the weights described earlier. The infant mortality rate and the under-five mortality rate are fitted independently. The only subjective element in the process is in the decision concerning how many slope variables to include in the model. The observations and fitted line are displayed graphically. In step two, the step one results are examined, and data sets that are clearly aberrant are identified - such as vital registration sequences that fall consis- tently below all other infant mortality estimates, or indirect estimates that are clearly inconsistent with the bulk of the other mortality estimates. In general, the weights for that entire data set are reduced by a constant factor that is usually zero. A9. Egypt provides an interesting example of the application of the methodology. Figure A1 shows the observations and final regression estimates for infant mortality. The vital registration and observa-tions from the 1976 and 1986 population censuses are clearly out of line with all other survey estimates. The step one regression line (not shown) is pulled down by the registration data, particularly for the 1960s, giving the almost certainly erroneous impression of rising infant mortality in the 1960s. A10. In addition, and generally applied throughout the report, it is assumed that response errors are more likely to result in under- estimates of child mortality than in overestimates. Thus when two data sets indicate very different levels, that set indicating higher mortality is assumed, other things being equal, to be more likely to be right. In step two in the case of Egypt, the registration and census data were all given zero weights. A11. The intention of the methodology is to provide a transparent and partially objective way of fitting a smoothed trend to a set of observations, and of extrapolating the trend to cover the period from 1960 to the present. However, there are subjective judgements which still have to be made. Step one depends on the weights selected for different types of data. At the same time, while analysts might choose different weights, the weights used in the report are broadly supported by robust regression results. It is in the second step that subjective judgements are likely to have a significant impact - primarily in the decision as to whether, and if so by how much, to underweight entire data sets. Country specific results A12. Each of the 83 countries reported has its own two page section which is divided into four parts. The first part lists the data sets used in the analysis, and gives both the initial and final weights utilized in the model fitting. The second part notes any unusual characteristics of the application, such as overriding an apparent trend when extrapolating child mortality on the basis of known periods of civil disruption. The third part summarizes the results of the model in the form of estimates of under-five mortality and infant mortality for the period 1960 to 1992. In addition to the mortality estimates, the time period coefficients estimated by the model and the implied annual rates of change for five year periods are also given. The fourth section presents graphs showing all available observations of infant and under-five mortality by source, together with the fitted trend line. ------------------------------------------------------------------- Source Key for Figure A1 Figure A1: Egypt infant mortality CENSi76 - Census, 1976, indirect estimates EFSd80 - Egyptian fertility survey, 1980, direct estimates EFSi80 - Egyptian fertility survey, 1980, indirect estimates ECPSi84 - Egypt contraceptive prevalence survey, 1984, indirect estimates CENSi86 - Census, 1986, indirect estimates EDHSd89 - Egypt demographic and health survey, 1988-89, direct estimates EDHSi89 - Egypt demographic and health survey, 1988-89, indirect estimates EPSd91 - Egypt papchild survey, 1991, direct estimates EPSi91 - Egypt papchild survey, 1991, indirect estimates Vital reg.- Vital registration, 1960-87 Estimates - Regression estimates from step two. ------------------------------------------------------------------- AGENCY PROFILES The World Bank (IBRD) In the area of common data systems for monitoring child and maternal mortality, the World Bank uses the indicators published by UNICEF and WHO. Both the under-five mortality rate and the maternal mortality ratio are considered "priority poverty indicators" that are required by the Bank's Operational Directives to be included in Bank country economic reports. The Bank has also started to incorporate surveillance of both outcome and process indicators for reproductive health into project design and imple- mentation for several projects currently under preparation. A paper on indicators for reproductive health projects is currently in the final stages of preparation. The difficulties in measuring under-five and maternal mortality described in the report of the working group are well recognized, and the Bank endorses the efforts of WHO and UNICEF to improve the data, including the use of model-based maternal mortality estimates. United Nations Population Fund (UNFPA) Since its inception, UNFPA has encouraged and supported national efforts to formulate and implement population policies, helping developing countries to establish population planning units and has funded population analysis and research, as well as data collection activities. It has also provided support for national capacity building through training programmes at the national, regional and global levels. Establishing a common approach to national capacity building in tracking child and maternal mortality form an integral part of UNFPA's support to data collection and analysis activities. UNFPA provides support to numerous population and housing censuses. This is crucial in sub-Saharan Africa, where such support has enabled newly independent countries to undertake their first modern population censuses. Additionally, the Fund supported demographic surveys, such as the World Fertility Survey (WFS) programme, and more recently the PAPCHILD surveys undertaken in the Arab States. UNFPA's future strategy with regards to a common approach to national capacity building will maintain its emphasis on strengthening national data systems and analytical capabilities to provide timely and relevant information for policy formulation, programme development and monitoring, including support to intersectoral and inter-disciplinary efforts to streamline existing national and international approaches to generate and disseminate data. Special emphasis will be given to the development of innovative methodologies to generate, disseminate and use data in population and related areas. UNFPA will support the development of data systems that generate information that is disaggregated by gender as well as by geographic areas. UNFPA has, post ICPD, undertaken a number of initiatives directed at the improvement of monitoring reproductive health and family planning activities. The Fund is currently undertaking a pilot project aimed at establishing the feasibility of a system for the global monitoring of key indicators of family planning and reproductive health programmes. UNFPA is also spearheading an international initiative, with the active participation of the United Nations and bilateral agencies and organisations, to help establish comprehensive national and international data bases on reproductive health and family planning, inter alia, to facilitate the assessment of needs and the development of indicators including those measures agreed upon to track child and maternal mortality. ------------------------------------------------------------------- United Nations Children's Fund (UNICEF) In response to the ICPD Programme of Action recommendations, UNICEF will build on its on-going work, with other United Nations agencies, in assisting countries to strengthen their capacity to monitor progress toward the World Summit for Children goals and thus better address the ICPD goals. Most recently, UNICEF, in collabora-tion with WHO, UNFPA, UNESCO, the UN Statistical Division and regional centres of excellence, has been helping countries to build a statistical base for reporting progress towards specific goals at mid-decade. A key objective in monitoring has been to bring together the users and producers of data and to ensure that policy makers have access to understandable and current information to make decisions for programme and policy formulation, development and implementation. This has been a particular concern of UNICEF in tracking child mortality levels. In tracking maternal mortality particular emphasis is being placed on process indicators (i.e., indicators which monitor the processes which are known to reduce maternal mortality, including indicators which measure improvements in access to, utilization of and the quality of Emergency Obstetric Care services). UNICEF has issued a set of guidelines for monitoring progress in maternal mortality reduction which include a detailed description of the measurement issues and proposes a series of process indicators with a methodology for collecting the data needed to calculate these indicators. In addition, UNICEF, in collaboration with WHO, is in the process of developing model-based estimates of maternal mortality for those countries which have no data at all or very weak data on maternal mortality. ------------------------------------------------------------------- World Health Organization (WHO) WHO maintains global bibliographic and indicator databases on maternal mortality and associated women's health issues including coverage of maternity care, unsafe abortion, infertility, anemia in pregnancy, and fertility. The maternal mortality database comprises studies bringing together information on a country-by- country basis of all that is known about maternal mortality - the dimensions of the problem, causes and avoidable factors and the populations most at risk. This information provides the foundations upon which the regional and global estimates of maternal mortality and morbidity are made. The databases are available on diskette and have been widely distributed to countries, WHO Regional Offices, international agencies and researchers around the world. Tabulations of the indicators are reissued at regular intervals. WHO also convenes meetings of experts and produces guidelines on measurements issues, including methods for assessing maternal mortality at community level, indicators for monitoring progress towards the attainment of maternal health goals, and methodologies for measuring maternal morbidity. Guidelines on verbal autopsy for maternal deaths and conducting maternal death audits at facility level are currently in preparation. WHO is working with developing countries to improve health information systems in general and in particular to increase national capacity to gather and analyse basic information on births, deaths and cause of death. WHO's philosophy is that all data collection should be seen as a means towards an end rather than an end in itself. It is, therefore, recommended that in the context of maternal health indicators, countries focus increasingly on perfor-mance-based measures such as maternal audit, surveillance and other process measures. Such programme indicators should be useful for policy-making and be generated through data collection procedures that are useful for programme management at the level at which the data are collected.