Technical and Policy Paper No. 1
Issues in Measuring and Monitoring
Maternal Mortality
Implications for Programmes
Contents
- Expected Number of Maternal Deaths in a Population of 1 Million
Persons at Given Levels of the MMR and the CBR
- Sampling Error Associated with a Survey of 5,000 Households with an
Average of 4 Persons per Household at Given Levels of MMR and CBR in a Population
of 1 Million
- Approximate Number of Respondents According to Level of Maternal
Mortality Ratio and Desired Margin of Error
- Comparison of Direct and Indirect Sisterhood Estimates of the
Maternal Mortality Ratio, DHS Surveys in Selected African Countries, 1989-1995
- Country Ranking in Female Life Expectancy at Birth 1990/5 Compared
with WHO/UNICEF MMR and Other Indicators for Same Period, Selected Asian Countries
- Country Ranking in Female Life Expectancy at Birth 1990/5 Compared
with WHO/UNICEF MMR and Stanton and Hills Modeled MMR, Selected Asian Countries
- Maternal Mortality - ICPD Goals
- Definition, Methods of Computation and Hospital Records
- The Sisterhood Questions - Indirect Method
- The Sisterhood Questions - Direct Method
Reproductive health care entails the provision of a comprehensive range
of quality services and information. Amongst the elements of reproductive health to which
UNFPA gives significant attention are family planning, maternal care, harmful practices
and violence against women, prevention and management of reproductive tract infections and
sexually transmitted diseases (STDs), and prevention of HIV/AIDS. Despite progress in many
of these areas, maternal mortality continues to be unacceptably high in much of the
developing world and particularly so in poor countries, or areas of countries where the
prevalence of poverty is still acute. Such countries tend to be characterized by high
levels of fertility, high rates of infectious diseases and neglect for the proper care for
women during (and after) pregnancy.
Just over a decade has elapsed since the Safe Motherhood Initiative was
first launched in 1987 with the aim of reducing the high levels of maternal mortality that
were evidenced in many developing countries. During this period the concept of safe
motherhood has evolved and broadened. Thus the Programme of Action of the
1994 International Conference on Population and Development (ICPD), and subsequently the
Fourth World Conference on Women in Beijing in 1995, explicitly stated that programmes to
reduce maternal mortality must contain a number of complementary elements.
UNFPAs approach to reducing maternal mortality includes, among
other measures, assistance towards:
- expanding access to quality family planning services and information;
- prevention of abortion, management of the complications arising from unsafe abortions,
as well as post-abortion family planning and counseling;
- management of the complications of pregnancy and delivery;
- training for health personnel involved in providing assistance during pregnancy,
delivery and the postpartum period;
- advocacy for measures that lead to improvements in referral systems and services and
transport for complications arising from emergencies;
- targeted IEC campaigns to promote reproductive health, and obstetrical and other health
practices associated with maternal care; and
- research to evaluate the efficiency and effectiveness of innovative approaches to the
delivery of maternal care services.
With the heightened international interest in programmes aimed at reducing levels of
maternal mortality, it is important that the immense difficulties faced in reliably
measuring levels and changes in its incidence are widely understood. The measurement of
maternal mortality is important for assessing the types of programmes to be implemented,
as well as progress that they make in reducing its incidence. This technical report has
been prepared to help those concerned with formulating, monitoring and evaluating
programmes to reduce maternal mortality; to better understand the scale and frequency of
maternal deaths in populations; to recognize the problems of measuring maternal mortality
and to suggest indicators that can be used to monitor progress.
For many developing countries, fully reliable estimates of maternal mortality will not
become available until they have complete systems of civil registration. In the meantime,
there is not much to be gained from spending large amounts of scarce resources on
expensive methods of data collection merely to estimate broad or overall levels of
maternal mortality. UNFPA will continue to work with others in seeking to improve the
measurement of maternal mortality, but our priority will certainly be on programmes and
activities to reduce maternal mortality from its unacceptably high level in many
countries.
I would like to thank Mr Richard Leete, Senior Technical Officer, Technical and
Evaluation Division, (TED), who was responsible for the preparation of this report, as
well as his colleagues Dr Nicholas Dodd, Mr Richard Osborn, Dr Charlotte Gardiner, Dr
Laura Laski, Ms Catherine Pierce and Mr Michael Vlassoff for making helpful comments
during the course of its preparation. I would also like to thank Mr Joseph Chamie,
Director, United Nations Population Division; Mr Hermann Habermann, Director, United
Nations Statistics Division; and Professor John Cleland of the London School of Hygiene
and Tropical Medicine, for reviewing an earlier draft.
I hope that this report will be read and used by all those concerned with programmes to
reduce maternal mortality.
Sethuramiah Rao
Director, Technical and Evaluation Division
United Nations Population Fund
January 1998
The relatively high rates of maternal mortality prevailing throughout much of
the developing world are a consequence of the continued serious neglect of womens
reproductive health, particularly that of the poorest women in the least developed
countries, coupled with a legacy of ineffective programme interventions. UNFPA is fully
committed to addressing and improving reproductive health through its support for country,
regional and inter-regional programmes, in a manner that is consistent with the Programme
of Action (PoA) of the ICPD. In particular, UNFPA helps contribute towards reducing
maternal mortality through a wide range of measures that support increased access to a
comprehensive range of quality reproductive health services and information.
The ICPD PoA, in support of earlier UN-sponsored International Conferences, endorsed
the goal that developing countries should aim to halve 1990 levels of maternal mortality
by the year 2000 and set targets for levels to be reached beyond that date (Box 1).
Subsequently, UNFPAs Executive Board endorsed the use of maternal mortality as one
of seven threshold indicators relating to ICPD goals to be employed by the Fund for
resource allocation to country programmes. This heightened focus on maternal mortality has
major implications for establishing reasonably accurate levels as well as for monitoring
changes in them over time. However, maternal mortality is exceedingly difficult to
estimate with any reasonable level of confidence in countries without a complete
and reliable system of civil registration, which includes most developing countries.
Box 1 Maternal Mortality -
ICPD Goals
Countries should strive to effect significant reductions in maternal mortality by the
year 2015; a reduction in maternal mortality by one half of the 1990 levels by the year
2000 and a further one half by 2015. The realization of these goals will have different
implications for countries with different 1990 levels of maternal mortality. Countries
with intermediate levels of mortality should aim to achieve by the year 2005 a maternal
mortality ratio below 100 per 100,000 live births and by the year 2015 a maternal
mortality ratio below 60 per 100,000 live births. Countries with the highest levels of
mortality should aim to achieve by 2005 a maternal mortality ratio below 125 per 100,000
live births and by 2015 a maternal mortality ratio 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 in maternal mortality within
countries and between geographical regions, socio-economic and ethnic groups should be
narrowed (United Nations, 1995). |
Objectives
The purposes of this paper are to help those concerned with programmes
to improve reproductive health better understand: (i) the scale and frequency of maternal
deaths in populations; (ii) the advantages and disadvantages of using surveys to measure
maternal mortality; (iii) model-based estimates of maternal mortality and their
limitations; and (iv) the potential of relevant process indicators of maternal health for
programmatic purposes. The paper concludes by considering some of the implications
arising from the discussion of these issues for programmes.
Why Many Countries Cant Measure
Maternal Mortality
In developed countries, annual estimates of levels of maternal
mortality are generally made from data obtained as a by-product of the civil registration
of births and deaths. However, most developing countries do not have complete and
reliable systems of civil registration and, given the huge constraints affecting these
systems, particularly in the poorest countries, this situation is unlikely to improve
quickly. On the demand side, for example, rural people often have no incentive, and may
indeed incur costs, to register a household members death. On the supply side,
governments often lack sufficient resources to finance the cost of staffing, equipping and
effectively maintaining a network of registration offices.
The resource requirements for sustaining civil registration systems are
too large to be met by international funding agencies. Moreover, on the basis of past
experience it would appear that areal pilot registration schemes are unlikely to be
sustainable without continuous donor support. Civil registration systems tend to be
sustained only when their products have an essential role in the operation of the legal
and social welfare systems of a country.
Even in contexts where death registration is complete the cause-of-death
certification is generally not sufficiently accurate to capture all maternal deaths (Box
2) - this is the case in both developed and developing countries. It is generally
considered that more maternal deaths are assigned to deaths from non-maternal causes than vice
versa. In some cultures this may occur, for example, when the deaths are the result of
complications arising from illegal, or poorly performed, induced abortions.
In summary, in most developing countries the vital registration data
required to compute outcome indicators of maternal ill-health, such as the maternal
mortality ratio, the maternal mortality rate, or the lifetime risk of maternal death (Box
2), are generally either unavailable or insufficiently robust. As a result these countries
have to rely mainly on survey- or model-based estimates.
Box 2 Definition, Methods of
Computation and Hospital Records
What is a maternal death?
Simply stated it is a death of a woman who is pregnant, or within 6 weeks of her
pregnancy, from any cause directly or indirectly related to the pregnancy or its
management (AbouZahr and Royston, 1991). The tenth revision of the International
Classification of Diseases makes provision for including late maternal deaths occurring
between 6 weeks and one year post partum.
Five obstetric complications account for most maternal deaths, namely, post-partum
haemorrhage, sepsis, unsafe induced abortion, hypertensive disorders of pregnancy, and
obstructed labour. Some 75 per cent of maternal deaths occur in the last trimester and
first week following the end of pregnancy (Campbell and Graham, 1996).
How is a maternal mortality ratio conventionally computed?
The number of maternal deaths in a given year is divided by the number of live births
in the same year, and multiplied by 100,000. The maternal mortality ratio measures
obstetric risk, with live births approximating for all pregnancies in the denominator.
Is the maternal mortality ratio the same as the maternal mortality rate?
No. The maternal mortality rate relates the number of maternal deaths in a given year
to the estimated number of women of reproductive ages, taken as 15 to 49, at the mid-point
of the same year - it measures the risk of women dying from a maternal death.
What is the lifetime risk of a maternal death?
Each time a woman becomes pregnant she risks dying of a maternal cause. The higher the
level of maternal mortality the greater is that risk. The risk is cumulative: the more
times she becomes pregnant the chances of dying increase. Lifetime risk is generally
calculated with implicit assumptions about average levels of maternal mortality and
fertility over the course of a womans reproductive ages. For example, with a
maternal mortality ratio of 500 per 100,000 live births, the average lifetime risk of a
maternal death is 1 in 100 if a woman has an average of 2 pregnancies, increasing to 1 in
40 (2.5 per cent) if she has 5 pregnancies. Average lifetime risk conceals wide variation
with the poorest (rural) woman generally facing the highest risk.
Can hospital records be used to estimate national levels of maternal mortality?
Generally not, on account of selection biases. Hospital-based births tend to be
unrepresentative of all deliveries. Hospitals frequently tend to disproportionately cater
for high-risk women and emergency admissions. Further, in some developing countries a high
proportion of rural women do not have ready access to hospitals, which tend to be
concentrated in urban areas. However, hospital records of maternal deaths do provide a
useful starting point for retrospective studies and enquiries that may provide useful
insights about the factors that contributed to such deaths. |
2 Frequency of Maternal Deaths
Maternal deaths are statistically infrequent events even in countries
with high mortality and high fertility. Table 1 (third column) presents the results of
some straightforward calculations to show the annual number of maternal deaths that can be
expected in a population of 1 million at given levels of the maternal mortality ratio and
the crude birth rate. The figures show, for example, that in a country with a relatively
high maternal mortality ratio, at say 500 per 100,000 live births, and relatively high
fertility, with say a crude birth rate of 40 per 1000 population, there would be just 200
maternal deaths per million population over the course of a year. It is unlikely that
these 200 maternal deaths would be geographically evenly distributed among the population.
In many settings, it is highly probable that there would be a clustering of these deaths
among the poorest women, generally those in rural areas lacking access to emergency
obstetric care. The small number and geographic concentration of maternal
deaths will have important implications for the targeting and the nature of programmatic
interventions to reduce their incidence.
Table 1. Expected number of maternal deaths
in a population of 1 million persons at given levels of the MMR and the CBRa
Maternal Mortality
Ratio
(per 100,000 live births) |
Crude Birth Rate
(per 1,000 population) |
Expected Maternal
Deaths b
(per 1,000,000 population) |
1000 |
50 |
500 |
1000 |
40 |
400 |
1000 |
30 |
300 |
1000 |
20 |
200 |
500 |
50 |
250 |
500 |
40 |
200 |
500 |
30 |
150 |
500 |
20 |
100 |
250 |
50 |
125 |
250 |
40 |
100 |
250 |
30 |
75 |
250 |
20 |
50 |
100 |
50 |
50 |
100 |
40 |
40 |
100 |
30 |
30 |
100 |
20 |
20 |
Notes: a For populations greater than 1
million the expected number of maternal deaths can be obtained by multiplying by the
factor that they exceed 1. For example, with a population of 5.3 million simply multiply
the expected number of deaths associated with the given MMR and CBR (given in the third
column) by 5.3.
b Obtained as the CBR * (MMR/100). |
3 Survey-Based Estimates
Sample surveys offer an important potential means for collecting data to
estimate levels of maternal mortality in countries lacking reliable civil registration
systems. They can be used in a variety of different ways to help policymakers and
programme managers study maternal mortality.
Basic Approach
A basic survey approach for collecting information about maternal deaths
is through a retrospective enquiry about deaths in a household. Maternal deaths can be
obtained through a set of filtered questions starting with one about any death that has
occurred to a female member of the household in one, or n, years before the
reference date of the survey. In theory, the relevant questions can be readily asked in
the context of a fertility or health survey. In practice, there are a number of problems
that limit the potential of this type of approach for estimating levels of maternal
mortality. The main reasons are:
(i) the relatively small sample size of most household surveys tends to
lead to unacceptably high standard errors of estimates of levels of maternal mortality.
Using the figures in Table 1 as a starting point and extending the calculations, Table 2
illustrates the expected number of maternal deaths and the associated minimum margin of
error that would result in a probability survey of 5,000 households with an average of 4
persons per household in a population of 1 million. The figures show just how few maternal
deaths could be expected and the high margin of error at all levels of the maternal
mortality ratio. For example, if the prevailing maternal mortality ratio was 500 per
100,000 live births and the crude birth rate 40 per 1000 population the survey would be
expected to net just 4 maternal deaths. The 95 per cent confidence interval around this
figure is plus, or minus, almost 100 per cent. In other words random errors alone can
render meaningless, or very seriously distort, estimates of the maternal mortality ratio
obtained by this type of survey approach; and
(ii) non-sampling errors including: (a) respondents simply not knowing,
or wanting to say, whether a woman has died as a result of a maternal cause - early
maternal deaths or those arising from abortion complications in particular are exceedingly
difficult to capture (Shahidullah, 1995); (b) telescoping the date of events, and (c) the
general methodological and cultural difficulties of collecting information about deaths in
surveys.
The trade-off with this survey approach is between canvassing a larger
number of respondents, or increasing the span of years of recall data from a smaller
number of respondents. The latter strategy raises the sample size and improves the
stability of the estimates, but increases the possibility of non-sampling errors through
recall biases for example. In summary, this type of survey approach is generally not
considered to be a cost-effective means for estimating levels of maternal mortality,
primarily because of the very high costs involved in canvassing a sufficiently large
sample of households to obtain estimates that are not subject to very wide margins of
sampling error.
Table 2: Sampling Error Associated with a Survey of 5,000
Households with an Average of 4 Persons per Household at Given Levels of MMR and CBR in a
Population of 1 Million
MMR
(per 100,000 live births) |
CBR
(per 1,000 population) |
Expected Maternal Deaths
(per million
population) |
Expected Maternal Deaths in Sample a |
Standard Error b |
Margin of Error
(+ or -) % c |
1000 |
50 |
500 |
10 |
156 |
63 |
1000 |
40 |
400 |
8 |
140 |
70 |
1000 |
30 |
300 |
6 |
121 |
81 |
1000 |
20 |
200 |
4 |
99 |
99 |
500 |
50 |
250 |
5 |
111 |
89 |
500 |
40 |
200 |
4 |
99 |
99 |
500 |
30 |
150 |
3 |
86 |
114 |
500 |
20 |
100 |
2 |
70 |
140 |
250 |
50 |
125 |
2.5 |
78 |
125 |
250 |
40 |
100 |
2 |
70 |
140 |
250 |
30 |
75 |
1.5 |
61 |
162 |
250 |
20 |
50 |
1 |
49 |
198 |
100 |
50 |
50 |
1 |
49 |
198 |
100 |
40 |
40 |
0.8 |
44 |
221 |
100 |
30 |
30 |
0.6 |
38 |
256 |
100 |
20 |
20 |
0.4 |
31 |
313 |
Notes: a Expected maternal
deaths in sample = EMD * (20,000/1,000,000).
b Standard error = SQRT[{N (N-n)/(n-1)} p
(1-p)] where N is population; n is sample numbers of persons and p is EMD/1,000,000.
c Margin of error = (2*SE/EMD) * 100. |
Reproductive Age Mortality Surveys
(RAMOS)
RAMOS, or reproductive age mortality surveys, are generally used to
establish what proportion of deaths of women of reproductive ages were due to maternal
causes. In this type of survey, numbers of deaths of women of reproductive ages can be
obtained from a civil registration system (where cause of death information is
incomplete), hospital and clinical records, or from other sources. Assignment of
reproductive age deaths to a maternal cause is made after an in-depth consideration of
evidence from one or a combination of medical records, oral autopsies and interviews with
family members or medical personnel. From a programmatic perspective, an important
advantage of a RAMOS approach is that it can identify the underlying cause groups of
maternal deaths and their geographic clustering. This can provide a basis for further
study of the reasons why they occurred through the use rapid assessment qualitative
methods for example.
The main difficulty with the RAMOS approach is in establishing a
complete, or representative, frame of deaths of women of reproductive ages. In the absence
of such a frame any resulting estimates of maternal mortality are unlikely to be reliable
at the national, or local, levels.
Surveys Using the Indirect
Sisterhood Method
The indirect sisterhood method estimates maternal mortality from
information obtained by asking four basic questions (Box 3) of all adults in household
canvassed in socio-economic, demographic and health surveys. The approach is analogous to
the well-established Brass technique for estimating child mortality from information on
children ever born and children surviving (Brass, 1975). The sisterhood method uses the
proportion of adult sisters dying due to maternal causes to derive an estimate, based on a
set of underlying modeling assumptions, of the lifetime risk of maternal mortality - from
which the maternal mortality ratio can be readily derived (Graham et al., 1989). The
attraction of this approach is its simplicity and sampling efficiency.
Box 3. The Sisterhood
Questions - Indirect Method
Four questions are asked of all adults interviewed during the survey
(analysis is generally limited to responses from those aged 15 to 49 so as to keep the
reference period of estimates of maternal mortality to about 12 years before the date of
the survey):
How many sisters have you ever had who have reached aged 15?
How many of these sisters are alive now?
How many of these sisters are dead?
How many of these sisters died during pregnancy, childbirth, or during the 6 weeks after
the end of pregnancy?
Aggregate data are used to calculate the proportion of sisters dying due
to maternal causes. Adjustment factors are used to convert these proportions into
estimates of maternal mortality (Graham et al., 1989). |
A relative strength of the indirect sisterhood method is the smaller sample size
required to produce estimates of maternal mortality with given margins of sampling error.
This is because each respondent reports information about the life experience of all
sisters, and generally there will be an average of two respondents per household. Table 3
shows, for example, that in a country with a maternal mortality ratio of say 500 per
100,000 live births, and a high crude birth rate of around 40, sisterhood estimates with a
margin of error of 10 per cent and 20 per cent would require, respectively, replies from
13,000 and 3,200 respondents aged below 50.
Table 3: Approximate Number of Respondents According to Level
of Maternal Mortality Ratio and Desired Margin of Error
Maternal Mortality Ratio |
Reported Deaths Per 1000 Respondents |
Respondents
Needed by Margin of Error |
|
|
(+ or -) 30% |
(+ or -) 20% |
(+ or -) 10% |
|
|
r a > 43 |
r a > 97 |
r a >
385 |
750 |
45 |
1000 |
2100 |
8000 |
500 |
30 |
1500 |
3200 |
13000 |
250 |
15 |
3000 |
6400 |
25000 |
Note: a r denotes total number of deaths.
Source: Adapted from Table 2 in Hanley et al. (1996). |
However, despite its relative efficiency there are a number of technical problems that
limit the usefulness of the sisterhood method for programmatic purposes, particularly for
assessing trends in maternal mortality. These are:
(i) that the sisterhood method, using respondents aged 15-49, produces an estimate of
the maternal mortality ratio that relates on average to approximately 10-12 years prior to
the date of the survey. The fact that the data cannot be used to produce current estimates
of maternal mortality means that sisterhood based estimates will be of very limited use
for programme evaluation or for monitoring purposes. However, in some settings where
fertility has not changed significantly and hence the exposure to the risk of a
pregnancy-related death, and the provision of maternal health care services has also not
changed markedly, the reference point of estimates may not matter very much;
(ii) the use of the sisterhood method is not advised in settings where there have been
rapid changes in the level and pattern of fertility and mortality, or where fertility has
declined to low levels (WHO, 1998). Further, simulation studies suggest that sisterhood
estimates tend to be significantly more variable than direct estimates (Garenne and
Friedberg, 1997); and
(ii) there are several potential non-sampling errors that can affect sisterhood data,
some of which have been noted above in relation to surveys using the basic approach to
estimate maternal mortality. For example, there are cultural problems relating to the
meaning of sisters in some developing country settings; there are also
problems where respondents have lost touch with sisters who have migrated, for example,
and there are further problems that occur when respondents do not know why their sister
died.
Surveys Using the Direct Sisterhood Method
The Demographic Health Survey (DHS) project has extended the use of the original
sisterhood methodology through the incorporation of a sibling history module in selected
surveys (Rutenberg et al., 1990; Bicego et al., 1997) - the so-called direct
sisterhood method. Respondents are asked to provide fairly detailed information about each
of the children born to their mother, that is, all siblings (Box 4). By obtaining
information about all siblings, and not just sisters, it is considered that the reporting
of female deaths will be more complete. However, asking all the relevant questions is much
more demanding on both interviewers and interviewees, particularly when the respondent has
many siblings, than is the case with the four indirect sisterhood questions. Hence this
methodology has cost and trade-off implications. However, in theory the data yielded
provide the opportunity for computing direct and somewhat more recent estimates of
maternal mortality.
Box 4. The Sisterhood Questions - Direct Method From each respondent
How many children did your mother give birth to including yourself?
How many of these births did your mother have before you were born?
What was the name given to your oldest (next oldest) brother or sister?
About each sibling (survivorship status and age)
Is .... male or female?
Is .... still alive?
How old is ....?
About siblings who have died
In what year did .... die?
How old was .... when s/he died?
About each dead sister
Was .... pregnant when she died?
Did .... die during childbirth?
Did .... die within 2 months after end of a pregnancy/childbirth?
Was her death due to complications of pregnancy/childbirth?
How many children did .... give birth to during her lifetime? |
A detailed evaluation of the results of sibling history data collected in 14 DHS
surveys showed that a significant proportion of respondents had considerable difficulties
in the placement, or dating, of adult deaths, and that there was severe under-reporting of
maternal deaths that had occurred several years before the date of the survey (Stanton, et
al., 1997). The net result is that the DHS data cannot be used for analysing time
trends in maternal mortality, and even point estimates calculated by the direct method are
subject to very wide margins of sampling error. Further, the authors of the evaluation
advise that given the large sampling errors associated with the survey-derived measures of
maternal mortality, the sibling history module should be not be included in surveys more
frequently than once in every ten years (Stanton, et al., 1997).
Choice Between Direct and Indirect
Sisterhood Methodology
The relatively complex nature of the questions used in the direct sisterhood
approach suggests that it is unlikely to be an appropriate tool for use in poor countries
with low literacy levels. Indeed, a comparison of the maternal mortality ratio estimates
relating to a similar reference period made by the direct method with those made by the
indirect method (but not using the exact approach as in the original sisterhood method),
for a selection of African countries, shows that those made by the indirect approach tend
to be significantly higher (Table 4). While it is not to be expected that differing
methodologies will yield the same estimates, the magnitude of the differences is such as
to seriously question the reliability of the figures made by the direct approach. The less
expensive and more efficient indirect sisterhood questions should be the choice for use in
surveys in countries where adult literacy remains relatively low (a different conclusion
is reached by Garenne and Friedberg, 1997) - although the limitations outlined above need
to be kept in mind.
Table 4: Comparison of Direct and Indirect Sisterhood
Estimates of the Maternal Mortality Ratio, DHS Surveys in Selected African Countries,
1989-1995
Country |
Direct estimates of MMR
9-15 years before survey |
Indirect estimates of MMR
12 years before survey |
Central African Republic |
783 |
1205 |
Madagascar |
574 |
730 |
Malawi |
269 |
525 |
Morocco |
429 |
416 |
Namibia |
158 |
384 |
Niger |
805 |
859 |
Senegal |
377 |
462 |
Sudan |
322 |
450 |
Zimbabwe |
178 |
255 |
Note: The direct estimates of the MMR shown here are based on
maternal deaths reported by respondents relating to the period 9-15 years before the
survey. The number of reported deaths for this period is likely to be subject to large
recall errors. The indirect estimates of the MMR are based on all maternal deaths reported
by respondents that relate up to the date of the survey but on average 12 years before it.
Source: Data from Table 4.4 in Stanton et al., (1997). |
Censuses of Population
With the increasing realization of the limitations of surveys for
measuring maternal mortality, particularly the need for very large samples to obtain
reasonably reliable estimates, pressures have grown for the inclusion of maternal
mortality questions, such as the four sisterhood questions, in national decennial
population censuses. However, in general, census organizations may have to resist such
pressures because of the additional financial and opportunity costs, as well as the
perception that information of acceptable quality on the topic of mortality is unlikely to
be forthcoming from a population census in most settings. Indeed, there are strong
reservations about the extent to which adult mortality can be adequately measured from
census data, particularly with respect to the use of the sisterhood approach (United
Nations, 1998). One reason is the relative complexity of the information given the level
of training of census enumerators and the time allocated for field work. Another is a
great reluctance among some cultural groups to report retrospectively information about
family or household member deaths (Leete and Kwok, 1986). Yet another is some tendency for
households to break up following a maternal death. In summary these, and other
non-sampling error problems encountered in surveys and noted above, may well apply to an
even greater extent with census data.
4 Model-Based Estimates
Partly because of the various limitations inherent in population-based
estimates of maternal mortality, and partly because of the lack of any estimates for many
developing countries, sets of model-based country estimates have been made. The main
method has been to use regression techniques to develop estimation models based on data
for countries (developed and developing) with relatively reliable maternal mortality
estimates. These models are then used to predict maternal mortality for countries lacking
such estimates (for an alternative method, see Mari Bhat et al, 1995).
This methodology has been used in two different ways to obtain estimates
of maternal mortality ratios. First, through models that estimate the maternal mortality
ratio directly on the basis of selected independent variables (Stanton and Hill, 1994);
and second, through models that estimate the proportion of deaths of women of
reproductive ages that are maternal (PMDF) on the basis of selected independent
variables (Stanton et al., 1995). In order to obtain maternal mortality ratios with
this second approach, the predicted PMDF values are applied to estimated numbers of deaths
of women of reproductive ages in a given year, and then related to estimated numbers of
births in the same year - reproductive age deaths and births coming from different
sources, but mainly from the UN Population Divisions population estimates. With
either approach, a major challenge is to select appropriate independent, or predictor,
variables that are not highly correlated with one another, and for which comparable data
are available.
WHO/UNICEF Model-Based Estimates
In 1996 WHO and UNICEF published a set of model-based country estimates
of maternal mortality ratios for 1990 made primarily on the basis of this second approach
(WHO and UNICEF, 1996) - in eight developing countries, RAMOS estimates were available and
accepted. The model used to estimate the PMFD is built on a data set of 49 observations,
roughly equally split between developed and developing countries. Essentially, the model
predicts the PMDF from two independent variables - the general fertility rate (GFR)
and the proportion of live births assisted by a trained birth attendant (TRATT),
but excluding traditional birth attendants.
The 1990 model-based estimates of maternal mortality ratios are
described as representing orders of magnitude rather than precise figures and as being
subject to wide margins of error (WHO and UNICEF, 1996). They are difficult to evaluate
without reliable knowledge of the individual countries except, perhaps, by comparing them
with 1990 estimates of female life expectancy at birth, levels of fertility or
contraceptive prevalence rates (see Figures 1 and 2). For example, in general terms one
might expect there to be an inverse relationship between contraceptive prevalence rates
(CPR) and the level of maternal mortality, albeit with exceptions in particular countries.
In positing this relationship it is implicitly being assumed that the CPR is a proxy
indicator of the availability and accessibility of health personnel.
A further comparison of the WHO/UNICEF maternal mortality ratios with
the three previously mentioned variables for Asian countries shows several of the figures
to be counter-intuitive (Table 5). For example, the estimate for low-fertility Indonesia
at 650 per 100,000 live births is almost double that of high-fertility Pakistan (340) and
higher than for India (570); Nepal is shown to have a rate of 1500, nearly double that of
Bangladesh (850) despite similar levels of female life expectancy at birth in the two
countries. Affluent and very low fertility Republic of Korea is shown as having a level
(130) almost twice as high as the very poor Democratic Peoples Republic of Korea
(70). Similarly, Viet Nam (160) is shown to have a lower rate than Thailand (200) and much
lower than the Philippines (280). The magnitude and pattern of these differences, although
theoretically possible, are implausible - they are unlikely to be due to factors such as
ethnic and cultural diversity, or spatial differences in population distribution between
these countries.
Table 5: Country
Ranking in Female Life Expectancy at Birth 1990/5 Compared with WHO/UNICEF MMR and Other
Indicators for Same Period, Selected Asian Countries |
Country |
Female life expectancy at
birth (years) |
Maternal mortality ratio
(per 100,000 live birts) |
Total fertility rate
(per woman) |
Contraceptive prevalence
rate
(%) |
Korea, Rep. of |
74.8 |
130 |
1.6 |
79 |
Sri Lanka |
74.2 |
140 |
2.2 |
62 |
Korea, DPR |
73.9 |
70 |
2.1 |
72 |
Malaysia |
73.0 |
80 |
3.6 |
48 |
Thailand |
71.7 |
200 |
1.9 |
66 |
China |
70.5 |
95 |
1.9 |
83 |
Philippines |
68.2 |
280 |
4.0 |
40 |
Iran |
68.0 |
120 |
5.3 |
65 |
Viet Nam |
67.3 |
160 |
3.4 |
53 |
Indonesia |
64.5 |
650 |
2.9 |
50 |
Pakistan |
62.6 |
340 |
5.5 |
12 |
India |
60.6 |
570 |
3.4 |
43 |
Myanmar |
59.3 |
580 |
3.6 |
50 |
Bangladesh |
55.6 |
850 |
3.4 |
45 |
Nepal |
54.1 |
1500 |
5.4 |
23 |
Source: The figures in
columns 2 and 4 are from United Nations (1996a); those in column 3 are from WHO and
UNICEF (1996), and those in column 5 are from United Nations (1996b)
and UNFPA (1997b). |
The 1990 WHO/UNICEF estimates are sensitive to errors in the two
independent variables used in the model. Thus the independent health variable TRATT, or
proportion of births attended by trained birth attendants, can be quite different in
meaning from country to country even where the observed levels are similar, as can the
availability of, and access to, relevant back-up facilities needed in times of obstetric
emergency. Further, the independent fertility variable, the GFR, is essentially model
based for many countries, and different model age patterns of fertility could give quite
different levels of GFRs.
More importantly, even if the predicted PMDF values were error free,
the number of deaths of women of reproductive ages that are determined primarily on the
basis of model life tables, and to a lesser extent the number of live births, that are
used in the estimation procedure, appear to be insufficiently robust for use in making
reliable estimates of maternal mortality ratios. Further, the actual numbers of
reproductive age deaths and births for any given year, obtained from the UN's population estimates, can be subject to significant
changes in the biennial revisions made of them as new census and survey data become
available.
Future Directions For Model-Based
Estimates
Further efforts to model the maternal mortality ratio directly,
experimenting with different independent variables, such as the level of child mortality,
contraceptive prevalence rate, per cent of population with access to essential obstetric
care, per cent of population living in rural areas and nutritional and educational status
of women of reproductive ages, would appear to be justified. This modeling approach, using
female life expectancy at birth as the main independent variable, has been attempted by
Stanton and Hill (1994), and selections of the resulting predictions for Asian countries
from two of the models are shown in columns 4 and 5 of Table 6. The country patterns of
the predicted maternal mortality ratios seem to be more readily acceptable than that
yielded in the WHO/UNICEF set. For example, the maternal mortality ratio for Indonesia is
shown to be lower than that for Pakistan, and that for the Republic of Korea lower than
for the Democratic People's Republic of Korea -
differentials that are consistent with expectations based on other population and social
development indicators. While it is much less certain which set of figures, if any,
accurately represents the true values, the use and evaluation of both modeling approaches
would provide range estimates and would appear to be a safer statistical strategy.
Moreover, it would be useful to try to develop criteria for assessing the correctness of
predicted maternal mortality ratios, so as to avoid the real danger of accepting sets of
model-based figures that are systematically over- or under-estimated.
Table 6: Country Ranking in Female Life
Expectancy at Birth 1990/5 Compared with WHO/UNICEF MMR and Stanton and Hill's Modeled MMR a, Selected Asian Countries
Country |
Female life
expectancy at birth (years) |
WHO/UNICEF
maternal mortality ratios |
Maternal
mortality ratios b |
Model 1A |
Model 2 |
Korea, Rep. of |
74.8 |
130 |
26 |
40 |
Sri Lanka |
74.2 |
140 |
89 |
129 |
Korea, DPR |
73.9 |
70 |
37 |
57 |
Malaysia |
73.0 |
80 |
59 |
88 |
Thailand |
71.7 |
200 |
105 |
152 |
China |
70.5 |
95 |
107 |
156 |
Philippines |
68.2 |
280 |
103 |
152 |
Iran |
68.0 |
120 |
100 |
141 |
Viet Nam |
67.3 |
160 |
138 |
198 |
Indonesia |
64.5 |
650 |
221 |
318 |
Pakistan |
62.6 |
340 |
331 |
446 |
India |
60.6 |
570 |
245 |
350 |
Myanmar |
59.3 |
580 |
253 |
362 |
Bangladesh |
55.6 |
850 |
584 |
706 |
Nepal |
54.1 |
1500 |
750 |
1037 |
Notes: a All the MMR values shown in this table are
subject to large standard errors.
b Both sets of model estimates were determined largely on the basis of
female life expectancy at birth. Model 2 differs from model 1A in that the maternal
mortality ratios used in the estimation model were increased by a factor of 1.5 to take
account of assumed under-reporting.
Source: The figures in columns 2 are from United Nations
(1996a); those in column 3 are from WHO and UNICEF (1996), and those in columns 4 and 5
are from Stanton and Hill (1994). |
5 Process
Indicators of Maternal Health
It will rarely be practicable to devise projects to monitor levels
and changes in maternal mortality ratios. The targeting and assessment of programmes
designed to improve maternal health can be more readily and efficiently assessed through
the development of a selection of relevant proxy and process indicators, taking into
account local-level conditions and institutional factors. Such indicators can be used to
point to the types of problems that need to be tackled in the delivery of services to
reduce maternal mortality. However, there is no consensus yet about the most relevant
indicators that will inform about the level and changes in maternal mortality. Some
indicators that were initially thought of as suitable proxy measures of the maternal
mortality ratio, such as the perinatal mortality rate, have been shown not to be reliable
(Akalin et al., 1997). Further, while several process indicators relating to
maternal health have been proposed in various lists compiled by international
organizations and other groups, many of them have not yet been thoroughly field-tested,
evaluated and used for programmatic purposes.
UNFPA has recently proposed a comprehensive set of indicators,
including a sub-set relating to maternal health, for use in monitoring national population
and reproductive health programmes (UNFPA, 1997a). However, many of the indicators
included in the UNFPA list have also not yet been field tested, particularly in terms of
issues relating to the cost and feasibility of collecting the data necessary for their
construction, the frequency at which they should be monitored, their applicability at
different geographical levels, and which combination of indicators to use for specific
programming purposes. Work on these issues is currently in progress. Nevertheless, the
UNFPA list includes a choice of indicators about which there is an emerging international
consensus on their potential usefulness and a selection of these are summarized below -
additional details about concepts, definitions, computation and sources of data can be
found in Koblinsky et al., 1995; UNFPA, 1997a; UNICEF, WHO and UNFPA, 1997; WHO,
1997.
Indicators of unmet need for family planning
and obstetric services
(i) % of obstetric and gynaecological admittances due to
abortion complications - aims to show the extent of progress towards the goal of
minimal admittances due to induced abortion complications; and
(ii) % of delivering women who developed obstetric
complications and received emergency obstetric care (EmOC) -
aims to inform about the extent to which obstetric care is being received by those in
need, with the goal being universal access to EmOC.
Indicators of utilisation, coverage and access
(iii ) % of births attended by skilled health
personnel (excluding traditional birth attendants) - aims
to inform about the extent of progress towards the goal of universal access to, and
utilization of, intrapartum care;
(iv) % of service delivery points able
to provide basic essential obstetric care (EOC) (including parenteral antibiotics; parenteral oxytocic drugs; parenteral
sedatives for eclampsia; manual removal of placenta; and manual removal of retained
products) - aims to inform about the extent of availability of basic EOC
and progress towards the goal of universal access to basic EOC; and
(v) % of district hospitals able to provide
C-sections and blood transfusions - aims to inform about the extent of progress
towards the goal of the availability of EOC at all district hospitals, as well as the
quality of available services.
Indicators of quality of care
(vi) % of deliveries that are C-section -
aims to inform about whether EOC facilities are providing life-saving obstetric services
and the quality of care of the services provided. The proportion of C-section births
should be within the acceptable range; that is, they should account for a minimum of 5 per
cent and a maximum of 15 per cent of all births as defined by WHO standards; and
(vii) % of pregnant women attending antenatal
services who received (a) iron/folate (100 tablets), (b) tetanus immunization (two doses)
- aims to inform about extent to which pregnant women receive iron folate tablets and
tetanus immunization. These data provide important pointers for assessing the quality of
antenatal services that are being provided.
In addition to the above process indicators, obstetric complications
that lead directly to maternal deaths may be taken as outcome measures for programmatic
purposes. Information could be collected on post-partum haemorrhage and sepsis for
example, using a combination of retrospective oral autopsies and clinical and
observational records, to examine quality-of- care issues and to identify preventable
causes of death.
There is little to be gained from spending scarce programme
resources on very expensive surveys merely to try to estimate broad levels of maternal
mortality. Where no better figures are available, a cost-effective way of getting an
order-of-magnitude estimate of maternal mortality, relating to about 10-12 years in the
past, is through the inclusion of the four sisterhood questions in a suitable household
survey. Model-based figures may serve as very rough working estimates where no
population-based estimates are available. However, it is important to be aware of their
limitations and that they might provide an unstable and potentially misleading benchmark
against which to measure progress. In general, estimates of the maternal mortality ratio
derived from different sources and/or through the use of different methodologies are
unlikely to be comparable and may give misleading impressions about change, and sometimes
even in its direction.
For the various reasons outlined in the chapters above, in the majority
of developing countries it will not be possible to regularly monitor changes in maternal
mortality through the maternal mortality ratio. Further, the maternal mortality ratio is
itself a notoriously unreliable indicator for countries with small populations. Even if
there were effective mechanisms for collecting high quality data, random errors arising
from the small number of events would make interpretation difficult. Hence greater use
will have to be made of proxy and process indicators for assessing the impact of programme
support for improving maternal health.
While RAMOS studies may not always be able to provide reliable national
estimates of the level of maternal mortality, this approach has much potential for
programmatic purposes. It offers the potential to focus on the underlying cause groups of
maternal deaths and their geographic clustering in particular settings. Such results can
provide a basis for further policy-relevant study, through the use of cost-effective rapid
assessment procedures, of the reasons why they occurred and what interventions are
required.
The expected number of maternal deaths (and their spatial distribution)
in any particular country will have important implications for the type of programme
activity proposed. For example, in countries with small and medium population sizes with
maternal mortality ratios at, say, around 250, it will be extremely difficult to design
direct programmes to reduce maternal mortality to target levels without knowledge of the
spatial distribution of deaths, simply because of the small number of maternal deaths
involved. In such settings it may be best to concentrate on measures to improve access to,
and availability of, reproductive health services in the poorest and least-served areas.
Programme Interventions
The types of interventions that can be implemented to reduce levels of
maternal mortality in a given country will depend on many factors (including scale,
geographical distribution and cause) and will need to be considered within the context of
the country's overall reproductive health
framework.
In general, high levels of maternal mortality indicate weaknesses in
the coverage and quality of reproductive health services, including family planning.
Improving access to, and availability of, quality family planning information and
services, particularly among high-risk groups such as teenagers, women over age 39 and
those of high parity, helps reduce maternal mortality (Fortney, 1987). Although it has to
be recognised that many maternal deaths occur among women though to be at lower risk.
In some countries, significant proportions of mothers give birth
without the help of a trained birth attendant and are out of reach of emergency care to
deal with obstetric complications. The lack of services, both human and physical, results
in many readily avoidable maternal deaths, particularly among the rural poor. This unmet
need requires both the extension of emergency obstetric services and improvements in the
quality of delivery (Maine et al., 1997). Specifically, this requires: (i)
strengthening of referral systems; (ii) support for the training of health providers who
work at the primary health care level, involved in providing assistance during pregnancy,
delivery and the postpartum period, particularly in life-saving skills, ensuring the
availability of adequate supplies of medicines, and ensuring hygienic delivery practices;
and (iii) measures that support the availability and access, on a 24-hour basis, to fully
equipped emergency obstetric services for pregnant women with complications.
Further, targeted IEC campaigns, particularly among the poorest women,
can help to promote reproductive health and obstetrical and other health practices
associated with safe motherhood. Advocacy efforts directed towards political leaders and
policy makers emphasizing the seriousness of maternal morbidity and mortality may lead to
additional funds for the promotion of improvements in services and the training of
personnel to treat complications arising from emergency situations.
Maternal mortality is often high in settings where induced abortion is
illegal but its incidence is nevertheless fairly widespread. In countries where maternal
mortality is high because of unsafe abortions, efforts should be directed at the
prevention of abortion by meeting the unmet demand for family planning services. Further,
programmes are needed to support the management of the complications arising from unsafe
abortions, as well as post-abortion family planning and counseling.
Interventions to reduce maternal deaths generally need to be targeted
at the poorest areas and groups at highest risk, where the provision of health services
and amenities is least. They should take into consideration issues relating to coverage,
equity, quality of care, women's satisfaction
.and cost-effectiveness (Mother Care Matters, 1997). Meetings with women's groups and appropriate NGOs about how best to
develop and deliver maternal health services that are conducive to women's needs will also help ensure their suitability and
acceptance at the community level.
Against the background of heightened international and national
interest in maternal health, it is important that the immense difficulties faced in
reliably measuring levels and changes in maternal mortality in developing countries are
widely understood. Programme managers need to be aware of the main measurement problems
and at the same time be cognizant with the scale, dimensions and spatial distribution of
maternal mortality. Regular current monitoring of maternal health status through the
maternal mortality ratio, at intervals of less than 10 years, is not viable in most
developing countries. Robust and regular estimates will generally only become available in
the long run when countries have complete and reliable systems of civil registration.
For the short run it will often be necessary to rely on proxy and
process indicators for targeting interventions and assessing their impact, although the
task of compiling the relevant indicators poses many challenges. In particular, it needs
to be recognized that in many developing countries the capacity for collecting and
processing the data required for producing the proposed indicators is very weak. There
appears to be an inverse relationship between the availability of data on maternal
mortality and the magnitude of the problem. The coverage and quality of the statistical
information obtained as a by-product of service and administrative systems are highly
variable, and surveys tend to be ad hoc and donor driven. A new strategy for the
development of sustainable integrated population and health information systems, including
the role of technical assistance, will have to emerge to give meaning to the indicator
movement.
The efforts and attention of programme managers should be focused on
sub-populations in greatest need of reproductive health services, particularly in poor
(rural) areas where health personnel and obstetric care facilities are most lacking.
International and national attempts to improve the measurement of maternal mortality
should continue. Maternal mortality is too high in most developing countries, and
this needs to be repeatedly emphasized to show that much remains to be done to reduce it
to acceptable levels.
AbouZahr, C., and E. Royston. (1991). Maternal Mortality:
A Global Factbook. Geneva: World Health Organization.
Akalin, M. Z., D. Maine, A. de Francisco and R. Vaughan. (1997).
"Why Perinatal Mortality is Not a Proxy for
Maternal Mortality", Studies in Family
Planning, Vol. 28, No. 4.
Bicego, G., S. Curtis, H. Raggers, S. Kapiga and S. Ngallaba.
(1997). Sumve Survey on Adult and Childhood Mortality Tanzania, 1995. Macro
International Inc: Calverton, Maryland.
Brass, W. (1975). Methods for Estimating Fertility and
Mortality from Limited and Defective Data, Chapel Hill, North Carolina: Laboratories
for Population Statistics.
Campbell, D. M. R., and W. J. Graham. (1996). "Measuring Maternal Mortality: Data Sources for
Direct and Indirect Approaches", Paper
presented to the IUSSP Seminar on Innovative Approaches to the Assessment of Reproductive
Health. Manila, September 24-27, 1996.
Fortney, J. A. (1987). "The
Impact of Family Planning in Reducing Maternal Mortality",
Studies in Family Planning, Vol. 18, No.2.
Garenne, M., and F. Friedberg. (1997). "Accuracy of Indirect Estimates of Maternal
Mortality", Studies in Family Planning,
Vol. 28, No.2.
Graham, W., W. Brass, and R. W. Snow. (1989). "Estimating Maternal Mortality: The Sisterhood
Method", Studies in Family Planning,
Vol. 20, No. 3.
Hanley, A. J., Hagen, A. C., and T. Shiferaw. (1996). "Confidence Intervals and Sample-Size
Calculations for the Sisterhood Method of Estimating Maternal Mortality", Studies in Family Planning, Vol. 27,
No. 4.
Koblinsky, M., K. McLaurin, P. Russell-Brown and P. Gorbach.
(1995). Indicators for Reproductive Health Program Evaluation: Final Report of the
Subcommittee on Safe Pregnancy, Chapel Hill: University of North Carolina.
Leete, R., and K. K. Kwok. (1986). "Demographic Changes in East Malaysia and Their
Relationship with Those in the Peninsula 1960-1980", Population Studies, Vol.
40, No. 1.
Maine, D., M. Z. Akalin, V. M. Ward, and A. Kamara. (1997). The
Design and Evaluation of Maternal Mortality Programs, School of Public Health,
Columbia University.
Mari Bhat, P. N., K. Naveneetham and S. I. Rajan. (1995). "Maternal Mortality in India: Estimates from a
Regression Model", Studies in Family
Planning, Vol. 26, No.4.
Mother Care Matters. (1997). "A
Tribute to José Luis Bobadilla", Mother
Care Matters, Vol. 6, No.2.
Rutenberg, N., T. Boerma, J. Sullivan, and T. Croft. (1990). "Direct and Indirect Estimates of Maternal
Mortality with Data on the Survivorship of Sisters: Results from the Bolivia DHS". Paper presented at the Population Association
of America, Toronto, Ontario, May 3-5, 1990.
Shahidullah, M. (1995). "The
Sisterhood Method of Estimating Maternal Mortality: The Matlab Experience", Studies in Family Planning, Vol. 26,
No. 2.
Stanton, C., and K. Hill. (1994). Model-Based Estimates of
Maternal Mortality, (unpublished mimeographed report submitted to UNICEF/WHO).
Stanton, C., K. Hill, C. AbouZahr, and T. Wardlaw. (1995). Modeling
Maternal Mortality in the Developing World, (unpublished WHO/UNICEF mimeograph).
Stanton, C., N. Abderrahim, and K. Hill. (1997). DHS Maternal
Mortality Indicators: An Assessment of Data Quality and Implications for Data Use, DHS
Analytical Reports No. 4. Calverton, Maryland.
United Nations. (1995). Report of the International
Conference on Population and Development, Cairo, 5-13 September 1994. New York: United
Nations Publications.
United Nations, Department for Economic and Social Information
and Policy Analysis. (1995). World Population Prospects: The 1994 Revision, New
York: United Nations.
______. (1996a). World Population Prospects: The 1994
Revision, New York: United Nations.
______. (1996b). Levels and Trends of Contraceptive Use as
Assessed in 1994. New York: United Nations.
______. (1998 forthcoming). Principles and
Recommendations for Population and Housing Censuses. New York: United Nations.
United Nations Population Fund. (1997a). Indicators for
Population and Reproductive Health Programmes. New York: UNFPA.
______. (1997b). The State of World Population 1997. New
York: UNFPA.
UNICEF, WHO and UNFPA. (1997). Guidelines for Monitoring the
Availability and Use of Obstetric Services. New York: UNICEF.
WHO and UNICEF, (1996). Revised 1990 Estimates of Maternal
Mortality: A New Approach by WHO and UNICEF. Geneva.
World Health Organization. (1997). Monitoring Reproductive
Health: Selecting a Short List of National and Global Indicators. Geneva, WHO.
______. (1998). The Sisterhood Method for Estimating Maternal
Mortality: Guidance Notes for Potential Users. Geneva: WHO.
© 1998 United Nations Population Fund |