Best Practice & Research Clinical Obstetrics and GynaecologyVol. 22, No. 3, pp. 425–445, 2008

doi:10.1016/j.bpobgyn.2007.12.001

available online at http://www.sciencedirect.com

1

Measuring progress in reducing

maternal mortality

Wendy J. Graham* DPhil Oxon

Professor of Obstetric Epidemiology

Department of Obstetrics and Gynaecology, University of Aberdeen, UK

Immpact, University of Aberdeen, UK

Lauren B. Foster MA, MSc

Research Assistant

Immpact, University of Aberdeen, UK

Lisa Davidson MA Hons

Research Assistant

Immpact, University of Aberdeen, UK

Elizabeth Hauke MBBS, BSc

Research Assistant

Department of Epidemiology and Population Health, London School of Hygiene

and Tropical Medicine, London, United Kingdom

Oona M.R. Campbell PhD

Professor in Epidemiology and Reproductive Health

Department of Epidemiology and Population Health, London School of Hygiene

and Tropical Medicine, London, UK

The need to monitor progress in reducing maternal mortality has a long history, which can betraced back to the 1700s in some parts of the Western world. Today, however, this need is feltmost acutely in developing countries, where the priority is to stimulate, evaluate and sustain ac-tion to prevent these essentially avoidable deaths. Over the last two decades, considerable ef-forts have been made to understand and overcome the measurement challenges of maternal

* Corresponding author. Health Sciences Building, University of Aberdeen, Foresterhill, Aberdeen AB25

2ZD, Scotland, United Kingdom. Tel.: þ44 (0) 1224 553924; Fax: þ44 (0) 1224 555704.

E-mail address: w.graham@abdn.ac.uk (W.J. Graham)

1521-6934/$ - see front matter ª 2007 Elsevier Ltd. All rights reserved.

426 W. J. Graham et al

mortality in the context of weak information systems, and new and enhanced methods and toolshave emerged.

Key words: developing countries; maternal deaths; maternal mortality; measurement; methods;progress; tools.

INTRODUCTION

Today, the desire to show progress in reducing the burden of mortality is universal:across developing and developed countries; at international, national and local levels;and for all causes and conditions. What is also universal is the challenge this pres-ents. Capturing deaths and assigning causes is not straightforward, and this alonemakes showing changes – increases or decreases – problematic.1 Two-thirds ofthe world’s population reside where routine registration of deaths is missing, andfor the remaining third, misclassification of particular causes or circumstances cansometimes lead to spurious trends and conclusions about progress.2 It is importantto acknowledge this wider reality of mortality measurement even when specific sub-groups are the focus of attention. In this chapter, the focus is on maternal mortalityand a reminder of the wider challenge is particularly relevant. Measuring maternalmortality is often regarded as one of the most problematic outcomes to track,especially in developing countries.3 There certainly are significant challenges, butalso many distinctive characteristics to maternal deaths, which aid reporting andmeasurement, as illustrated later. The aim of our paper is to introduce the variousapproaches that are possible in different country settings, and so illustrate the com-parative wealth of opportunities for measuring maternal mortality. Why is this illus-tration important now?

2007 marked the twentieth anniversary year of the international Safe MotherhoodInitiative, which was launched to galvanize action to reduce maternal mortality.4 2007/08 also marks the mid-point to achieving the Millennium Development Goals (MDG),one of which – MDG5 – will be judged on the basis of a 75% reduction in maternalmortality by 2015.5 Together, these events have shone a spotlight on the weaknessesof existing statistics, nationally and thus internationally, on the magnitude and trends inmaternal mortality. Ironically, the weaknesses are often discussed together with appar-ent conclusions that no progress has been made and that MDG5 is ‘off-track’.6 Con-fusing ‘no progress’ with ‘no measurement of progress’ is the fundamental problemhere, as is confusing ‘no options for measurement’ and ‘no resources to do so’.Such confusions have repercussions at local, national and international levels, leadingin the extreme to disillusionment and disinvestment in actions to reduce these deaths,because ‘what you count is what you do’.7 Currently, there are indeed no standardizedmethods and sources for measuring maternal mortality that can be universally appliedand are universally reliable for the purposes of international monitoring of maternalmortality. This does not, however, mean that at subnational and national levels thereare no options for measurement. There is currently a high demand and need withincountries to measure maternal mortality to stimulate, evaluate and sustain action toprevent these essentially avoidable deaths.

This chapter synthesizes the published literature and draws on key resource mate-rials (Box 1) to present the current range of measurement options, and to highlight thescope for further improvement and innovation. The definitions and common

Box 1. Literature search strategy and key resources

Published literature on the measurement of maternal mortality in developingcountries was sought from the electronic databases MEDLINE, EMBASE,CINAHL, and POPLINE. These databases were searched for the years 1980–2007, using search terms: developing countries, maternal deaths, maternal mor-tality, measurement, progress, methods, and tools.In addition, a number of web-based resources were used, in particular the newlycreated Maternal Mortality Measurement Resource (http://www.maternal-mortality-measurement.org), developed by the authors of this chapter. Otherwebsites of particular relevance include:

1. International Classification of Disease and Injuries. ICD-10 Online: http://www.who.int/classifications/apps/icd/icd10online/

2. Health Metrics Network: https://www.who.int/healthmetrics3. Initiative for Maternal Mortality Programme Assessment (Immpact): http://

www.immpact-international.org4. MEASURE Evaluation: http://www.cpc.unc.edu/measure/5. United States Centers for Disease Control: http://www.cdc.gov/mmwr/

preview/mmwrhtml6. WHO Reproductive Health & Research: http://www.who.int/reproductive-

health7. United Nations Statistics Division: http://unstats.un.org/unsd

Measuring progress in reducing maternal mortality 427

indicators for maternal mortality are introduced. The main reasons for, and history of,measurement is then discussed. Next, the challenges to measurement are outlined,followed by the specific issues raised by tracking progress. An overview is then givenof the measurement approaches for finding deaths and for categorizing them as mater-nal. The final section proposes priority areas for research and development.

WHAT ARE THE KEY DEFINITIONS AND INDICATORSFOR MATERNAL MORTALITY?

Definitions

There are two main bases for defining maternal death: cause of death and time ofdeath relative to pregnancy status.8 These have direct implications for the choiceand suitability of alternative data sources and methods for measuring maternal mortal-ity, such as civil registration, routine health information systems or population-basedhousehold surveys. For example, in many developing countries only deaths occurringin health facilities are assigned a medical cause, as certification of deaths at home rarelyhappens and relatives’ reports of signs and symptoms might not be sufficiently preciseor reliable to draw conclusions. For deaths at home, verbal autopsy techniques9 arefrequently used to gauge likely causes, as discussed later.

The International Classification of Diseases (ICD) is the international standardfor diagnostic classification.10 It was developed by, and is regularly updated by, theWorld Health Organization (WHO) and is currently in its tenth version (ICD-10).

428 W. J. Graham et al

The ICD provides a comprehensive coding system for diseases and health conditionsto be used on death certificates, hospital and other vital records. ICD-10 definesa maternal death as:

.the death of a woman while pregnant or within 42 days of termination of preg-nancy, irrespective of the duration and site of the pregnancy, from any causerelated to or aggravated by the pregnancy or its management but not fromaccidental or incidental causes.

This definition requires cause-of-death information so that incidental causes can beexcluded and maternal deaths can be subdivided into two groups: direct obstetricdeaths and indirect obstetric deaths. A direct obstetric death is defined as one:

.resulting from obstetric complications of the pregnant state (pregnancy,labour and the puerperium), from interventions, omissions, incorrect treatment,or from a chain of events resulting from any of the above.

An indirect obstetric death is defined as one:

.resulting from previous existing disease or disease that developed duringpregnancy and which was not due to direct obstetric causes, but was aggravatedby physiologic effects of pregnancy.

ICD-10 differs from earlier versions11 in that it also provides a definition for latematernal death:

.the death of a woman from direct or indirect obstetric causes more than 42days but less than one year after termination of pregnancy.

In view of the difficulty of identifying maternal deaths on the basis of medical causes,there has long been reliance on using the pregnant or parturient status of the deceasedwoman as a proxy for cause. Events captured by this so-called time-of-death definitionare now referred to as pregnancy-related deaths. A pregnancy-related death is defined as:

.the death of a woman while pregnant or within 42 days of termination of preg-nancy, irrespective of the cause of death.

As can be seen by comparison with the ICD definition given earlier, using a time-based approach means that deaths owing to accidental or incidental causes are also in-cluded. Although there is limited evidence from developing countries on the size of theoverestimation of maternal mortality by including these additional causes, it is generallyregarded as insignificant and balanced-out in many settings by the underestimation dueto omitted maternal deaths from sensitive causes, such as induced abortion or early inthe first trimester when the pregnancy is unknown to relatives or the woman. Onestudy in Bangladesh, for example, cites a figure of 15% more pregnancy-related com-pared to maternal deaths12, and most of the published literature tends to use the termsmaternal and pregnancy-related deaths synonymously. The principal advantage of thelatter definition in settings where the primary source of information on deaths is rela-tives’ reports is the avoidance of questions on signs and symptoms and thus proxy as-certainment of cause. Some methods for estimating maternal mortality, such as thesisterhood method13, rely on time-of-death reporting; and others use both cause andtime definitions, such as Sample Registration of Vital Events with Verbal Autopsy(SAVVY; this is discussed in more detail later).14 Misclassification of medical cause iscommon for many categories of death, including in developed countries with advancedstatistical systems15 as well as in developing countries.16

Measuring progress in reducing maternal mortality 429

Indicators

Five main measures or indicators are used for maternal mortality. These are shown,together with their mathematical equivalences in Box 2.

WHY MEASURE MATERNAL MORTALITY?

Information on maternal mortality is needed for a wide range of purposes and at local,national and international levels.17 These purposes can be broadly grouped into four:

Box 2. Main measures of maternal mortality

1. Maternal mortality ratio (MMR): number of maternal deaths in a period pernumber live births during same period:

Number of maternal deaths

Number of live births� 100;000

MMRatio¼MMRate=General fertility rate

MMRatio¼ 1� ð1�LTRÞ1=TFR

LTR, lifetime risk; TFR, total fertility rate.2. Maternal mortality rate (MMRate): number of maternal deaths in a period per

number of women of reproductive age during the same period:

Number of maternal deaths

Number of women aged 15-49� 100;000

MMRate¼MMRatio�General fertility rate

MMRate¼ 1� ð1�LTRÞ1=35

3. Lifetime risk (LTR): probability of a woman dying from maternal causes overcourse of her reproductive life span. Can be expressed as:

LTR ¼ 1� ð1�MMrateÞ35

1� ð1�MMratioÞTFR

Sometimes approximated as 35�MMRate

4. Proportion maternal among deaths of women of reproductive age (PMDF): re-flects contribution of maternal deaths to overall mortality among women ofreproductive age

PMDF¼ Number of maternal deaths in a period

Number of deaths among women 15-49 in same period

5. Case fatality rate: reflects the number of women who die from a specific com-plication among all pregnant or recently delivered women who experience thatcomplication.

430 W. J. Graham et al

1. To establish levels, trends and differentials in maternal mortality.2. To identify characteristics and determinants of maternal deaths.3. To monitor and evaluate the effectiveness of activities designed to reduce maternal

mortality.4. To monitor progress towards international development targets, such as MDG5.

However, in reality these purposes overlap as all measurement ultimately aims toreduce the problem and is not an end in itself. The primary intended purpose affectsthe required scope and accuracy of the information, and so also the suitability of alter-native measurement options. For instance, if the need is for local-level data to improvethe quality of care at district hospitals, then it is less important to know the MMR orMMRate (see Box 2) than to be able to identify preventable death cases. The lattercould be achieved through, for example, an audit of medical records and interviewswith staff and relatives, so providing rich and relevant detail going beyond the occur-rence of death to include determinants and circumstances.18 Alternatively, if informa-tion is needed on the trends in maternal mortality for a country to monitordevelopment targets, then the MMR, reported annually through a complete civil reg-istration system for, say, the last 5 years would be a suitable source, if available. In thischapter, we focus particularly on the need for information to show progress in reduc-ing maternal mortality at national and major subnational scales, and thus on purposes 1and 4 as mentioned above.

HISTORY OF MEASURING MATERNAL MORTALITY

The history of measuring maternal mortality – ‘deaths in childbed’ – goes back overthree centuries in some developed countries.19 Here, the measurement of mortalityusually evolved with the overall civil registration system, moving from being primarilya means of ensuring legal transference of inheritance rights at death to being used forstatistical purposes. However, in some settings, maternal deaths were initiallyrecorded as part of the maternity service, with healthcare providers primarily respon-sible for reporting cases.20 As experience of measuring maternal mortality increased,so did reliance on multiple data sources and capture mechanisms. This evolution waspartly as a consequence of the increasing rarity of maternal deaths in now-developedcountries and so the need for intensified surveillance. It also reflected a realization thatall measurement approaches have advantages and disadvantages, both in terms ofcomplete ascertainment of maternal deaths and provision of additional informationfor programmatic purposes. Today, in many countries with advanced statistical sys-tems, maternal deaths are still identified from multiple sources, as for example inthe UK Confidential Enquiries into Maternal Deaths (CEMD)21, and with additionalspecial efforts, such as inclusion of a pregnancy-status check-box on death certificatesas well as active surveillance in the US.22

In developing countries, heightened interest in maternal mortality coincided witha series of subnational studies in the 1980s, described in a WHO report that revealeda higher than expected frequency of maternal deaths.23 These dedicated studies alsohighlighted the serious underreporting in routine statistics and gave early insights intothe challenges of capturing maternal deaths, particularly where the vast majority occurat home without contact with the health system. At this time, the options for measur-ing maternal mortality at a population level in developing countries were essentiallylimited to those used for all adult mortality – via incomplete vital registrationor large-scale population surveys. Alternatively, deaths to women of reproductive

Measuring progress in reducing maternal mortality 431

age were adjusted by the proportion estimated to be due to maternal causes(PMDF) – often assumed to be 25–33%.17 Since then, the measurement opportunitiesfor maternal mortality have increased markedly, as outlined later. Some reflect widerdevelopments in information-gathering in developing countries, such as the Demo-graphic and Health Surveys24, or capacity-strengthening initiatives such as the HealthMetrics Network (see Box 1). Other opportunities reflect measurement advances, forinstance the development of the sisterhood method13, the use of non-probability sam-pling or sampling at service sites (SSS)25, the addition of questions on pregnancy-related deaths to the decennial census26, the use of analytical techniques to adjustincomplete data, such as capture–recapture27, or of statistical regression models topredict levels in the absence of empirical data.28 The crucial point is that for most sit-uations and purposes there is now an opportunity and a method suitable for generat-ing estimates of maternal mortality, provided resources are available. The diversity ofoptions is partly a consequence of the diversity of measurement contexts in develop-ing countries, and partly a response to the challenges of capturing maternal deaths, asnow to be discussed.

WHAT ARE THE CHALLENGES IN MEASURINGMATERNAL MORTALITY?

Over the last 25 years, considerable experience has accumulated in understanding andgrappling with the challenges of measuring maternal mortality in the context of weakor non-existent routine reporting of deaths. This experience has been widely publi-cized7,17,29–31, giving a higher level of awareness of the problems than exists in manyother areas of mortality estimation.32 Broadly speaking, the challenges fall into two in-terrelated categories: problems with meeting the definition of maternal mortality, andproblems with finding the deaths.

The definitional challenge

The two main bases for defining maternal deaths were introduced earlier as medicalcause and time of death. Both of these are prone to errors of reporting or omission,be these on death certificates or from interviews with relatives. The ICD-10 definitionrequires that deaths from accidental and incidental causes should be excluded, and thisis obviously not possible if cause is unreported. This definition also encompassesdeaths irrespective of the duration and site of the pregnancy and, as noted earlier,events in early pregnancy related to, for example, ectopic pregnancy or unsafe inducedabortion are particularly prone to underreporting by relatives and might also bemissed in health records. In the absence of routine autopsy, many deaths in developingcountries remain recorded as ‘unknown’. Both cause and time-of-death definitions arealso challenged by the cut-off at 42 days postpartum, which might be hard for relativesto recall, and so might also be omitted from routine health records, especially wherethe case is related to indirect obstetric causes, such as tuberculosis (TB) or HIV/AIDS.33 Various measurement methods and tools have been devised to address mis-reporting or omission of cause or time of death (these are discussed later). None ofthese, however, can totally eliminate the definitional challenge. Finally, it is also impor-tant to highlight the distinct features of maternal death that make ascertainment easierthan for other mortality subgroups. First, in many cultures pregnancy and childbirthare highly memorable events, and deaths in otherwise healthy women at this timeare particularly striking. Second, special studies have shown that deaths cluster around

432 W. J. Graham et al

labour, delivery and the 24 hours immediately postpartum, hence when the womanwould have been visibly pregnant.34

The challenge of finding deaths

Clearly, before any definition can be applied, deaths that are potentially maternal orpregnancy-related must be found. The wider group of deaths to which maternal orpregnancy-related events belong are those to women of reproductive age (WRA).The latter are usually defined as 15–44 years or 15–49 years, although in some situ-ations where early age at first birth is common, WRA are regarded as those aged 10–44 years or 10–49 years. In terms of identifying WRA deaths, there are active, passiveand mixed approaches. Active identification is where the enumeration process goesout to find cases in the population through surveys, a census or surveillance, andcan be further subdivided on the basis of whether the death is in respondents’own households or among their sisters. Passive approaches are where the deathsare already captured by an existing system, such as with civil registration or healthfacility statistics, and the maternal or pregnancy-related fraction then needs to beextracted. Mixed identification involves both active and passive identification ofdeaths, as with approaches collectively know as Reproductive Age Mortality Studies(RAMOS).35

The difference between active and passive approaches to finding deaths is impor-tant for two main groups of reasons: completeness and reliability, and feasibility andprecision. Passive approaches are reliant on deaths being notified to a point of datacapture, such as a civil registration office, and thus are prone to incompleteness(bias of omission), particularly where deaths occur in the community rather than inhealth facilities. The proportion of maternal deaths occurring at home variesenormously, with studies reporting figures from 2 to 56%.34 The other bias in passivesystems comes from selectivity, in other words, where deaths to particular subgroupsof women, such as the poorest, or from certain causes, such as haemorrhage, areunderrepresented.

Active approaches seek out deaths and, with appropriate attention to the size andcoverage of the data capture, can yield more complete and representative estimatesand patterns of maternal mortality. However, to find deaths in a population involvesintensive field efforts and, given the comparative rarity of maternal deaths at a popula-tion level and over short periods of time (say, less than 2 years), require large samplesizes or complete enumeration. Active approaches thus have disadvantages in terms offeasibility and cost compared to passive approaches, which rely upon existing report-ing systems. Moreover, where the coverage of the population is insufficient owing, forexample, to inadequate sample size, then imprecise estimates can result. If the sampledesign is also inappropriate, there might also be bias owing to selection. Much hasbeen written about sample size requirements for measuring maternal mortality, andseveral useful reference resources exist (see Box 1).36,37 One compromise oftenmade because of the large sample sizes needed to produce recent estimates is touse long reference periods, such as 10 years. Such a compromise is problematic fortwo main reasons, one related to increasing recall errors when relatives are askedto think back over such long periods, and the other where time trends are soughtfor programmatic planning purposes, since 10 years is often too large an interval tobe useful.38

Having described the overall challenges to measuring maternal mortality, we willnow look at these in the context of the specific need to demonstrate progress.

Measuring progress in reducing maternal mortality 433

WHAT DO WE MEAN BY PROGRESS?

Defining progress might seem obvious, implying at its simplest reduced burden orincreased health gain. However, progress is a multidimensional concept39 and can bediscussed on four main, interrelated bases: the type of outcome in which progressis measured, the scale or scope of the population of interest, the criteria and metricused for judging progress, and the reliability and attribution of apparent progress.

Choice of outcome

In the case of progress in reducing maternal mortality, the direct outcome of interestis maternal death. However, as this is a composite of deaths, change might be sought inspecific subgroups defined on the basis of causes, such as direct obstetric, or time-of-death (e.g. during labour or delivery), or for particular groups of women, such asyoung adults. For example, progress in a population in reducing maternal mortalitymight be distorted by new or re-emergent diseases, such as HIV/AIDS or TB, maskingdeclines in specific direct obstetric causes, such as eclampsia.40 Moreover, instead oflooking for progress in maternal mortality, proxy indicators might be used, whichare causally related or associated with the risk of maternal complications and/or death.For example, progress might be measured in terms of a reduced number or incidenceof near miss cases (specific instances of life-threatening complications as describedelsewhere in this volume), as an indication of increased deaths averted.41 Alternatively,markers of the coverage of services relevant to preventing maternal deaths might betracked as proxies of progress, such as indicators of emergency caesarean sections or‘met need for obstetric’ care.42 These so-called process indicators43 have been pro-moted over the last 20 years not only as useful markers of the need for interventionor of progress in meeting the need, but also as proxies for monitoring maternal mor-tality.31,39 Indeed, although the main target outcome for measuring progress towardsMDG5 is a 75% reduction in the MMR, the proxy indicator also being tracked atnational and international levels is the percentage of deliveries with skilled attendants,5

and there are now several other international initiatives monitoring coverage indica-tors, such as the Countdown to 2015.44 Most process or coverage indicators alsopresent measurement challenges, and although their sample-size requirements aregenerally much less onerous than for maternal mortality, they should be seen as com-plementary rather than substitutes where maternal mortality is the target outcome ofinterest.45

Scale or scope of progress

Progress might need to be judged at a variety of levels, with major implications formeasurement. Whereas, for example, MDG5 is tracked at national, regional and globallevels, within countries decision-makers might also want to judge progress in reducingdeaths at a very local level of a particular subdistrict or for a catchment population ofa specific hospital. These two types of scenarios naturally have different options andrequirements for acquiring data, in the former case often relying on readily availablestatistics or modelling estimates where no empirical sources exist. Such an approachhas been used since 1995 to arrive at national and global estimates based on a regres-sion model, and figures have just been released for 2005.28

Progress at a macro-level can mask major differences within and between worldregions, as well as at a country level. For instance, in terms of the poor–rich gap in

434 W. J. Graham et al

maternal mortality, differences of the order of 4- to 6-fold have recently been revealedusing an analytical approach called the familial technique.34,46 Moreover, at the macro-level of international comparisons, the source and the reliability of estimates onmaternal mortality vary enormously, presenting major challenges for interpretation.31

There are also problems at subnational, and especially local levels, such as the costs ofapplying an active approach to data capture in the absence of useable passive sources.Moreover, at local levels, the size of the population and thus the number of deathsinfluences both the effort needed to identify events and the stability and reliabilityof estimates, where, for example, less than 5 deaths occur per year and thus justone less death constitutes a 20% decline.

Criteria and metric used

Progress can be judged in absolute terms, as with changes in numbers of deaths, orin the various measures of maternal mortality mentioned in Box 2. Alternatively,progress is frequently considered against a target, as with MDG5, where the targetis a 75% decline in the MMR. In measuring progress or change there is necessarilya temporal dimension, although this is not necessarily confined to one area or pop-ulation but is rather tracked by comparisons between groups, such as the reductionover time in the poor–rich differential in maternal mortality. Whatever the compar-ative basis, monitoring implies measurement more than once, and thus a repeatedtackling of the challenges mentioned earlier in the chapter. The judgement ofwhether progress is adequate may be subjective, as with, for example, a situationwhere one less death a year is regarded as success at a local level, or in relationto the rate of change needed to achieve a longer-term goal.28 Using absolute num-bers of deaths is problematic above a local level or across long time intervals, asother factors can confound the measurement of change, such as in- or out-migrationor fertility trends that affect the population at risk.38 Generally speaking, rates, ratiosor proportions are thus preferable metrics for showing change, and this introducesthe fourth basis for discussing progress – the extent to which changes are real andattributable.

Reliability and attribution of progress

The challenges in measuring maternal mortality described earlier – of defining andidentifying deaths – mean there is always uncertainty in the estimates for any singlepoint in time or comparative group. Such uncertainty is even more problematicwhen changes are being monitored. For example, weaknesses in a baseline assessmentof the MMR for a particular country could mask or exaggerate the progress apparentfrom a later and more reliable assessment. Moreover, where the sources of estimatesare sample surveys, then uncertainty also arises from the size of the sample, as re-flected in the confidence intervals.37 Overlapping confidence intervals imply that anydifference in the central estimates might be an artefact rather than real.29 Althoughwith complete enumeration of deaths using active approaches like the Census andcomplete surveillance do not, by definition, have sampling errors, non-sampling errorsmight be significant and necessitate adjustments to the estimates.26 A number of de-mographic techniques exist for assessing and adjusting mortality data from deficientsources.47,48

Finally, it is important to distinguish between tracking progress with or withoutattribution of the change, as this has significant implications for measurement.

Measuring progress in reducing maternal mortality 435

Assigning progress in mortality reduction to a specific programme or intervention ismore complicated than showing a change per se, owing to confounding or interact-ing factors.38 For reliable attribution, experimental study designs are regarded asthe gold standard. Unfortunately, these are rarely feasible for major complex pro-grammes made up of composites of interventions delivered to communities ratherthan individuals and in real-world contexts in which other interventions andchanges cannot be excluded or monitored adequately.49 More realistic designsfor assessing progress associated to, but not necessarily causally related to, declinesin maternal mortality are primarily descriptive, ideally with the opportunity for be-fore-during-and-after comparisons. Such designs should be underpinned by a soundconceptual framework that enables any observed change in mortality to be relatedto the intervention programme or strategy on the basis of plausibility.50 In contextsin which large-scale plausibility assessments are conducted, not only should mater-nal mortality be tracked, but also the inputs, outputs and processes, so building upa picture of how any change in outcome could have occurred. Today there are stillvery few examples of such composite evaluations51, in part owing to the resourcesrequired, and most historical and more recent case studies of maternal mortalitydecline are unable to definitively attribute progress to specific programmes orinitiatives.52,53

WHAT ARE THE OPTIONS FOR MEASURINGMATERNAL MORTALITY?

This section provides an introduction to the principal methods, techniques andapproaches to measuring maternal mortality. These are best grouped according toa hierarchy of considerations that are more easily conveyed using interactive devices,such as the web-based guideline in the main resource referred to earlier (http://www.maternal-mortality-measurement.org). The first level of consideration distin-guishes between empirical and analytical approaches, with the former relying onnew data on deaths and the latter based on assessing, adjusting or modelling dataonce they have been gathered. This chapter focuses on empirical approaches, andreaders interested in analytical approaches are referred to a number of key re-sources.37,47,48,54–56 Many of the former approaches and tools can also be appliedto other deaths besides maternal57 and have evolved considerably over the last 20years in response to the continuing weaknesses of routine information systems indeveloping countries. The second level of consideration for differentiating empiricalapproaches and tools separates those primarily intended to find deaths from thoseused to determine which deaths are maternal or pregnancy-related. These twostreams are of course interrelated, such that an approach for finding deaths willthen use a particular tool to determine cause or time of death. The third level of con-sideration relates to availability, differentiating between sources or approaches that areroutine or ongoing and not peculiar to maternal mortality, such as death registration,those that involve special or ad hoc activities or studies, such as population surveys,and those that involve a combination of routine and special data capture, such asRAMOS. Eighteen empirical measurement items are introduced in the remainder ofthis section. For each there is a brief description, and reference is made to key guide-lines or scientific articles (further details, including advantages and disadvantages, aswell as measurement requirements, can be found at http://www.maternal-mortality-measurement.org). Selecting between these options for any particular context inwhich measuring progress is sought depends on a trade-off between practical

436 W. J. Graham et al

considerations of the availability of an ongoing data source, of budget and statisticalresources and of time, versus the scientific dimensions of measuring progress, suchas repeatability, reliability and attribution (discussed earlier).

Routine or ongoing systems or opportunities for findings deaths

Civil registration

Civil registration is a routine, permanent, nationally mandated data source thatcaptures vital life events (namely live births, deaths, fetal deaths, marriages and di-vorces). It provides legal documentation of such events and is the ideal source forvital statistics.58 However, even in contexts where civil registration is complete,underreporting and misclassification of maternal deaths is a frequent challenge.15,16

Civil registration can also be done on a sample basis, identifying deaths either activelyor passively.

Sample Vital Registration with Verbal Autopsy

The SAVVY method was developed by MEASURE Evaluation59 as a means of collectingmore detailed information on deaths and mortality in developing countries by combin-ing sample vital registration and verbal autopsy techniques. Inherent in the SAVVYmethod is the assumption that, alone, the latter tools cannot produce sufficiently de-tailed data on scope and causes of mortality14, but that by combining such tools, morecomprehensive data can be yielded. SAVVY is essentially a variation of demographicsurveillance (described further below). Where demographic surveillance systemsare established in a representative sample of areas (with annual or semi-annual updaterounds), SAVVY introduces active death reporting, which occurs in parallel. Deathsidentified are then followed-up with verbal autopsy by interviewers to determinecause of death.60,61

Decennial census

A national population census involves collecting, evaluating, analysing and disseminatingdemographic and socioeconomic data on all the persons in a country (or in a well-de-limited part of a country) at a specified time. Most countries do a national census ev-ery 10 years.62,63 Where countries already ask about deaths in the household in theyear before the census (or some other reference period), adding questions onwhether the deaths were pregnancy-related can be an efficient way to get nationaland subnational estimates of maternal mortality64,65, including the PMDF, MMR,MMRate and LTR (see Box 1). However, if the census is used for the measurementof maternal mortality, it is essential that these data be evaluated and adjusted, if nec-essary.26,66 Demographic techniques have been developed to adjust for under- andover-reporting of births55 or deaths.56

Routine health information system

In most countries, the health information system (HIS) can be regarded as a continuousand ongoing source of data on deaths, generated from health facilities including hospi-tals and health centres and reported by health professionals. Facilities compile statis-tics on maternal deaths and births and report these on to a central level, with codingof cause usually according to ICD-10. However, data quality can be variable7,15, and the

Measuring progress in reducing maternal mortality 437

usefulness of the HIS is often limited by the high proportion of deaths occurring in thecommunity, as discussed earlier. Tools, such as RAPID (see below), exist to checkwhether maternal deaths are being under-reported in facilities, particularly thosefrom indirect causes and those dying on non-obstetric wards.67,68 Strengthening ofroutine HISs is widely regarded as an international and national priority; the HealthMetrics Network is the current lead initiative in this regard.69

Special or ad hoc opportunities for findings deaths

Population-based household surveys

Household surveys are one of the most important data-capture platforms for maternaldeaths in settings where routine information systems are weak or non-existent. Prob-ability sampling ensures that target populations are representative.37 A particular ad-vantage is that confidence intervals can be calculated around maternal mortalityestimates.12 In addition, depending on the approach used, household surveys canalso gather useful information on causes, timing, place and consequences of deathas well as healthcare-seeking behaviour prior to death. There are currently threemain ways in which population-based household surveys are used to measure maternalmortality:

1. Using direct mortality questions: this involves the ascertainment of deaths in thehousehold in a recent interval of time, and is the approach also used in the decen-nial census.

2. Using indirect sisterhood method questions: the original (indirect) sisterhoodmethod13 was developed in the late 1980s as an efficient means of measuring ma-ternal mortality through population-based surveys, generating a variety of indica-tors: the PMDF, MMR, MMRate, LTR, and the adult female death rate. Adultrespondents are asked four questions about the survival of all their adult sistersborn to the same mother.36,70 The method reduces the need for large sample sizesbecause there is often more than one respondent per household, more than onesister per respondent, and because the time period of death is not restricted.36,71,72

3. Using direct sisterhood method questions: the direct sisterhood method is a variantof the indirect sisterhood method, and is currently used by Demographic and HealthSurveys (DHS).73 Although it requires larger sample sizes than its predecessor, theadvantage of the direct method is the targeting of a more limited reference periodfor sister deaths.24,74,75 Using a more detailed set of questions, in particular to ascertaindeaths among all siblings, and then those that are pregnancy-related, as well as the yearwhen the death occurred, point estimates for maternal mortality are generated, usuallyrelating to 0–6 years and>6 years prior to the survey. A retrospective MMR can thus becalculated for the reference period in question. The method is not recommendedfor duplication at short time intervals due to the large confidence intervals.74

Non-probability sampling

Non-probability sampling differs from probability sampling in that it does not use randomselection. Random selection ensures that all units being studied (i.e. members of a pop-ulation) have an equal opportunity of being selected for participation. Although this isusually considered to be the most rigorous sampling method, probability samples mightnot be feasible in certain contexts due to limited time, resources or other constraints.Non-probability samples can be useful in such circumstances. They fall into two main

438 W. J. Graham et al

categories: convenience and purposive sampling. Purposive sampling is more commonlyused and involves targeting social groups, experts or key informants76, as discussed fur-ther below. Results of non-probability techniques might not be representative of the gen-eral population, although this can be assessed for some techniques such as sampling atservice sites (see below), and conventional confidence intervals cannot be generated.

Sampling at service sites (SSS) is an adaptation of non-probability sampling for mea-suring maternal mortality. It samples respondents at services or other sites (such asantenatal care sites, child health services, or at marketplaces) and asks sisterhood-method questions.25,70 SSS uses high-coverage sites, data on respondent characteris-tics and reliance of reporting on siblings to overcome some biases.

Key informants

Key informants are people within communities who might have access to detailed in-formation about deaths or maternal deaths. Non-medical key informants might beparticularly useful in settings where many maternal deaths occur outside health facil-ities. Key informants can be used in a range of studies and are particularly common innon-probability samples76, in which they are targeted for their expertise or communityaccess, and might lead to the identification of other key respondents.77 They can alsobe used in RAMOS, active surveillance of pregnancy-related and maternal deaths78,and in CEMD. Key informants can include: community health workers, traditional birthattendants (TBAs), healthcare providers, village leaders, teachers, political cadres,cemetery workers and coffin makers. Key informants can be interviewed on a one-off or periodic basis. Accuracy can be improved if two networks of informants ortwo data sources are used with capture–recapture adjustment.27,79

Surveillance

Surveillance involves actively following populations to detect births and deaths. Thereare three main approaches: demographic surveillance systems (DSS), prospective stud-ies, and active surveillance of WRA deaths. In addition, CEMDs are a form of surveil-lance (discussed elsewhere in this volume), as are the SAVVY and sample vitalregistration systems described earlier.

Demographic surveillance systems began in the 1960s as a means of tracking lon-gitudinal demographic changes to populations in developing countries. Unlike prospec-tive (cohort) studies, DSS are can monitor entire populations and are usually largerand longer term.80 Field sites collect data on births, deaths (including causes) and mi-gration, which provide an important resource for evaluating healthcare interventions.They also offer a starting point for new studies.81 The INDEPTH network (an Inter-national Network for the continuous Demographic Evaluation of Populations andTheir Health in developing countries), is an international network of 31 DSS field sitesin 17 countries spanning Africa and Asia.82

Prospective studies

Prospective studies follow a group of individuals (a cohort or selected population)over a period of time to observe health outcomes, including mortality. Within thecontext of maternal mortality, a prospective study could involve following a populationof pregnant women or women of reproductive age.83,84 When an entire population isfollowed over time, this is usually termed a demographic surveillance system, as de-scribed above. Prospective studies often seek to test for an association between

Measuring progress in reducing maternal mortality 439

a certain exposure and the outcome of interest.85 Randomized controlled trials area special form of prospective study in that the exposure (treatment) is deliberatelyadministered at random to individuals in the cohort.

Active surveillance of pregnancy-related and maternal deaths

Active surveillance of pregnancy-related and maternal death is most often done as partof a CEMD, but can be used to obtain an estimate of the MMR. A key feature is theongoing effort to actively identify potential maternal deaths as they occur. In theUK CEMD21, for example, health staff – including midwives, obstetricians and generalpractitioners – are encouraged to report any possible maternal death to a confidentialenquiry assessor. In the US, active surveillance is done using a modified death notifi-cation form, birth and death record linkage and active notification by individual prac-titioners, maternal mortality committees and newspaper reports.22,86,87 In Egypt,maternal mortality surveillance is ongoing based on a modified death notificationform and interviews with the families of all pregnancy-related deaths.88

Composite approach using routine and special enquiries

Reproductive age mortality studies (RAMOS) use varied sources, depending on thecontext, to identify all deaths of women of reproductive age and ascertain which ofthese are maternal or pregnancy-related.89 The distinctive features of RAMOS arenot well defined in the literature. We characterize its key feature as starting with allreproductive aged female deaths (or a representative sample).90 Others have charac-terized RAMOS as relying on multiple sources of deaths, or as taking an in-depth lookat multiple determinants.30 RAMOS provide an estimate of the PMDF, but can be com-bined with other data to obtain the MMR, MMRate and LTR.

Tools for identifying maternal or pregnancy-related deaths

These are used within the context of the alternative approaches for initially findingdeaths described above.

Death certificates

Death certificates are vital records that document the medical causes and circum-stances of all deaths. A standard death certificate form asks for primary and underlyingcauses of death91, which might or might not capture those related to pregnancy. Asa result, even in settings where vital records are routine and complete, pregnancy-re-lated deaths can be missed. A Modified Death Notification Form92, which includes ad-ditional questions and tick-boxes related to pregnancy22, is preferable. In somesettings, such as South Africa, a separate Maternal Death Notification Form isused.93 This supplements a standard death certificate with more detailed informationabout the circumstances of a maternal death.

Autopsy

Also known as a post-mortem examination, an autopsy helps to determine primary andunderlying causes of death by examining the body and organs of a deceased person. Tissueand other samples can be taken during autopsy and tested by pathologists to provide fur-ther information. Findings are detailed in autopsy and associated pathology reports.94,95

440 W. J. Graham et al

Verbal autopsy

Verbal autopsy (VA) is an approach used to obtain cause of death by interviewing lay re-spondents on the signs and symptoms experienced by the deceased before death.96 It isused where vital registration systems are weak or the proportion of a population undermedical care is low. VAs usually involve three steps: (1) data collection by interviewingbereaved relatives orothers familiar with the circumstances of the death and who, ideally,were with the deceased during the events leading to death; (2) assignment of cause ofdeath using either individual or multiple physician reviews, expert algorithms or data-driven algorithms (regression or neural networks, or Bayesian approaches with proba-bilities of various diagnoses)97,98; and (3) coding and tabulation of causes, ideally usingthe ICD-10. VAs aim to identify maternal deaths that occur in communities (either withinor outside health facilities) and to identify broad subcauses of maternal mortality.9,99

They are often used as part of community-based maternal death reviews or CEMD,and are coupled with questions to ascertain both the medical and non-medical factorsthat precipitated a maternal death via in-depth interviews and questionnaires (includingopen-ended verbatim accounts, symptom checklists or checklists with filter questions).VAs can be done on a one-off basis or routinely as part of SAVVY, DSS, or active surveil-lance of pregnancy-related deaths, as described earlier.

RAPID

The Rapid Ascertainment Process for Institutional Deaths (RAPID) is a method de-signed to identify unreported maternal deaths within health facilities, and to highlightareas for improvement in routine institutional death reporting. Developed by Im-mpact100, RAPID involves reviewing health facility records of all deaths of women ofreproductive age and subsequently identifying pregnancy-related deaths. This processuncovers pregnancy-related deaths that might have been missed in routine reports.Once the level and circumstances of under-reporting is assessed, recommendationscan be made on how to improve gaps.

CONCLUSIONS

The need to monitor progress in reducing maternal mortality has a long history, whichcan be traced back to the 1700s in some parts of the Western world. Today, however,this need is felt most acutely in developing countries. By drawing attention to these avoid-able maternal deaths, the international Safe Motherhood Initiative, marked the beginningof two decades of efforts to understand and overcome the measurement challenges inthe context of weak information systems. Evidence of the success of these efforts exitsin terms of both the wide array of approaches, methods and tools presented in this chap-ter, and the declining proportion of countries now relying on modelled estimates of ma-ternal mortality, which has fallen from over 50% for 1990 to near a third in 2005.28 Thereare remaining challenges, which it would be foolish to deny, and much room for continu-ing research and development. The limitations of civil registration and routine health in-formation systems in many countries are still serious and further investment in theirimprovement is certainly needed. This does not, however, mean an indefinite wait forgood-quality data on maternal mortality, as there is now a suitable and feasible methodfor most country situations and most purposes, including tracking progress.101 Althoughsome of these approaches generate estimates at a reasonably low cost, none of them haszero costs. Greater investment is thus needed in measurement strategies for maternal

Measuring progress in reducing maternal mortality 441

mortality in developing countries. This must include strengthening technical capacity notonly to acquire the data but also to act on it, and so achieve the ultimate purpose – re-ducing the burden of preventable maternal deaths.

Research agenda

� Development and testing of composite approaches to data capture, using mul-tiple methods and tools, to help overcome the limitations of each individually.� New techniques dedicated to capturing deaths in early pregnancy, such as from

unsafe induced abortion.� Further field trials of innovative new tools, such as SSS and RAPID.� Meta-analytic methods that pool data from more than one source.� Improved techniques for validating estimates and for assessing and quantifying

uncertainty in estimates.� Capacity-strengthening initiatives to improve awareness of the need to assess

the quality of data on maternal mortality and apply relevant adjustmentprocedures.� Development of composite indicators that integrate aspects of processes of

care (coverage) with measures of mortality outcomes.� Enhanced indicators for looking at differentials in the risk of maternal mortality

according to social and/economic disadvantage.� Application of Geographic Information System (GIS) techniques to understand

spatial patterns in maternal mortality.� Better understanding of the value for information held by end-users of data on

maternal mortality and their acceptance of uncertainty in estimates.

ACKNOWLEDGEMENTS

W.J.G. is funded partially by the University of Aberdeen. O.M.R.C. is funded partially bythe London School of Hygiene and Tropical. W.J.G. and O.M.R.C. are also partiallyfunded, and L.B.F., E.H and L.D. fully funded by an international research programme –Immpact (http://www.immpact-international.org) supported by the Bill & Melinda GatesFoundation, the Department for International Development, the European Commissionand USAID. The funders have no responsibility for the information provided or views ex-pressed in this paper. The views expressed herein are solely those of the authors.

CONFLICT OF INTEREST

We declare we have no conflict of interest.

REFERENCES

1. Hill K. Making deaths count. Editorial. Bull World Health Organ 2006; 84: 162.

2. Mathers CD, Ma Fat D, Inoue M et al. Counting the dead and what they died from: an assessment of

the global status of cause of death data. Bull World Health Organ 2005; 83: 171–177.

3. Yazbak AS. Comment: challenges in measuring maternal mortality. Lancet 2007; 370: 1291–1292.

442 W. J. Graham et al

4. Berer M. Editorial: maternal mortality and morbidity: is pregnancy getting safer for women. Reprod

Health Matters 2007; 15(30): 6–16.

5. World Health Organization. Health and the Millennium Development Goals. Geneva: World Health

Organization, 2005.

6. Lancet. Editorial. Women: more than mothers. Lancet 2007; 370: 1283.

*7. Graham WJ & Hussein J. Universal reporting of maternal mortality: an achievable goal? Int J Gynaecol

Obstet 2006; 94(3): 234–242.

8. UNFPA. Issues in measuring and monitoring maternal mortality: implications for programmes. Technical and

Policy Paper No. 1. New York: United Nations, 1998.

*9. Ronsmans C, Vannest AM, Chakraborty J et al. A comparison of three verbal autopsy methods to as-

certain levels and causes of maternal deaths in Matlab, Bangladesh. Int J Epidemiol 1998; 27: 660–666.

10. International statistical classification of diseases and related health problems, 10th Revision.

2nd edition, Volume 2 Instruction Manual. Geneva: World Health Organization.http://

www.who.int/classifications/icd/ICD-10_2nd_ed_volume2.pdf

11. Laurenti R & Buchalla CM. Maternal and child health indicators: implications of the tenth re-

vision of the International Classification of Diseases. Pan American Journal of Public Health/Re-

vista Panamericana de Salud Publica 1997; 2(1): 13–17.

*12. Hill K, El Arifeen S, Koenig M et al. How should we measure maternal morality in the developing

world? A comparison of household deaths and sibling history approaches. Bull World Health Organ

2006; 84: 173–180.

13. Graham WJ, Brass W & Snow RW. Estimating maternal mortality: the sisterhood method. Stud Fam

Plann 1989; 20: 125–135.

14. Setel PW, Sankoh O, Velkoff VA et al. Sample registration of vital events with verbal autopsy:

a renewed commitment to measuring and monitoring vital statistics. Bull World Health Organ

2005; 83: 611–617.

15. Deneux-Tharaux C, Berg C, Bouvier-Colle MH et al. Underreporting of pregnancy-related mortality

in the United States and Europe. Obstet Gynecol 2005; 106(4): 684–692.

*16. Songane FF & Bergstrom S. Quality of registration of maternal deaths in Mozambique: a community-

based study in rural and urban areas. Soc Sci Med 2002; 54(1): 23–31.

17. AbouZahr C. Maternal mortality overview. In Murray CJ & Lopez AD (eds.). Health Dimensions of Sex

and Reproduction. Global Burden of Disease and Injury Series. Geneva: World Health Organization,

Harvard School of Public Health, World Bank, 2001.

*18. World Health Organization. Beyond the numbers: reviewing maternal deaths and complications to make

pregnancy safer. Geneva: WHO, 2004.

19. Loudon I. Death in Childbirth. An international study of maternal care and maternal mortality 1800–1950.

Oxford: Oxford University Press, 1992.

20. Poppel F & Van Dijk JP. The development of cause-of-death registration in the Netherlands, 1865–

1955. Contin Chang 1997; 12: 265–287.

*21. Macfarlane A. Enquiries into maternal deaths during the 20th century. In: Why Mothers Die 1997–

1999. The Confidential Enquiries into Maternal Deaths in the United Kingdom. London: RCOG Press,

2001, pp. 346–357.

22. MacKay AP, Rochat R, Smith SC et al. The check box. Determining pregnancy status to improve

maternal mortality surveillance. Am J Prev Med 2000; 19: 35–39.

23. WHO. Maternal mortality ratios and rates: a tabulation of available information. Geneva: World Health

Organization, 1991.

24. Stanton C, Abderrahim N & Hill K. DHS maternal mortality indicators: an assessment of data and

implications for data use. Demographic and Health Surveys Analytical Report No 4. Calverton, Maryland

USA: Macro International Inc.

25. Immpact, University of Aberdeen. Sampling at Service Sites (SSS), Module 4, Tool 1. In: Immpact Tool-

kit: a guide and tools for maternal mortality programme assessment. Aberdeen: University of Aberdeen,

2007, http://www.immpact-international.org/toolkit/module4/sss/sss.pdf.

*26. Stanton C, Hobcraft J, Hill K et al. Every death counts: measurement of maternal mortality via a cen-

sus. Bull World Health Organ 2001; 79: 657–664.

27. Amstrup SC, McDonald TL & Manly BFJ. Handbook of Capture-Recapture Analysis. Princeton Univer-

sity Press, 2005.

Measuring progress in reducing maternal mortality 443

28. Hill K, Thomas K, AbouZahr C et al. Estimates of maternal mortality worldwide between 1990 and

2005: and assessment of valid data. Lancet 2007; 370: 1311–1319.

29. Graham W. Now or never: the case for measuring maternal mortality. Lancet 2002; 359(9307): 701–

704.

30. World Health Organization. Studying maternal mortality in developing countries: a guidebook. WHO/

FHE/87.7. Geneva: World Health Organization, 1987.

31. AbouZahr C & Wardlaw T. Maternal mortality at the end of a decade: signs of progress? Bull World

Health Organ 2001; 79(6): 561–573.

32. Gakidou E, Hogan M & Lopez AD. Adult mortality: time for a reappraisal. Int J Epidemiol 2004; 33:

710–717.

33. Betran AP, Wojdyla D, Posner SJ et al. National estimates for maternal mortality: an analysis based on

the WHO systematic review of maternal mortality and morbidity. BMC Public Health 2005; 5: 131.

34. Ronsmans C & Graham WJ. Maternal mortality: who, when, where and why. Lancet 2006;

368(9542): 1189–1200.

35. Saleh S, Gadalla S & Fortney J. Maternal mortality in Menoufia: a study of reproductive age mortality.

Cairo: Social Research Center, The American University of Cairo, 1987.

36. WHO. The sisterhood method for estimating maternal mortality: guidance notes for potential users. Ge-

neva: World Health Organization, 1997, http://www.who.int/reproductivehealth/publications/

rht_97_28/sisterhood_method_estimating_maternal_mortality.pdf.

37. United Nations. Household sample surveys in developing and transition countries. New York: United Nations,

ST/ESA/STAT/SER.F/96., 2005, http://unstats.un.org/unsd/hhsurveys/pdf/Household_surveys.pdf.

38. Campbell OMR. Measuring progress in safe motherhood programmes: uses and limitations of health

outcome indicators. Reprod Health Matters 1999: 31–42.

39. Walker N, Bruce J & Black RE. Interpreting health statistics for policymaking: the story behind the

headlines. Lancet 2007; 369: 956–963.

40. Bicego G, Boerma JT & Ronsmans C. The effect of AIDS on maternal mortality in Zimbabwe and

Malawi. Aids 2002; 16: 1078–1081.

41. Say L, Pattinson RC & Gulmezoglu M. WHO systematic review of maternal morbidity and mortality:

the prevalence of severe acute maternal morbidity (near miss). Reprod Health 2004; 1: 3.

42. Ronsmans C, Campbell OMR, McDermott J et al. Questioning indicators of need for obstetric care.

Bull World Health Organ 2002; 80: 317–324.

43. UNICEF & WHO. Guidelines for monitoring the availability and use of obstetric services. New York:

WHO, 1997.

44. Bryce J, Terreri N, Victora CG et al. Countdown to 2015: tracking intervention coverage for child

survival. Lancet 2006; 368: 1067–1076.

45. Campbell OMR & Graham WJ. Strategies for reducing maternal mortality: getting on with what

works. Lancet 2006; 368(9543): 1284–1299.

46. Graham WJ, Fitzmaurice AE, Bell JS et al. A new method for linking maternal death with poverty: the

familial technique. Lancet 2004; 363: 23–27.

47. United Nations. Manual X: indirect techniques for demographic estimation. Department of Eco-

nomic and Social Affairs. Population Studies, No. 81. 1983. ST/ESA/SER.A/81.http://www.un.org/

esa/population/publications/Manual_X/Manual_X.htm

48. Bennett NG & Horiuchi S. Mortality estimation from registered deaths in less developed countries.

Demography 1984; 21(2): 217–234.

49. Milne L, Scotland G, Tagiyeva-Milne N et al. Safe motherhood program evaluation: theory and prac-

tice. J Midwifery Women’s Health 2004; 49: 338–344.

50. Victora CG, Habicht JP & Bryce J. Evidence-based public health: moving beyond randomised trial. Am

J Public Health 2004; 94: 400–405.

51. Initiative for Maternal Mortality Programme Assessment (Immpact), http://www.immpact-

international.org.

52. Van Lerberghe W & De Brouwere V. Of blind alleys and things that have worked: history’s lessons on

reducing maternal mortality. In De Brouwere V & Van Lerberghe W (eds.). Safe motherhood strate-

gies: a review of the evidence. Studies Health Serv Organ Policy 2001; 17:, pp 1273–1280.

53. Chowdhury ME, Botlero R, Koblinsky M et al. Determinants of reduction in maternal mortality in

Matlab, Bangladesh: a 30-year cohort study. Lancet 2007; 370: 1320–1328.

444 W. J. Graham et al

54. MortPak manual, http://www.un.org/esa/population/publications/MortPak_SoftwarePkg/MortPak_CH1.

pdf.

55. UnitedNations.Handbookon the collection of fertility and mortality data. DepartmentofEconomic andSocial

Affairs, Statistics Division. Series F No.92; ST/ESA/STAT/SER.F/92. New York: United Nations, 2004,

http://unstats.un.org/unsd/demographic/standmeth/handbooks/Handbook_Fertility_Mortality.pdf.

56. United Nations. Methods for estimating adult mortality. Population Division, Department of Economic

and Social Affairs UN Secretariat. P21-26. ESA/P/WP.175, 2002, http://www.un.org/esa/population/

publications/adultmort/Complete.pdf.

57. Brass W. Demographic data analysis in less developed countries: 1946–1996. Popul Stud 1996; 50: 451–

467.

58. United Nations. Principles and recommendations for a vital statistics system. Revision 2. Department of

Economic and Social Affairs. Statistics Division. New York: United Nations, 2001. ST/ESA/STAT/

SER.M/19/Rev.2.

59. MEASURE Evaluation. MEASURE Evaluation’s SAVVY: a toolset for counting every person. FS-06–04 (12/

03/06). Chapel Hill: MEASURE Evaluation, 2006, http://www.cpc.unc.edu/measure/publications/

sheets/FS-06-04%20SAVVY.pdf.

60. Setel PW, Rao C, Hemed Y et al. Core verbal autopsy procedures with comparative validation re-

sults from two countries. PLoS Med 2006; 3(8): e268.

61. Setel PW, Whiting DR, Hemed Yet al. Validity of verbal autopsy procedures for determining cause of

death in Tanzania. Trop Med Int Health 2006; 5: 681–696.

62. United Nations Statistics Division site, http://unstats.un.org/unsd/demographic/sources/census/.

63. US Census, http://www.census.gov/ipc/www/cspro/index.html.

64. Hill K & on behalf of participants at the Expert Panel on Capturing Maternal Mortality in the 2010

Census Round. Measuring maternal mortality. Lancet 2006; 368(9553): 2121.

65. Immpact/Health Metrics Network meeting Census meeting, http://www.abdn.ac.uk/immpact/news/

meeting2.php; September 2006.

66. Hill K, Stanton C & Gupta N. Measuring maternal mortality from a census: Guidelines for potential users.

Chapel Hill: MEASURE Evaluation, 2001, http://www.cpc.unc.edu/measure/publications/pdf/ms-01-

04.pdf.

67. Graham W & Hussein J. The right to count. Lancet 2004; 363: 67–68.

68. Sombie I, Meda N, Hounton S et al. Missing maternal deaths: lessons from Souro Sanou University

Hospital in Bobo-Dioulasso, Burkina Faso. Trop Doct 2007: 96–98.

69. Health Metrics Network. Strengthening country health information systems: assessment and monitoring

tool. Version 1.96: 04. Geneva: World Health Organization, https://www.who.int/healthmetrics/doc-

uments/hmn_assessment_tool_guide_english_v1_96.pdf.

70. Danel I, Graham W, Stupp P et al. Applying the sisterhood method for estimating maternal mortality

to a health facility-based sample: a comparison with results from a household-based sample. Int J Epi-

demiol 1996; 25(5): 1017–1022.

71. Hanley JA, Hagen CA & Shiferaw T. Confidence intervals and sample size calculations for the

sisterhood method of estimating maternal mortality. Stud Fam Plann 1996; 27(4): 220–227.

72. Shahidullah M. The sisterhood method of estimating maternal mortality: the Matlab experience. Stud

Fam Plann 1995; 26(2): 101–106.

73. DHS Maternal Mortality Questionnaire, http://www.measuredhs.com/basicdoc/modules/DHSIV/

MATERNAL.DOC.

74. Stanton C, Abderrahim N & Hill K. An assessment of DHS maternal mortality indicators. Stud Fam

Plann 2000; 31(2): 111–123.

75. Rutenberg N & Sullivan JM. Direct and indirect estimates of maternal mortality from the sisterhood

method. Washington, DC: IRD/Macro International Inc., 1991.

76. Singh P, Pandey A & Aggarwal A. House-to-house survey vs. snowball technique for capturing ma-

ternal deaths in India: a search for a cost-effective method. Indian J Med ResIndian J Med Res 2007;

125: 550–556.

*77. Walker GJA, McCaw-Binns A, Ashley DEC et al. Identifying maternal deaths in developing countries:

experience in Jamaica. Int J Epidemiol 1990; 19: 599–605.

78. Immpact/Population Reference Bureau. Delivering safer motherhood: sharing the evidence. Policy brief.

Washington, DC: Population Reference Bureau, http://www.prb.org/pdf07/IMMPACTpolicybrief.pdf.

Measuring progress in reducing maternal mortality 445

79. Buescher PA, Harper M & Meyer RE. Enhanced surveillance of maternal mortality in North Carolina.

Raleigh: Center for Health Informatics and Statistics, http://www.schs.state.nc.us/SCHS/pdf/

CHIS125.pdf.

80. Baiden F, Hodgson A & Binka FN. Demographic surveillance sites and emerging challenges in

international health. Bull World Health Organ 2006; 84(3): 163–164.

81. Nhacolo AQ, Nhalungo DA, Sacoor CN et al. Levels and trends of demographic indices in southern rural

Mozambique: evidence from demographic surveillance in Manhica district. BMC Public Health 2006; 6:291

doi:10.1186/1471-2458-6-291.

82. INDEPTH Network. Population and health in developing countries: Volume1: Population, health and sur-

vival at INDEPTH sites. Ottawa: International Development Research Centre, 2002.

83. Greenwood AM, Greenwood BM, Bradley AK et al. Prospective survey of the outcome of pregnancy

in a rural area of the Gambia. Bull World Health Organ 1997; 65(5): 635–643.

84. Thonneau PF, Matsudai T, Alihonou E et al. Distribution of causes of maternal mortality during de-

livery and post-partum: results of an African multicentre hospital-based study. Eur J Obstet Gynecol

Reprod Biol 2004; 114(2): 150–154.

85. Fenton PM, Whitty CJM & Reynolds F. Caesarean section in Malawi: prospective study of early ma-

ternal and perinatal mortality. BMJ 2003; 327: 587.

86. Horon IL. Underreporting of maternal deaths on death certificates and the magnitude of the prob-

lem of maternal mortality. Am J Public Health 2005; 95(3): 478–482.

87. Jocums S, Mitchel Jr. EF, Entman SS et al. Monitoring maternal mortality using vital records linkage.

Am J Prev Med 1995; 11(2): 75–78, http://www.ncbi.nlm.nih.gov/pubmed/7632453?ordinalpos¼168&itool

¼EntrezSystem2.PEntrez.Pubmed.Pubmed_ResultsPanel.Pubmed_RVDocSum.

*88. Campbell O, Gipson R, Issa AH et al. National maternal mortality ratio in Egypt halved between

1992–93 and 2000. Bull World Health Organ 2005; 83: 462–471.

89. Measure Evaluation Compendium, http://www.cpc.unc.edu/measure/publications/html/ms-02-09-

tool20.html.

90. Smith JM & Burnham G. Conceiving and dying in Afghanistan. Lancet 2005; 365(9462): 827–828.

91. Instructions for Classifying the Underlying Cause-of-Death, ICD-10, http://www.cdc.gov/nchs/data/

dvs/2a2007.pdf; 2007.

92. Hoyert DL. Maternal mortality and related concepts. National Center for Health Statistics. Vital

Health Stat 2007; 3(33): 4–11.

93. Pattinson RC. Guidelines for completing the maternal death notification form. 2nd ed. Pretoria: South

African Department of Health, 1999, http://www.doh.gov.za/docs/forms/guide_maternal.pdf.

94. Hoyert D. The autopsy, medicine, and mortality statistics. National Center for Health Statistics. Vital

Health Stat 2001; 3(33). Centers for Disease Control, http://www.cdc.gov/nchs/data/series/sr_03/

sr03_032.pdf.

95. Millward-Sadler GH. Pathology of maternal deaths. In Kirkham N & Shepherd N (eds.). Progress in

Pathology 2003; vol. 6. London: Greenwich Medical Media Ltd, 2003, pp. 163–185.

96. Chandramohan D, Maude GH, Rodrigues LC et al. Verbal autopsies for adult deaths: issues in their

development and validation. Int J Epidemiol 1994; 23: 213–222.

97. Fottrell E, Byass P, Ouedraogo TW et al. Revealing the burden of maternal mortality: a probabilistic

model for determining pregnancy-related causes of death from verbal autopsies. Popul Health Metr

2007; 5: 1.

98. InterVA-M, http://www.interva.net/.

*99. Chandramohan D, Maude GH, Rodrigues LC et al. The validity of verbal autopsies for assessing the

causes of institutional maternal death. Stud Fam Plann 1998; 29: 414–422.

100. Immpact, University of Aberdeen. Rapid ascertainment process for institutional deaths (RAPID).

Module 4, Tool 2. In: Immpact Toolkit: a guide and tools for maternal mortality programme assessment.

Aberdeen: University of Aberdeen, 2007, http://www.immpact-international.org/toolkit/module4/

rapid/rapid.pdf.

101. Fillippi V, Ronsmans C, Campbell OMR et al. Maternal health in poor countries: the broader context

and the call for action. Lancet 2006; 368(9546): 1535–1541.