© 2011 ReproduAll righ

Reproductive Health M0968-8080/11PII: S0968-8www.rhm-elsevier.com

ctive Health Matters.ts reserved.atters 2011;19(37):117–128$ – see front matter080 (11 ) 37550-7 www.rhmjournal.org.uk

DISCUSSION

New estimates of maternal mortality andhow to interpret them: choice or confusion?

Carla AbouZahr

Consultant, health statistics; formerly World Health Organization staff member, Geneva, Switzerland.E-mail: abouzahr.carla@gmail.com

Abstract: Two independent exercises to estimate levels of maternal mortality took place during2010, one published by the Institute for Health Metrics and Evaluation in Seattle, USA, the otherpublished by four UN agencies (UNICEF, UNFPA, World Bank and World Health Organization).Although both approaches are based on similar sets of empirical country data, their statisticalmethods differ in important respects – with implications for the resulting global, regional andcountry estimates. This paper examines the differences, discusses both the value and inherentlimitations in such exercises, proposes ways of interpreting the different estimates and suggestshow such exercises could be made more relevant to the needs of country-level decision-makers.It calls on the global community to invest seriously in working with countries to generateprimary data on maternal mortality using measurement methods that reduce uncertainty andgenerate data on a continuing basis. The best routine source of data on maternal deaths is acivil registration system that assures permanent, compulsory and universal recording of theoccurrence and characteristics of vital events, including births and deaths, and causes ofdeath. The record of deaths among women of reproductive age derived from civil registration isoften the first step in conducting a confidential enquiry into and preventing maternal deaths.©2011 Reproductive Health Matters. All rights reserved.

Keywords: maternal mortality, statistical modelling, vital statistics, health policies and programmes

2010 was the year of maternal mortality esti-mates. In April, the Lancet published mater-nal mortality figures developed at the

Institute for Health Metrics and Evaluation(IHME), an academic institution based at theUniversity of Washington, Seattle, USA.1 In Sep-tember, a different set of numbers was issued byUN agencies UNICEF, UNFPA, World Bank andWorld Health Organization, working in collabo-ration with technical experts from the Univer-sity of Berkeley, California, USA.2 Both sourcesincluded data for nearly all countries (IHME 181,UN 174), along with regional and global totals.Both covered similar time spans (IHME 1980–2008,UN 1990–2008). Both calculated overall andannual average rates of change. Both claimed tobe based on a systematic review of all available

data. Both applied adjustments to country-reported data in order to improve comparabilityand correct for bias. Both used statistical model-ling to generate estimates for countries or timeperiods where data are lacking. Both acknow-ledged the important impact that HIV has hadon maternal mortality, especially in sub-SaharanAfrica. Both claimed to have found substantialdeclines in maternal mortality in recent years.Both included estimates of uncertainty aroundthe numbers.Despite these broad measures of agreement,

there are important differences, both in theregional and global totals and in the individualcountry estimates. So what is a potential user tomake of these two sets of numbers? Is one superiorto the other? How can users choose between the

117

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

two? Why are the UN estimates for 2008 so dif-ferent from those issued in 2005? For those con-fused by the sudden upsurge in numbers – andyou are not alone – here are some frequently askedquestions and answers that may help in under-standing, and using, the new estimates.

What are the differences between the twosets of numbers?• Globally, the differences in maternal deathsare smallIn terms of numbers of maternal deaths in 2008,the difference between the two sets of estimatesis around 4%. IHME estimated some 342,900maternal deaths compared with UN estimatesof 358,000.

• Uncertainty ranges are significantly differentWith regard to the estimates of the maternalmortality ratio (MMR), IHME estimated 251 per100,000 live births in 2008 (range 221–289)compared with the UN estimate of 260 (range200–370). As is clear from Figure 1, the uncer-tainty range is wider for the UN estimates thanfor those developed by IHME. This is due to thestatistical methods used to calculate uncer-tainty, but does not mean that the UN esti-mates are inherently less precise that IHME's;both are uncertain.

• Rates of change are significantly differentAn important difference between the two sets ofestimates is the rate of change since 1990. IHMEestimated a global annual average rate of declinebetween 1990 and 2008 of 1.3%, (range 1.0, 1.5)compared with a decline of 2.3% estimated bythe UN agencies (range 2.1, 2.4).* The expla-nation lies in the differences in the estimated1990 starting point, with IHME estimating a1990 maternal mortality ratio of 320 per 100,000compared with the UN estimate of 400 per 100,000(Figure 2). Either way, both sets of estimates showrates of changewell below the annual 5.5%declinethat would be needed to attain the MDG5 target. Itis worth noting that an attempt in 2005 by the UNagencies to estimate rates of change suggested anannual rate of decline between 1990 and 2005 ofonly 0.4% overall, but 2.5%when limited to coun-tries with several empirical data points.3

• There are important differences incountry estimatesWhereas the global figures and trends appearbroadly similar, there are important differenceswhen it comes to individual country values.IHME estimates are higher than UN estimatesin some countries; the reverse is true in others(Figure 3).

*However, IHME estimated a rate of decline in a “no HIVscenario” of 2.4% annually.

118

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

What are the causes of these differences?The differences between the two sets of country-level estimates are largely due to:

• Differences in the underlying empiricallyavailable data.

• The way available data are adjusted to accountfor bias.

• The way deaths among HIV-positive preg-nant women are dealt with.

• The use of different estimates of total deathsin women of reproductive age.

• The specifications of the statistical modelsused to generate missing values.

• Underlying data availabilityThe IHME and UN estimates are based onsomewhat different country-reported data. Ofcourse, both groups made every effort to takeinto account as many country data points aspossible in developing their estimates. The UNconstructed a database of 484 empirical dataobservations, several of which covered multipleyears, thus generating 2961 country-years of data.The IHME database of 2651 observations wasconstructed by analysing microdata from surveyscovering multiple years to generate individualvalues for each year. Because data points werecounted in different ways, the databases are not

strictly comparable. Inevitably, there will be datapoints that have been missed by one group or theother and this will give rise to differences in thefinal estimates. However, the impact of such dif-ferences is generally small and could be reducedfurther by sharing databases more effectivelythan has been the case up to now.

• Data adjustment proceduresA more important source of difference betweenthe two sets of estimates is the way country-reported data were handled. Both the IHMEand the UN estimates took as the starting pointnot the maternal mortality ratio itself but,rather, the proportion of all deaths occurring inwomen of reproductive age due to maternalcauses (PMDF). Both groups adjusted these datato account for biases andmisreporting, for exam-ple due to misclassification of maternal deaths toother causes. In adjusting data from civil regis-tration, IHME conducted a detailed empiricalexamination of deaths classified to “ill-definedand unspecified causes”. The UN adjustmentto the PMDF derived from civil registration wasbased on a review of published literature on theextent of misclassification of maternal deathsreported in reproductive age mortality studies.4,5

Interestingly, the two different approaches cameup with similar estimates of misclassification,

119

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

leading to a 40% increase in maternal deaths inthe IHME study compared with 50% in the UNestimates. Two adjustments were applied to thePMDF derived from surveys, an upward one toaccount for under-identification of maternaldeaths due to abortion, and a smaller downwardadjustment to account for “incidental” deaths, thatis, deaths in pregnantwomen thatwere not relatedto or aggravated by the pregnancy. According tothe ICD definitions, these should not be countedamong maternal deaths (see Box 1). The resultingupward and downward adjustments to the PMDFwere almost identical, resulting in a net changeof around 1%. At each step there is room fordebate about the choices made and, inevitably, adegree of subjectivity is involved. For example,both datasets initially included data elementsdetermined, following qualitative review, to be“implausible” or “outliers”. These were not sub-sequently included in the estimation process.Unlike the IHME exercise, the UN estimation alsoexcluded all sub-national data sets.

• Deaths in HIV-positive pregnant womenThe two sets of estimates also have differentways of dealing with HIV infection in pregnantwomen and its impact on overall maternal mor-

120

tality. The IHME analysis used HIV prevalence asa covariate for the model. The UN approach wasto take available data on total deaths in womenof reproductive age as the starting point, andapply a statistical model and various assump-tions to generate estimates of HIV-associatedmaternal deaths.

• The envelope of deaths of women ofreproductive agePerhaps the single most important difference inmethods is the way the proportion of mater-nal deaths among women of reproductive agepredicted by the statistical model was used to cal-culate total maternal deaths. Because few deve-loping countries have reliable counts of deathsby age and sex, total deaths of reproductive-age women are usually estimated from life tables.However, IHME and the UN used different lifetables and this is an important source of the dif-ferences between the two sets of data. The IHMEdeveloped country-specific life tables for adultfemale mortality, taking into account availablesurvey data on sibling survival, adjusted usingthe Gakidou-King procedure to correct for sur-vivor bias.6 The UN estimates, on the otherhand, used existing published UN estimates ofadult female mortality as the envelope. Unfortu-nately, it is not possible to quantify the effect ofthis factor on the overall maternal mortalitynumbers, because to date IHME has not pub-lished the details of its envelope estimates.

• Statistical model specificationsBoth sets of estimates used statistical models topredict values for countries and/or time periodsfor which empirical data are not available. Inevi-tably, differences in statistical methods, modelspecifications and covariates give rise to dif-ferent results. The UN estimates were derivedfrom a multilevel regression model that includedrandom elements at the level of observations,countries and regions. In simple terms, themodel takes into account both the nature ofthe underlying empirical data as well as coun-try and regional specificities. The underlyingempirical data drive the overall level of mater-nal mortality but the covariates, especially GNIper capita are major drivers of the trends overtime. The IHME used a two-stage approach:a linear model and a spatial-temporal model

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

designed to capture real systematic variationnot captured by covariates.In the UN analysis, covariates included Gross

National Income (GNI) per capita, general fer-tility rate and proportion of deliveries attendedby a skilled health worker. IHME covariatesincluded total fertility rate (TFR), GNI per capita,HIV seroprevalence, neonatal mortality, age-specific female education, and age. Skilled birthattendance was included but did not add to pre-dictive validity. In practice, GNI is the most sig-nificant driver of trends in both sets of estimates.It is important that this finding not be mis-interpreted: it does not imply that improvementsin maternal mortality are simply a matter ofincreasing national wealth as opposed to otherinterventions, such as access to reproductive andmaternal health care. The statistical model isdescriptive rather than explanatory.In several instances, the modelled estimates

are based on no country-specific data relatedto maternal mortality, and trends are entirelydriven by changes in the covariates, especiallyGNI per capita. Examples include Angola andEquatorial Guinea (no empirical data), Lao PDR(one data point for 1990), and Myanmar (noempirical data). This is a major weakness in bothsets of estimates. The apparent decline in mater-nal mortality in a country such as EquatorialGuinea, for example, seems to be entirely amatter of changes in national income due tothe rising price of oil and has little to do withinterventions to reduce maternal mortality.It is legitimate to ask whether one statistical

approach is better than the other but perhapsthis is the wrong question. In both models theunderlying data inputs are sparse and biased,the definitions used are inconsistent, the model-ling is complex and the uncertainty of the result-ing estimates is large. The big problem that needsto be addressed is the absence of country leveldata which no amount of tinkering with statisti-cal models can overcome.

How do the 2008 estimates differ fromthose issued previously by the UN?This paper is focused on the differences betweenthe estimates issued by the UN and IHME for2008. It does not directly address the differencesbetween the UN estimates for 2008 and thoseissued previously by the UN for 1990, 1995,

2000 and 2005. As noted in the documentationaccompanying each set of UN estimates, thedata adjustment and statistical methods usedon each occasion differed and the resultingglobal, regional and country estimates shouldnot be considered as comparable. Having saidthis, there are a number of constants in theapproaches used by the UN since 1990. Theseinclude the focus on the PMDF rather than onthe maternal mortality ratio and the adjustmentsapplied to empirical country data depending onthe source and methods used. The innovations inthe 2008 UN estimates approach include a morerigorous effort to account for “incidental” deathsamong pregnant women, including those due toHIV, and a more complex statistical modellingapproach designed to better estimate both levelsand trends in maternal mortality.

Why are the estimates often so differentfrom those reported by countries?Both the IHME and UN estimates aim to maxi-mise comparability, both over time and acrosscountries. But this is difficult when country-reported data derive from different sources anddata collection methods, use different defini-tions, and have varying time reference points.For example, when maternal mortality is derivedfrom the census or from household surveys,what is actually measured is pregnancy-relatedmortality rather than maternal mortality (Box 1).7

Depending on the data source, the resultingfigures relate to different definitions and dif-ferent time periods and have different levels ofprecision and reliability (Table 1). In some coun-tries, maternal mortality is reported through thecivil registration system but unless coverage isalmost complete (at least 90%) and all deathsare properly medically certified, the resulting dataare likely to be serious undercounts.8 Elsewhere,household surveys are used as the data source,using either direct or indirect (sisterhood) mea-surement methods. These generate values thathave important uncertainty bounds, due to a com-bination of sampling and non-sampling errors.In other settings, the census is used to estimatematernal mortality but here too, there are likelyto be problems of under-reporting; mortality dataderived from the census need to be reviewed andadjusted using demographic techniques, some-thing that not all countries do systematically.9

121

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

122

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

Many of the difficulties associated with moni-toring maternal mortality arise from the fact thatmaternal deaths are relatively rare events, onlyabout 5% as common as child deaths. The smallnumbers involved mean that national trendstend to be unstable, and sub-national ones evenmore so. This is especially true when mortalitylevels are very low and when household surveysare used as the data source. In countries withvery small absolute numbers of maternal deaths,changes of one or two deaths in the numeratorcan appear to have a disproportionate effect onthe maternal mortality ratio. Estimates generatedusing surveys are themselves subject to widevariations, which are not always described inthe country reports (Figure 4).

Can the estimates be used for both countryand global monitoring?The impetus to obtain better data to track prog-ress began to grow following the 1990 Child Sur-vival Summit, and has been further magnifiedby monitoring and evaluation related to theMillennium Development Goals. Both the IHMEand the UN estimates sought to address two chal-lenges: (1) how to bring together, in a common

format and using similar sets of assumptions,data for multiple time periods and from diversedata sources in order to generate a set of figurescomparable across countries and over time; and(2) how to predict maternal mortality in countriesand for time periods with no empirical data. Bothstatistical models generate figures for countrysettings where no primary data are available andthey produce updated, future or retrospectivevalues, such as for the MDG base year of 1990.The results of these global estimation exercises

should be used carefully. Estimates generatedusing statistical “forecasting” and “far-casting”techniques can be used for advocacy, planning,strategic decisions, and identifying research anddevelopment priorities. However, they should notbe used for monitoring progress towards agreedtargets and for an assessment of what is effectiveand what is not.10

What country decision-makers actually needare data that are accurate, frequent and rapidlyavailable at national and sub-national levels.Such data are essential to identify whether andto what extent their policies and programmesare achieving the anticipated results. Estimatesbased on statistical models do not answerthese needs.

123

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

Do these new estimates mean thatmore data on maternal mortality arenow available?In their MDG reports, UN agencies have regu-larly drawn attention to the paucity of reliableand comparable country data and the ensuingchallenge for monitoring progress.11 The IHMEstudy contests the claim that maternal mortalityis inherently hard to measure, noting that “Moredata are available for maternal mortality thanfor other main causes of child or adult death…the number of data points directly measuringtuberculosis as a cause of death is much lowerthan that for maternal mortality. A compari-son… with HIV and many causes of child mor-tality is similarly favourable.”1 There is sometruth in this statement. The maternal mortalityratio has been an international target at leastsince the 1990 World Summit for Children andresearchers have devoted considerable attentionto developing measurement approaches, such asthe sisterhood method.12 Despite this, at countrylevel, maternal mortality data are sparse. In theUNdatabase, during the period from the late 1980sto 2008, there were an average of only 2.8 obser-vations per country.13

The paucity of cause-specific mortality data isa widespread challenge. Despite the growingattention directed to measuring mortality dueto AIDS, TB and malaria, monitoring is usuallylimited to indicators of disease incidence andprevalence and access to preventive and/or treat-ment interventions such as insecticide-treatedmosquito nets, coverage of TB treatment, andpeople in need receiving antiretroviral thera-pies for advanced HIV infection. The successesachieved in recent years in methods to measurechild mortality have not been matched by similarprogress in tackling the measurement challengesfor causes of death or for adult mortality overall.

Why are there two sets of estimatesanyway? Would it not be better to havejust one agreed set of estimates?Until 2010, UN agencies took the lead on deve-loping global estimates, not only for maternalmortality but for child mortality, HIV and TBincidence and prevalence and other health indi-cators. The arrival of the IHME – an academicgroup based at the University of Washingtonin Seattle and funded largely by the Bill and

124

Melinda Gates Foundation – has important impli-cations, many positive, others less so. On the onehand, the scientific method is defined by inno-vation, transparency, critical evaluation and theconfrontation of alternative hypotheses. On theother hand, what is good from a scientific pointof view may be less so from a country perspec-tive, where multiple, often contradictory esti-mates create confusion and scepticism whichneed to be avoided.Both the UN and IHME bring together tech-

nical expertise from around the world in orderto ensure a strong scientific foundation for theirwork. In this instance, this gave rise to differ-ences of opinion on the most appropriate sta-tistical methods. Disagreements and debate area normal part of scientific discourse and canhelp improve the knowledge base and stimu-late better empirical data collection in country.Forcing a scientific consensus around estimatesthat are characterised by a high degree of uncer-tainty is not the way forward. But it is incumbentupon the agencies and academic institutionsundertaking such work to minimise confusionand maximise transparency by sharing data andopenly debating statistical methods and dataadjustments.Moreover, the production of estimatesshould go hand in hand with the development oftools and methods that build capacity in countriesfor data generation, analysis and interpretation.Furthermore, there has been a spirit of com-

petitiveness and unseemly haste in the releaseof estimates, often linked to high profile politi-cal events such as the meetings of the G8 inJune in Canada (which included special atten-tion to maternal and child health)14 and the UNGeneral Assembly in September 2010 (whichfocused on progress towards the MDGs.)15 Thepublication of the IHME results in the Lancetwas criticised for a rushed peer review process.16

The UN estimates were released independently,bypassing peer review journals and provokingcritical comments in the Lancet.17

Are there risks involved in these globalestimation exercises?Global estimation efforts have some negative,unintended consequences, especially at countrylevel. Neither the UN nor the IHME involved sci-entists or institutions from developing countries.Because of this lack of country engagement in

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

and ownership of the process, the resulting esti-mates are perceived as externally managed,data-mining exercises, bringing few benefitsto countries themselves. The UN process didinvolve a measure of country consultation, butthis was insufficient to build capacities for dataanalysis and interpretation. For country policy-makers, the existence of different values forsame indicators creates confusion and risksfostering a culture of “gaming” around thevarious figures. Global estimates also discour-age a culture of local accountability: if exter-nal institutions produce country numbers, thereis no need to allocate resources in country togenerate empirical data. On the other hand, ifno estimate is available, national policy-makershave an excuse for inaction.Following the release of the estimates, both

the IHME and the UN agencies have sponsoredinter-country workshops designed to enhanceunderstanding of the rationale and methodsfor the estimates. This is a welcome develop-ment but sadly, the two groups are still workingentirely separately. What a missed opportunity!Global estimation efforts can only benefit fromdirect discussions of the methods on a sharedplatform with country data producers, technicalexperts and data users. The maternal healthcommunity should advocate for better collabo-ration in any future country workshops.

Should UN agencies leave estimationto academic institutions?In the light of these debates, questions havebeen raised as to whether the UN agencies shouldcontinue to invest resources in global estima-tion.18 The argument is made that academicinstitutions have stronger scientific integrity,are better resourced, and are independent ofpolitical pressures.19 However, arguably, the UNagencies can draw on a broad range of scientificexpertise from around the world and have a his-tory of open discussion with countries on theavailability and quality of statistical reporting.20

Indeed, it is essential that in developing esti-mates, the UN system should work on a strongscientific foundation and in a spirit of indepen-dence and objectivity. The Child Health Epide-miology Reference Group,21 which producesestimates of causes of child death, and Count-down to 2015,22 which tracks maternal and

child health indicators, bring together scientistsand UN agencies, but act independently in theirscientific deliberations. A similar mechanism isbeing established for maternal health monitor-ing, bringing together both independent techni-cal experts and also involving UN agencies.23

There are other reasons why the UN shouldnot hand over this role to external bodies. TheUN system has a long history of global, regionaland country-level action. Multilateral agenciesare constituted by – and accountable to – nationalgovernments. This implies a time-unlimited com-mitment that academic institutions by them-selves cannot offer. Moreover, the UN system isbetter positioned than academic institutions togenerate productive interactions between globalmonitoring efforts and country information sys-tems. UN agencies have a responsibility to incor-porate country institutions and scientists into theestimation process. Countries are not simply sup-pliers of raw data to be analysed and adjusted byacademics and technicians. Moving the focusof estimation from institutions in the “north” toactors in the “south” would enable countries tobenefit from capacity-building around the collec-tion and sharing of data, development of scientificmethods of estimation, publication of estimates,and development and application of user-friendlyanalysis and estimation tools.

What should countries do to monitormaternal mortality trends?The debate around global estimates has so focusedthe attention on the maternal mortality ratio thatother indicators of progress have been neglected.Yet a better understanding of trendswould emergefrom an analysis of not only the maternal mor-tality ratio – a measure of obstetric risk – butalso of indicators that reflect the frequency atwhich women are exposed to this risk or levelsof fertility – reflected in the maternal mortalityrate. Valuable insights about maternal health canbe gleaned by examining the interplay betweenmaternal mortality and fertility.In settings where the maternal mortality ratio

(the obstetric risk) is high, the maternal mor-tality rate (the risk per reproductive age woman)may decline due to falling fertility. A decline inthe absolute number of births will result in fewermaternal deaths, even without improvements inthe uptake of maternal health interventions.

125

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

There is also value in tracking the numbers ofdeaths, especially in small countries or wherethe absolute number of maternal deaths is rela-tively small. A simple distribution of numbersof deaths by time of occurrence (during preg-nancy, during the intra-partum period, andpost-partum) provides valuable information forpolicy and programming. Moreover, each deathcan serve as the starting point for audits andconfidential enquiries that describe the circum-stance and causes of each death and identifyways of averting such deaths in the future.24 Inaddition, indicators of health care should bemonitored, such as use of skilled birth attendantat delivery, levels of fertility, nutritional status,and availability and use of essential obstetriccare. These data may well be more readily avail-able on a regular basis than mortality indicators.

Is maternal mortality really difficultto measure or are we just not trying?The reliance on global estimates and the factthat maternal deaths are so poorly counted, aresymptomatic of a broader problem, the neglectof the poorest and most vulnerable in societyand the lack of attention to their health andsurvival – the so-called “scandal of invisibility”.25

It is time for the global community to investseriously in working with countries to generateprimary data on maternal mortality using mea-surement methods that reduce uncertainty andthat generate data on a continuing basis, ratherthan at occasional intervals. This is not just anissue for maternal mortality but applies moregenerally across all causes of death. Good publichealth decision-making and accountability, inorder to make progress towards health targets,are dependent on reliable and timely statisticson births and deaths, including assessment ofcauses of death.

126

The best routine source of data on maternaldeaths is a civil registration system that assuresthe continuous, permanent, compulsory anduniversal recording of the occurrence and char-acteristics of vital statistics, including birthsand deaths, and causes of death.26 Civil regis-tration has a dual purpose, administrative andlegal on the one hand, and statistical, demo-graphic and epidemiological on the other. Forthe individual, the civil statistics records of birthor death provide essential legal documentationfor a wide range of purposes. From a populationperspective, birth and death records can provideimportant public health information. Vital sta-tistics derived from civil registration are the onlynationally representative, continuously avail-able source of information on cause-specificmortality. The record of deaths among womenof reproductive age derived from civil registra-tion is often the first step in conducting a confi-dential enquiry into maternal deaths.The UN Commission for Information and

Accountability forWomen's and Children's Healthis calling on countries and development partnersto prioritize investments for building robusthealth information systems to monitor women'sand children's health, with a focus on the systemsneeded to generate data on births, deaths, andcauses of death.27 The Health Metrics Network28

has launched an initiative to strengthen nationalcivil registration and vital statistics systems inorder to better monitor vital events. Through acombination of global advocacy; developmentof innovative solutions for registering births anddeaths and tracking pregnancy outcomes, andcompiling the lessons learnt from these experi-ences, countries will be empowered to generatetheir own primary data. Perhaps then, the globalcommunity will be able to dispense with statisti-cal estimation models.

References

1. Hogan MC, Foreman KJ,

Mohsen Naghavi AB, et al.Maternal mortality for181 countries, 1980–2008:a systematic analysisof progress towardsMillennium DevelopmentGoal 5. Lancet 2010;375(9726):1609–23. DOI: 10.1016/S0140-6736(10)60518-1.

2. WHO, UNICEF, UNFPA,World Bank. Trends inmaternal mortality: 1990 to2008. Geneva: WHO; 2010.

3. World Health Organization.Maternal mortality in 2005:estimates developed by WHO,UNICEF, UNFPA, and the WorldBank. Geneva: World HealthOrganization; 2005.

4. Horon IL. Underreporting ofmaternal deaths on deathcertificates and the magnitudeof the problem of maternalmortality. American Journal ofPublic Health 2005;95:478–82.

5. Lewis G, editor. TheConfidential Enquiry intoMaternal and Child Health(CEMACH). Saving mothers'

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

lives: reviewing maternal deathsto make motherhood safer.2003–2005. The seventh reporton confidential enquiriesinto maternal deaths in theUnited Kingdom. London,CEMACH, 2007.

6. Gakidou E, King G. Death bysurvey: estimating adultmortality without selectionbias from sibling survivaldata. Demography 2006;43(3):569–85.

7. WHO. International StatisticalClassification of Diseasesand related health conditions.10th revision. Geneva:WHO; 1992.

8. Deneux-Tharaux C, Berg C,Bouvier-Colle MH, et al.Under-reporting of pregnancy-related mortality in the UnitedStates and Europe. Obstetricsand Gynecology 2005;106:684–92.

9. For a fuller discussion ofthese issues, see AbouZahr C.Making sense of maternalmortality estimates.University of QueenslandHealth Information SystemsKnowledge Hub WorkingPaper, 2010 At: <www.uq.edu.au/hishub//?page=147837>.

10. Murray CJL. Towards goodpractice for health statistics:lessons from the MillenniumDevelopment Goal healthindicators. Lancet 2007;369:862–73.

11. United Nations The MillenniumDevelopment Goals Report2010. New York, 2010. At:<www.un.org/millenniumgoals/pdf/MDG%20Report%202010%20En%20r15%20-low%20res%2020100615%20-.pdf>.

12. Graham W, Ahmed S, StantonC, et al. Measuring maternal

mortality: an overview ofopportunities and optionsfor developing countries.BMC Medicine 2008;6:12.DOI:10.1186/1741-7015-6-12.

13. Wilmoth J. Technical paper onmaternal mortality estimation2011. (Forthcoming)

14. Accountability Commission forHealth of Women and Children.At: <www.who.int/topics/millennium_development_goals/accountability_commission/en/>. Accessed30 November 2010.

15. United Nations Summit on theMillennium Development Goals.20–22 September 2010. At:<www.un.org/en/mdg/summit2010/>. Accessed30 November 2010.

16. Graham WJ, Braunholtz DA,Campbell OM. New modelledestimates of maternal mortality.Comment on: Lancet 20108;375(9726):1609–23. In: Lancet2010;375(9730):1963. Authorreply 1968.

17. Soul searching at the UN[Editorial]. Lancet 2010;376(9746):1026.

18. PLoS Medicine Editors. Canwe count on global healthestimates? PLoS Medicine 2010;7(11):e1001–02. DOI:10.1371/journal.pmed.1001002.

19. Murray CJL, Lopez AD.Production and analysis ofhealth indicators: the role ofacademia. PLoS Medicine 2010;7(11):e1001–04. DOI:10.1371/journal.pmed.1001004.

20. Boerma JT, Mathers C,AbouZahr C. WHO and globalhealth monitoring: the wayforward. PLoS Medicine 2010;7(11):e10003–073. DOI:10.1371/journal.pmed.1000373.

21. Child Health EpidemiologyReference Group. At: <http://

cherg.org/main.html>. Accessed10 February 2011.

22. Countdown to 2015. At: <www.countdown2015mnch.org/>.Accessed 10 February 2011.

23. World Health Organization.Meeting of the MaternalMortality EstimationInter-Agency Group andthe Technical AdvisoryGroup on Maternal MortalityEstimates. Geneva: WHO,UNICEF, UNFPA, World Bank;26–27 January 2011.

24. World Health Organization.Beyond the Numbers:Reviewing Maternal Deathsand Complications to MakePregnancy Safer. Geneva: WHO;2010. ISBN 92 4 159183 8 (NLMclassification: WA 900).

25. Setel P, Macfarlane S, Szreter S,et al. A scandal of invisibility:making everyone count bycounting everyone. Lancet2007;370(9598):1569–77.DOI: 10.1016/S0140-6736(07)61307-5.

26. United Nations. Principlesand Recommendations fora Vital Statistics System.Revision 2. New York: UNStatistics Division; 2001.Sales No. 01.XVI.10.

27. WHO. AccountabilityCommission for Health ofWomen and Children. At:<www.who.int/topics/millennium_development_goals/accountability_commission/en/>. Accessed21 February 2011.

28. Health Metrics Networkstreamlines to focus onrecording vital events andtracking progress towardsstronger health informationsystems. At: <www.who.int/healthmetrics/en/>. Accessed21 February 2011.

127

RésuméDeux analyses indépendantes destinées à estimerles niveaux de mortalité maternelle ont étépubliées en 2010, l'une par l'Institute for HealthMetrics and Evaluation à Seattle, États-Unisd'Amérique, l'autre par quatre institutions desNationsUnies (UNICEF, FNUAP, Banquemondialeet OMS). Même si les deux approches sont fondéessur des ensembles similaires de données nationalesempiriques, leurs méthodes statistiques diffèrentsur des points importants, avec des conséquencessur les estimations mondiales, régionales etnationales ainsi obtenues. Cet article examineles divergences, discute de la valeur et deslimitations inhérentes à ces opérations, proposedes interprétations des différentes estimations etsuggère des moyens de mieux adapter cesétudes aux besoins des décideurs nationaux. Ilinvite la communauté internationale à investirvéritablement dans le travail avec les pays envue de créer des données primaires sur lamortalité maternelle au moyen de méthodes demesure qui réduiront l'incertitude et produirontdes données de manière continue. La meilleuresource systématique de données sur les décèsmaternels est un système obligatoire et généraliséde registre d'état civil qui consigne en permanenceles naissances, les décès et les causes des décès. Lacomptabilisation des décès survenus chez lesfemmes en âge de procréer, figurant dans lesregistres d'état civil, est souvent la premièreétape pour réaliser une enquête confidentiellesur la mortalité maternelle et la prévenir.

ResumenEn 2010 se realizaron dos ejercicios independientespara calcular los niveles de mortalidad materna:uno publicado por el Instituto de Métrica yEvaluación en Salud, en Seattle, EE. UU.; el otro,por cuatro organismos de la ONU (UNICEF,UNFPA, el Banco Mundial y la OrganizaciónMundial de la Salud). Aunque ambos enfoquesse basan en similares datos empíricos, susmétodos estadísticos difieren en importantesaspectos, con implicaciones para los consiguientescálculos nacionales, regionales e internacionales.En este artículo se examinan las diferencias, sediscute tanto el valor como las limitacionesinherentes en dichos ejercicios, se proponenmaneras de interpretar los diferentes cálculos yse sugiere cómo lograr que estos ejercicios seanmás pertinentes para las necesidades de laspersonas responsables de tomar decisiones encada país. Se hace un llamado a la comunidadglobal para que colabore con los países a fin degenerar datos principales sobre la mortalidadmaterna empleando métodos de medidas quedisminuyan la incertidumbre y generen datosde manera continua. La mejor fuente rutinariade datos sobre las muertes maternas es unsistema de registro civil que asegure el registropermanente, obligatorio y universal de la incidenciay características de estadísticas vitales, comonacimientos y muertes, así como las causas dedefunción. El registro de muertes de mujeres enedad reproductiva derivado del registro civilsuele ser el primer paso para la investigaciónconfidencial y prevención de lasmuertesmaternas.

C AbouZahr / Reproductive Health Matters 2011;19(37):117–128

128