Casuas Globales de Muerte Materna

8/11/2019 Casuas Globales de Muerte Materna

1/11www.thelancet.com/lancetghPublished online May 6, 2014 http://dx.doi.org/10.1016/S2214-109X(14)70227-X 1

Articles

Global causes of maternal death: a WHO systematic analysisLale Say, Doris Chou, Alison Gemmill, zge Tunalp, Ann-Beth Moller, Jane Daniels, A Metin Glmezoglu, Marleen Temmerman, Leontine Alkema

SummaryBackground Data for the causes of maternal deaths are needed to inform policies to improve maternal health. We developed and analysed global, regional, and subregional estimates of the causes of maternal death during200309, with a novel method, updating the previous WHO systematic review.

Methods We searched specialised and general bibliographic databases for articles published between between Jan 1,2003, and Dec 31, 2012, for research data, with no language restrictions, and the WHO mortality database for vitalregistration data. On the basis of prespecied inclusion criteria, we analysed causes of maternal death from datasets. We aggregated country level estimates to report estimates of causes of death by Millennium Development Goalregions and worldwide, for main and subcauses of death categories with a Bayesian hierarchical model.

Findings We identied 23 eligible studies (published 200312). We included 417 datasets from 115 countriescomprising 60 799 deaths in the analysis. About 73% (1 771 000 of 2 443 000) of all maternal deaths between 2003 and2009 were due to direct obstetric causes and deaths due to indirect causes accounted for 275% (672 000, 95% UI197375) of all deaths. Haemorrhage accounted for 271% (661 000, 199362), hypertensive disorders 140%(343 000, 111174), and sepsis 107% (261 000, 59186) of maternal deaths. The rest of deaths were due to abortion(79% [193 000], 47132), embolism (32% [78 000], 1855), and all other direct causes of death (96% [235 000],65143). Regional estimates varied substantially.

Interpretation Between 2003 and 2009, haemorrhage, hypertensive disorders, and sepsis were responsible for morethan half of maternal deaths worldwide. More than a quarter of deaths were attributable to indirect causes. Theseanalyses should inform the prioritisation of health policies, programmes, and funding to reduce maternal deaths atregional and global levels. Further efforts are needed to improve the availability and quality of data related to maternalmortality.

Funding USAID, the US Fund for UNICEF through a grant from the Bill & Melinda Gates Foundation to CHERG, andThe UNDP/UNFPA/UNICEF/WHO/The World Bank Special Programme of Research, Development, and ResearchTraining in Human Reproduction (HRP), Department of Reproductive Health and Research.

Copyright 2014 World Health Organization; licensee Elsevier. This is an Open Access article published without anywaiver of WHOs privileges and immunities under international law, convention, or agreement. This article shouldnot be reproduced for use in association with the promotion of commercial products, services, or any legal entity.There should be no suggestion that WHO endorses any specic organisation or products. The use of the WHO logois not permitted. This notice should be preserved along with the articles original URL.

IntroductionAn estimated 287 000 maternal deaths occured worldwidein 2010, most of which were in low-income and middle-income countries and were avoidable.1 Reduction ofmaternal mortality has long been a global health priorityand is a target in the UN Millennium Development Goals(MDG) framework2 and a key concern of the GlobalStrategy for Womens and Childrens Health launched bythe UN Secretary-General in September, 2010.3 To reach thetarget of the fth MDG, a 75% decrease in maternalmortality ratio (the number of maternal deaths per 100 000livebirths) between 1990 and 2015 is needed. Some progresstowards this target has been reported, especially in the pastdecade,1,46 but further improvements are needed.

A key requirement for further advances in reduction ofmaternal deaths is to understand the causes of deaths foreffective policy and health programme decisions. Thedenition of maternal mortality (the death of a woman

whilst pregnant or within 42 days of delivery or terminationof pregnancy, from any cause related to, or aggravated bypregnancy or its management, but excluding deaths fromincidental or accidental causes7) allows the identicationof maternal deaths on the basis of their causes, as eitherdirect or indirect. However, collection of routine andcomplete information about causes of maternal death hasnot been possible because of inadequacies of datacollection and absence of vital registration systems inmost countries. The rst systematic analysis of all availablepublished scientic literature and government reports oncauses of maternal death was published in 2006, andprovided an overall picture of the contribution of differentcauses to the burden of maternal deaths. In 2006, wereported haemorrhage and hypertensive disorders as theleading causes of maternal mortality in developingregions.8 More recently, the Global Burden of Disease(GBD) study9 provided estimates of maternal causes of

Lancet Glob Health 2014

PublishedOnlineMay 6, 2014http://dx.doi.org/10.1016/S2214-109X(14)70227-X

UNDP/UNFPA/UNICEF/WHO/The World Bank SpecialProgramme of Research,Development and ResearchTraining in HumanReproduction (HRP),Department of ReproductiveHealth and Research, WorldHealth Organization, Geneva,Switzerland (L Say MD,D Chou MD, A Gemmill MPH, Tunalp MD, A-B Moller MSc,A M Glmezoglu MD,M Temmerman MD);Department of Demography,University of California,Berkeley, CA, USA(A Gemmill);Clinical Trials Unit, College ofMedical and Dental Sciences,University of Birmingham,Birmingham, UK(J Daniels PhD); andDepartment of Statistics andApplied Probability and SawSwee Hock School of PublicHealth, National University ofSingapore, Singapore(L Alkema PhD)

Correspondence to:Dr Lale Say, UNDP/UNFPA/UNICEF/WHO/The World BankSpecial Programme of Research,Development and ResearchTraining in Human Reproduction(HRP), Department ofReproductive Health andResearch, World HealthOrganization, Geneva,[email protected]
http://crossmark.crossref.org/dialog/?doi=10.1016/S2214-109X(14)70227-X&domain=pdf


2/11

Articles

2 www.thelancet.com/lancetghPublished online May 6, 2014 http://dx.doi.org/10.1016/S2214-109X(14)70227-X

death for the main direct causes as part of the analysis of

all causes of death.We analysed global, regional, and subregional estimatesof the causes of maternal death during 200309, with anovel method. This period was chosen to avoid overlapwith the previous review that covered 19982002.8 Thestudy period did not include reported deaths from morerecent years to ensure increased comparability acrosscountries; more recent data were not available for mostcountries, especially within the WHO mortality databasethat includes vital registration datasets made available bythe countries. We also elaborated for the rst time furtherbreakdown of main cause of death categories, and providedcause of death estimates for disorders that are clinicallyimportanteg, antepartum and postpartum haemorrhage.

MethodsSearch strategy and selection criteriaWe used the International Classication of Diseases(ICD, 10th edition) denition of maternal mortality,7 andincluded maternal deaths reported during 200309, andgenerated regional estimates for the ten MDG regions.10 We searched for data for causes of maternal death fromtwo distinct sources. We used vital registration datasetsfrom the WHO mortality database, made available bycountries.11 We deemed vital registration data as goodquality if the completeness of death registration in thepopulation older than 5 years was more than 85% andthe proportion of ill-dened causes of death (coded asR99) were less than 20%. Details of how we assessedcompleteness and coverage of vital registration data byWHO mortality statistics are described elsewhere.1

We also did a literature search of bibliographic

databases by adapting a previously described searchstrategy12 (appendix). Two reviewers (ABM and DC)initially screened the citations identied by the searcheson the basis of their titles and abstracts. The full text ofthe article was obtained if both reviewers judged acitation as potentially eligible and a third reviewer (JPD)adjudicated on discrepant opinions. A second round ofscreening of the full reports was done in the same way.Additionally, we identied government reports includingcause-of-death information by hand searching WHOregional databases, websites of Ministries of Health andNational Statistical Offi ces, and archives of relevantreports received by WHO. We considered studiesidentied through the literature search and governmentalreports for inclusion if they reported data for the causesof maternal mortality between Jan 1, 2003, and Dec 31,2009. We excluded studies that contained data before2003 that could not be disaggregated from after 2002data and with midpoint of the data collection periodbefore 2003. In line with the 2006 review, we alsoexcluded studies reporting fewer than 25 deaths or fewerthan four major categories of death.8 Lastly, we excludedstudies in which more than 25% of deaths did not have acause assigned. We considered subnational studies forinclusion only if investigators explicitly reportedmethods and if the maternal deaths that they reportedwere deemed to be representative of the population theypertained to. We included studies from health facilitiesor institutions only if the institutional birth rate wasgreater than or equal to 50% in that setting during200309. The institutional birth rate was based on thenational country reported gure derived from sourcessuch as Demographic and Health Surveys, MultipleIndicator Cluster Surveys, and national health statisticalreports.

If different data sources overlapped for a countryperiod, we included only one data source, and nationaldata took precedence over data from subnational levels inthe following order: national enquiries, vital registration,and nationally representative surveys of maternal deaths.Thus, although countries might have had only onesource of data for any particular year, more than onesource might have been included in the entire studyperiod 200309.

Data extraction and classication of maternal deathsData extraction from studies and reports was done by onereviewer (ABM) and independently checked on the formby another (JPD or DC). ICD-10 codes were used to classifycauses of maternal death (appendix). Data using ICD-9codes were converted to ICD-10 codes with the WHOICD-10 Translator.13 We excluded vital registration datafrom Russia, Kazakhstan, Belarus, and Ukraine reportedwith ICD 10-1 because accurate assignment of deaths tothe equivalent ICD-10 codes or analytical pregnancy orobstetric categories was not possible. We assigned

SeeOnline for appendix

Figure : Study prole

338 datapoints from 79countries (vital registration)

56 datapoints from 32countries (governmentreports, special surveys,and condential enquiries)

Bibliographic databases

23 datapoints from 11 countries

11051 citations (200308)435 titles for review



62378 deaths reported in 417 datapoints from 115 countries

60 799 deaths reported in 417 datapoints from 115 countries

1579 deaths excluded because ofunascertained cause of death


3/11

Articles

www.thelancet.com/lancetghPublished online May 6, 2014 http://dx.doi.org/10.1016/S2214-109X(14)70227-X 3

research study data equivalent ICD-10 codes by matching

the closest diagnosis. When data presented wereambiguous, contradictory, or could not be disaggregated,we tried to contact the author for clarication. If this wasnot successful, we used the consensus view of two of theauthors (JPD, DC). Maternal deaths assigned to unknownas a cause of death were excluded from the analysis.

For analysis purposes, we grouped maternal causes ofdeath into seven main categories of direct and indirectcauses: abortion, embolism, obstetric haemorrhage,hypertensive disorders, pregnancy-related sepsis, otherdirect causes, and indirect causes. The abortion categoryincludes induced abortion, miscarriage, and ectopicpregnancy. We dened the category of indirect maternaldeaths in line with the WHO application of ICD-10(ICD-MM) to deaths during pregnancy, childbirth, and thepuerperium. 14 ICD-MM was published to enable astandardised grouping of causes of death and to avoidpresentation of highly aggregated data in differingclassication groups, complicating the task of datacomparability. The broad ICD-MM categories were furthersubdivided. The category haemorrhage was divided intosubcategories of antenatal, intrapartum, and postpartumhaemorrhage; other direct causes were subcategorised intocomplications of delivery, obstructed labour, and all otherdirect causes. Indirect causes of death were subcategorisedinto medical disorders, HIV-related maternal deaths, andall other indirect causes.

Statistical analysisFor every country, we estimated the causes of deathdistribution on the basis of country-specic data (ifavailable) and the regional causes of death distributionwith a Bayesian hierarchical model. We estimated allcauses and subcauses hierarchically except for theproportion of HIV/AIDS indirect deaths, which wemodelled separately because of the dependence of theproportion on the severity of the HIV/AIDS epidemic inthe country.

To construct country-specic HIV/AIDS-removed causesof death distribution, we divided countries into threecategories on the basis of data availability and data quality.We divided countries with recorded causes of deathdistribution into categories A and B, and combinedcountries without any data in group C.15 Category Aincluded all countries with good quality and complete vitalregistration data, where the sample of maternal deaths (forwhich the recorded causes of death distribution wasobtained) was deemed to be representative of the totalnumber of maternal deaths for the country during theperiod of interest. For countries in category B for whichvital registration data might have been available but notconsidered good quality, we assumed that the estimatedcauses of death distribution from the recorded sampleswere not necessarily representative of the causes of deathdistribution of all maternal deaths in the period of interest(appendix p 7).

Figure : Estimates for main causes of death worldwidePoint estimates are shown by bars (and squares) and 95% uncertaintyintervals are shown by the horizontal lines. CoDD=cause of death distribution.ABO=abortion. EMB=embolism. HEM=haemorrhage. HYP=hypertension.SEP=sepsis. DIR=direct causes. IND=indirect causes. Dev=developed regions.N Africa=northern Africa. SSA=sub-Saharan Africa. E Asia=eastern Asia.S Asia=southern Asia. SE Asia=southeastern Asia. W Asia=western Asia.CC Asia=Caucasus and central Asia. LAC=Latin America and the Caribbean.

Oceania

LAC

CC Asia

W Asia

SE Asia

S Asia

E Asia

SSA

N Africa

Dev

INDDIRSEPHYPHEMEMBABO

A Global CoDD

B Regional CoDDs

0 5 10 15 20 25 30 35 40%

ABO HEM SEPEMB HYP DIR

IND


4/11

Articles


We used the Bayesian hierarchical model for the HIV/

AIDS-removed cause of death distribution to exchangeinformation about these distributions between countries.With this method, the estimates in countries with scarceinformation were informed by the typical experience inthe region, which is the HIV/AIDS-removed causes ofdeath distribution that is unweighted by the country-specic total maternal deaths envelopes in the region.Additionally, in the hierarchical model, typical regionalpatterns in groups with scarce information were informedby patterns in other regions. For countries in group C, thecause of death distribution estimates were given by thetypical regional estimates, and we assessed uncertainty onthe basis of the estimated variability in, and correlationstructure of, country-specic causes of death distributionwithin regions. For countries in group A with good qualityvital registration data, the cause of death distribution wasestimated from the recorded causes of death distribution,accounting for stochastic uncertaintyie, uncertainty thatarises when dealing with small numbers of observeddeaths. For example, if only ve deaths were recorded in acountry, the underlying true cause of death distribution isuncertain. The extent of uncertainty in cause of deathdistribution for group A countries ranged from beingnegligible for countries where a large number of deathswere observed to substantial uncertainty for countrieswhere causes of death for only a small number of deathswere recorded (eg, Belgium, where the causes of deaths

for only ve maternal deaths were recorded). For countries

with substantial uncertainty on the basis of a smallnumber of deaths, the hierarchical model informed thecountry-specic estimate, whereas for countries with lessuncertainty, estimates were more data-driven and lessinformed by the hierarchical model. For countries incategory B (where the recorded cause of death distributionswere available for a subset of maternal deaths from vitalregistration or survey data), we assumed that the estimatedcauses of death distributions from the recorded sampleswere not necessarily representative of the causes of deathdistributions of all maternal deaths in the period ofinterest. For these countries, we assessed uncertainty inthe cause of death distribution estimates for all maternaldeaths without recorded causes on the basis of estimatedvariability in, and correlation structure of, country-speciccause of death distributions within regions. Therefore, thenal estimated cause of death distributions for countriesin group B accounted for stochastic uncertainty andadditional uncertainty in the cause of death distributionsfor the unrecorded subset of maternal deaths.

The estimation of the proportion of HIV/AIDSmaternal deaths was based on the approach used in theestimation of the total number of maternal deaths.1,16 This approach provides country-specic estimates for theproportion of HIV/AIDS maternal deaths among allAIDS deaths to women of reproductive ages. Thesecountry-specic estimates were combined with estimates

Abortion Embolism Haemorrhage Hypertension Sepsis Other direct causes Indirect causes

N % (95% UI) N % (95% UI) N % (95% UI) N % (95% UI) N % (95% UI) N % (95% UI) N % (95% UI)

Worldwide 193 000 79%(47132)

78 000 32%(1855)

661 000 271%(199362)

343 000 140%(111174)

261 000 107%(59186)

235 000 96%(65143)

672 000 275%(197375)

Developed regions 1100 75%(57116)

2000 138%(101220)

2400 163(111246)

1900 129%(100168)

690 47%(24111)

2900 200%(166275)

3600 247%(195339)

Developing regions 192 000 79%(47132)

76 000 31%(1754)

659 000 271%(199364)

341 000 140%(111174)

260 000 107%(59187)

232 000 96%(64143)

668 000 275%(197376)

Northern Africa 490 22%(0949)

720 32%(0989)

8300 369%(241516)

3800 169%(119229)

1300 58%(23129)

3800 171%(77308)

4000 180%(95302)

Sub-Saharan Africa 125 000 96%(51172)

27 000 21%(0845)

321 000 245%(169341)

209 000 160%(11721)

134 000 103%(55185)

119 000 90%(51157)

375 000 286%(199403)

Eastern Asia 420 08%

(0220)

6500 115%

(16406)

20 000 358%

(109682)

5900 104%

(39202)

1500 26%

(0497)

8000 141%

(20513)

14 000 249%

(64588)Southern Asia 47 000 59%

(15173)17 000 22%

(0568)238 000 303%

(140548)80 000 103%

(58166)107 000 137%

(33359)65 000 83%

(33177)229 000 293%

(122551)

Southeastern Asia 11 000 74%(28184)

18 000 121%(32334)

44 000 299%(152513)

21 000 145%(84227)

8100 55%(18150)

20 000 138%(56312)

25 000 168%(78342)

Western Asia 860 30%(1076)

2600 92%(33226)

8900 307%(174491)

3900 134%(75212)

1400 48%(15131)

4500 156%(66337)

6700 234%(113431)

Caucasus andcentral Asia

250 46%(2782)

590 109%(62182)

1200 228%(172303)

790 147%(116183)

460 85%(57136)

910 168%(126232)

1200 218%(162299)

Latin Americaand Caribbean

6900 99%(81130)

2300 32%(2647)

16 000 231%(197278)

15 000 221%(199246)

5800 83%(56125)

10 000 148%(117194)

13 000 185%(156226)

Oceania 290 71%(12229)

610 148%(19476)

1200 295%(85617)

560 138%(49258)

200 50%(06185)

510 124%(23387)

710 174%(47443)

Data shown are the estimated proportion of cause of death (%) with 95% uncertainty interval (95% UI).

Table : Distribution of causes of deaths by Millennium Development Goal regions


5/11

Articles


of the number of AIDS, maternal, and total deaths to

women of reproductive ages to obtain an initial meanestimate for the proportion of HIV/AIDS maternaldeaths. These estimates were updated with country-specic data on the proportion of HIV/AIDS deaths.

We used a Markov Chain Monte Carlo algorithm togenerate samples of the posterior distributions of allmodel parameters, including the country-specic causesof death distributions. Point estimates for proportionswere given by the posterior means of the proportion and95% uncertainty intervals were given by the 25th and975th percentiles of the posterior distributions. Wecalculated the resulting distribution for each region fromthe regional weighted averages of the estimated country-specic cause of death distributions in the region.Weights were based on the estimated number of maternaldeaths for each country for the year 2005 (which is theestimation year closest to the midpoint of the studyperiod).1 The Markov Chain Monte Carlo samplingalgorithm was implemented in R (version 3.0.1) and JAGS(version 3.3.0).17,18

Role of the funding sourceThe funder of the study had no role in study design, datacollection, data analysis, data interpretation, or writing ofthe report. The corresponding author had full access toall the data in the study and had nal responsibility forthe decision to submit for publication.

ResultsFigure 1 summarises the identication and selection ofdata for incorporation into the model. We included vitalregistration data from 79 countries, covering 338 country-year datapoints during 200309, of which 263 (78%) camefrom 58 countries with a good quality registration system.Of the 50 179 citations identied from bibliographicdatabases, 23 studies from 11 countries met the inclusioncriteria. Government reports, such as specialised statisticaltabulations and surveillance documents, were available fora further 32 countries. These reports included condentialenquiries, RAMOS, or specialised maternal mortalitysurveys. Four countries (France, UK, South Africa, andMexico) produced condential enquiries or reports ofenhanced surveillance systems covering 12 country-years,which were judged better than vital registration data forthat country. 26 datasets were informed by verbal autopsy;however, only ten of these datasets specied the verbalautopsy instrument used. The appendix shows furtherdetails of source of data by country and the studiesproviding the datasets.

All included data sources combined provided417 datapoints from 115 countries and reported62 378 deaths. Of these, we excluded 1579 deaths (25%),almost exclusively from studies and governmentalreports, because no main cause of death could beascertained (appendix). Of the 60 799 maternal deathsincluded in the nal database, 50% came from vital

registration data, 29% from sub-Saharan Africa, and 2%

from southern Asia. Considering the total estimatednumber of maternal deaths over 7 years, the study datarepresents 2.5% of all maternal deaths in that period.4

Figure 2 shows the regional and global estimates ofdistribution of causes of death. Nearly 73% of all maternaldeaths between 2003 and 2009 were due to direct obstetriccauses whereas deaths due to indirect causes accountedfor 275% (95% UI 197375) of all deaths from knowncauses (table 1). Haemorrhage was the leading directcause of maternal death worldwide, representing 271%(199362) of maternal deaths. More than two thirds ofreported haemorrhage deaths were classied aspostpartum haemorrhage (table 2). Hypertension was thesecond most common direct cause worldwide (140%,111174). Maternal mortality due to sepsis was 107%(59186), abortion accounted for 79% (47132), andembolism and other direct causes accounted for theremaining 128% of global deaths.

Table 3 shows the other direct causes of maternalmortality. Complications of delivery were responsiblefor 28% (1649) and obstructed labour for 28%(1455) of all maternal deaths worldwide, bothreported within the other direct category, whichaccounted for 96% of all maternal deaths worldwide.Further breakdown of deaths due to indirect causessuggests that more than 70% of indirect causes are

Antepartum Intrapartum Postpartum Haemorrhage total

N %(95% UI)

N %(95% UI)

N %(95% UI)

N %(95% UI)

Worldwide 158 000 65%(4396)

23 000 09%(0422)

480 000 197%(129289)

661 000 271%(199362)

Developedregions

700 48%(3379)

510 35%(16111)

1200 80%(47155)

2400 163%(111246)

Developingregions

157 000 65%(4396)

23 000 09%(0422)

479 000 197%(12929)

659 000 271%(199364)

NorthernAfrica

720 32%(1562)

380 17%(0368)

7200 320%(189473)

8300 369%(241516)

Sub-SaharanAfrica

110 000 84%(50137)

12 000 09%(023)

200 000 152%(86251)

321 000 245%(169341)

Eastern Asia 3800 66%

(16174)

210 04%

(0117)

16 000 287%

(61639)

20 000 358%

(109682)Southern Asia 30 000 38%

(1585)3400 04%

(0115)205 000 261%

(104514)238 000 303%

(140548)

SoutheasternAsia

7000 47%(20107)

3100 21%(0387)

34 000 231%(94461)

44 000 299%(152513)

Western Asia 1700 60%(29117)

710 25%(04104)

6400 222%(101415)

8900 307%(174491)


280 52%(3579)

230 42%(16107)

720 134%(94198)

1200 228%(172303)


4000 58%(4578)

2900 41%(2190)

9200 133%(109164)

16 000 231%(197278)

Oceania 200 48%(10138)

76 18%(01113)

940 229%(41578)

1200 295%(85617)

Percentages shown are the subgroup as a proportion of all deaths for that region in the input dataset.

Table : Subgroup analysis of haemorrhage deaths by Millennium Development Goal region


6/11

Articles


from pre-existing disorders, including HIV, when

exacerbated by pregnancy (table 4). HIV alone accountedfor 55% (3876) of global maternal deaths.The global distribution was affected by the two regions,

sub-Saharan Africa and southern Asia, that accounted for838% of all maternal deaths. Although estimated regionalcause of death distributions are quite uncertain for manycauses, point estimates show substantial differencesacross regions (table 1 and gure 3). Haemorrhageaccounted for 369% (241516) of deaths in northernAfrica, but only for 163% (111246) in developedregions. Hypertensive disorders were a particularlyimportant cause of death in Latin American and theCaribbean, contributing to 221% (199246) of allmaternal deaths in the region. Almost all sepsis deathswere recorded in the developing countries, and theproportion of such deaths was highest at 137% (33359)in southern Asia.

Only a small proportion of deaths are estimated toresult from abortion in eastern Asia (08%, 0220),where access to abortion is generally less restricted. LatinAmerica and the Caribbean, and sub-Saharan Africahave a higher proportion of deaths in this category thanthe global average; 99% (81130) and 96% (51172),respectively. Another direct cause, embolism, accountedfor more deaths than its global average in southeasternAsia (121%, 32334) and eastern Asia (115%, 16406).

The proportion of deaths due to indirect causes was

highest in southern Asia (293%, 122551), followed bysub-Saharan Africa (286%, 199403). Indirect causesalso accounted for nearly a quarter of deaths in thedeveloped regions. The overall proportion of HIVmaternal deaths is highest in sub-Saharan Africa, 64%(4688%). The appendix shows estimates for country-specic cause of death distributions.

DiscussionThis systematic analysis suggests that indirect causes andhaemorrhage are the largest causes of maternal deathworldwide. Of the direct causes of death, haemorrhagewas the leading cause of maternal death, followed byhypertensive disorders and sepsis. Regional estimatesvaried substantially.

We scanned and included many data sources includinggovernment reports and peer-reviewed scienticliterature (panel). Because of the paucity of data, all datafor a country were aggregated during the 7-year periodand model-based estimates were constructed for thelarge subset of countries without any information abouttheir causes of death distributions. This approach differsfrom the previous WHO systematic review8 in whichrecorded country-specic cause of death distributionswere weighted by the number of maternal deaths in thecountry to obtain regional estimates. The new approachwas implemented to overcome the drawback of theprevious study that a recorded cause of death distributionbased on a small sample size in a country with a largenumber of maternal deaths could unduly affect theregional estimates of the cause of death distribution. Inthe estimation method applied in this study, country-specic estimates of cause of death distribution wereinformed by the available data in the country and theregional average cause of death distribution, which canbe regarded as a typical pattern for the region (unweightedby the total maternal death envelopes of the countries inthe region) through a Bayesian hierarchical model. Theaccuracy of the regional and global estimates and 95%uncertainty intervals were validated through two out-of-sample validation exercises and suggested satisfactorymodel performance (appendix).

The different analytical approaches of the previousWHO review published in 2006 and the present analysislimit our ability to make comparisons between thendings of both. Furthermore, the limitations of thedataset and the methods used in the previous study didnot allow for generation of a worldwide cause-of-deathestimate, and only estimates for large world regions werecalculated. However, some of the region-specic trendsreported in the previous analysis also seem to be foundin the present study. These include, for example, thehighest share of haemorrhage deaths in Asian regions,the particular importance of hypertensive disorders ofpregnancy in Latin America and the Caribbean, and theimportance of indirect causes in sub-Saharan Africa.

Complicationsof delivery

Obstructed labour Other Other directcauses total

N %(95% UI)

N %(95% UI)

N %(95% UI)

N %(95% UI)

Worldwide 68 000 28(1649)

69 000 28(1455)

98 000 4(2275)

235 000 96(65143)

Developedregions

760 52(3790)

94 06(0317)

2100 141(118209)

2900 20(166275)

Developingregions

67 000 28(1549)

69 000 29(1455)

96 000 39(2174)

232 000 96(64143)

NorthernAfrica

1600 73(29159)

210 09(0233)

2000 88(26209)

3800 171(77308)

Sub-SaharanAfrica

43 000 33(1567)

28 000 21(0752)

48 000 37(1486)

119 000 9(51157)

Eastern Asia 250 04

(0114)

6900 123

(09501)

770 14

(0258)

8000 141

(2513)Southern Asia 14 000 18

(0452)21 000 27

(0585)30 000 38

(09103)65 000 83

(33177)

SoutheasternAsia

3000 21(0566)

9400 64(14206)

7800 53(13173)

20 000 138(56312)

Western Asia 2000 71(24179)

320 11(0242)

2200 75(2232)

4500 156(66337)


400 73(52109)

47 09(0419)

460 86(56144)

910 168(126232)


2300 33(2649)

3300 48(3283)

4700 67(48103)

10 000 148(117194)

Oceania 95 23(0389)

210 51(03255)

210 51(04235)

510 124(23387)


Table : Subgroup analysis of other direct causes of death by Millennium Development Goal region


7/11

Articles


Ultimately, the results of this analysis are constrained

by the accuracy of the data included. Cause of death dataare especially diffi cult to analyse because of inadvertenterrors such as misclassication and misinterpretation ofcause of death coding rules, or omissions or incorrectentries because of the nature of some of the disordersleading to maternal deaths such as abortion. Werecognise the limitations due to the particular issue ofmisclassication within maternal death certication.

Although algorithms have been developed to adjust formisclassication of deaths in so-called garbage codes,32 such algorithms are not able to improve on misclassicationwithin maternal cause categories or the under-reporting ofdeaths of specic maternal causes. In view of the range ofclassication and under-reporting issues, and the limitedinformation available to accurately adjust data, we opted toreport aggregated cause of death fractions as reported ingood vital registration or as predicted by the model basedon unadjusted data. Therefore, resulting estimates shouldbe interpreted as the estimates for the reported cause ofdeath distribution.

An alternative estimation method was used morerecently in the GBD Study 2010 to obtain estimates forthe all-cause cause of death distribution, whereby a subsetof misclassication issues were accounted for.9,33 TheGBD broad categorisation of causes of maternal deathdiffered from our approach, with important differencesfor some categories. For example, the GBD did not use acategory of indirect causes. Likewise, we includedobstructed labour as a subcategory of other direct causes,in line with the ICD-MM, whereas the GBD studyregarded it as a main category.9 Inclusion of variousdisorders within identied categories also differed. Forexample, long labour is included in the obstructed labourcategory in our analysis, but not in the GBD study.Nevertheless, broad agreement exists between our globalestimates and the GBD estimates (appendix).

However, estimates on maternal causes of death shouldbe viewed with caution. Although future research onimproved modelling approaches to deal with mis-classication errors might lead to improved estimates,the absence of reliable data is a more pressing issue thatdemands increased prioritisation.

Recommendations for policy and practiceOur results show two main concerns for policy andpractice related to data availability and quality in thecountries. First, where data are most needed, data areoften not available, which is unfortunately the case insome countries with high mortality where estimateswere obtained on the basis of modelling. In thisanalysis, India and Nigeria together accounted for athird of global maternal deaths, but only one datasetmet criteria for inclusion (India). Of the ten countrieswith the highest maternal mortality ratio in 2010, datawere available only for one, Cameroon. Moreover, only5% of all deaths included in the analysis were from

southern Asia where the second highest number of

maternal deaths were recorded. This means that causeof death distribution in a region is affected by thecountries that have data within that region. Althoughthe distribution is expected to be similar acrosscountries within a region, for some conditions whereavailability of interventions signicantly vary becauseof structural factors that are also highly contextual,such as the legal status of abortion, differences areexpected.

Second, where data are available, they are oftenincomplete. For example, we noted that indirect deathsaccounted for 275% (197375) of deaths, althoughthe actual indirect causes were not well delineated inmore than a fth of the reported indirect maternaldeaths. Although it might be diffi cult to establish withcertainty whether a womans pregnancy aggravated apre-existing medical disorder, or if their interactionresulted in her death, improved documentation on thesequence of events is paramount. For instance,differentiation between indirect maternal deaths due toHIV and direct maternal deaths in HIV-positive womenis important; this difference would have implications atboth clinical and programmatic levels. Acceleratedaction is needed to improve data acquisition and quality,especially relating to correct attribution of cause ofdeath information.

HIV-related Pre-existing medicalconditions

Other indirectcauses

Indirect causestotal

N %(95% UI)

N %(95% UI)

N %(95% UI)

N %(95% UI)

Worldwide 134 000 55%(3876)

361 000 148(92234)

177 000 72%(35146)

672 000 275%(197375)

Developedregions

400 27%(1051)

3000 203(161291)

250 17%(0944)

3600 247%(195339)

Developingregions

133 000 55%(3877)

358 000 148(91235)

177 000 73%(35147)

668 000 275%(197376)

NorthernAfrica

760 34%(1164)

2800 124(53241)

500 22%(0672)

4000 180%(95302)

Sub-SaharanAfrica

84 000 64%(4688)

168 000 128(70223)

122 000 93%(46184)

375 000 286%(199403)

Eastern

Asia

2200 39%

(03106)

12 000 207

(35544)

130 02%

(0011)

14 000 249%

(64588)Southern Asia 37 000 48%

(12102)143 000 182

(61419)49 000 63%

(05253)229 000 293%

(122551)

SoutheasternAsia

5900 40%(1483)

17 000 118(41287)

1400 10%(0235)

25 000 168%(78342)

Western Asia 1200 42%(1584)

4900 169(65365)

650 22%(0581)

6700 234%(113431)

Caucasus andCentral Asia

130 23%(1041)

920 169(119247)

140 25%(1451)

1200 218%(162299)


1300 18%(0930)

9800 140(117176)

1800 26%(1945)

13 000 185%(156226)

Oceania 170 42%(05111)

500 123% (19382)

36 09%(0144)

710 174%(47443)


Table : Subgroup analysis of indirect causes of death by Millennium Development Goal region


8/11

Articles


We included several sources of cause of death data, the

quality of which depends on who completed the deathcerticate or verbal autopsy, who interpreted theinformation from the death certicate or verbal autopsy,and whether medical records were available for review toconrm or revise the ascertained cause of death.Condential enquires and special maternal death reviewscan provide such information, but the feasibility of doingsuch detailed reviews is very restricted.

To support data acquisition and quality needs, revisionsare being made to the standard verbal autopsy instrumentto increase the feasibility of its implementation wherecause of death attribution is possible only by thosemeans. Furthermore, the ICD-MM will standardisedocumentation and analysis related to maternal causesof death and their attribution to direct and indirectcauses.14 Discrepancies exist in how some deaths arecategorised. For example, suicide in some contexts isregarded as coincidental whereas in other settings itmight be reported within direct or indirect maternaldeaths. Maternal suicides are known to happen in thecontext of undesired pregnancy, inability to accessabortion, and postpartum depression or psychosis. ICD-MM suggests that maternal suicides will be includedwithin the direct category of maternal death. With theprocess for the 11th revision of the ICD well underway,34 one can anticipate the possibility for improved granularityof data. But one must also recognise the responsibility ofthe certifying professional to provide accurate and usefulinformation for improved epidemiological monitoringand assessment to inform policies with the best availableevidence. Calls for inclusion of training on cause of deathcertication and the use of ICD use within medicalcurricula are well founded and should be supported.Still, these calls for better data need to acknowledge therealities in establishing, with accuracy, what the cause ofdeath was at time of certication. This need for accuracyis especially important in relation to identication ofindirect maternal deaths that aim to establish theaggravating effect between the physiological effects ofpregnancy and another disease.

With regard to clinical implications, we nd that, despiteestablished interventions to prevent and treat postpartumhaemorrhage (eg, active management of the third stage oflabour35), haemorrhage remains the leading individualcause of death. With available data, it is not possible toestablish whether the persistence of haemorrhage as theleading cause of death despite effective interventions isthe result of a failure to implement such interventions,whether there is a shift towards antepartum haemorrhageor a shift in delivery practice such as increasing rates ofcaesarean sections, or whether misclassications withregard to abortion and obstructed labour are erroneouslyincreasing the haemorrhage category.

Further analysis to elucidate the separate contribution ofantepartum and postpartum causes will have importantimplications for the planning and implementation of policy

Figure :Estimates for main causes of death by regionPoint estimates are shown by bars (and squares) and 95% uncertainty intervalsare shown by the horizontal lines. ABO=abortion. EMB=embolism.HEM=haemorrhage. HYP=hypertension. SEP=sepsis. DIR=direct causes.IND=indirect causes. Dev=developed regions. N Africa=northern Africa.SSA=sub-Saharan Africa. E Asia=eastern Asia. S Asia=southern Asia.

SE Asia=southeastern Asia. W Asia=western Asia. CC Asia=Caucasus and centralAsia. LAC=Latin America and the Caribbean.

IND

DIR

SEP

HYP

HEM

EMB

ABO

0 5 10 15 20 25 30 35 40%

Comparison of causes across regions

DevN AfricaSSAE AsiaS AsiaSE AsiaW AsiaCC AsiaLACOceania


9/11

Articles


and programmes, because interventions to address themare very different. Although the international communityhas rightly focused on postpartum haemorrhage, andspecically atonic postpartum haemorrhage, it is now theappropriate time to unpack the obstetric haemorrhagecategory. The improved method used in this analysisallowed delineation of haemorrhage deaths, reporting that

about 240% of all haemorrhage deaths happened duringpregnancy, and the remainder in the intrapartum orpostpartum period. Thus we provide for the rst time anevidence-based estimation of the proportion of maternaldeaths due to antepartum haemorrhage.

Also alarming is the proportion of deaths attributed tohypertensive disorders, which are the second highest

Panel: Misclassication and underreporting issues by cause of death

AbortionWe estimated that 79% (95% UI 47132) of all maternal deaths were due toabortion. This nding is lower than the previous assessments, which estimatedmortality due to unsafe abortion at 13%.19,20 Classication of maternal deathsdue to abortion, and more specically unsafe abortion, is associated with a riskof misclassication, which might lead to underreporting. Even where inducedabortion is legal, religious and cultural perceptions in many countries mean thatwomen do not disclose abortion attempts and relatives or health-careprofessionals do not report deaths as such. Under-registration of deaths mightbe the result of stigmatisation of abortion affecting what information isreported by relatives and informants or intentional misclassication byproviders when abortion is restricted.21

In these circumstances, the overall number of maternal mortality might not beaffected, whereas abortion-related deaths might be particularly underestimatedbecause of this under-reporting. Although these abortion-related deaths might beclassied mainly into sepsis and haemorrhage, this might over-simplify thecomplexity of death reporting. An analysis22 comparing InternationalClassication of Diseases 10 (ICD-10) codes for underlying cause of death with theremainder of information about the death certicate and verbal autopsy in ruralMexico found that deaths due to second trimester abortion were misclassiedinto both maternal and non-maternal deaths. Examples of misclassicationincluded assigning underlying cause of death from amniotic uid embolism tocerebral anoxia, rather than abortion, either induced or spontaneous. Validationstudies like this provide needed insight into the quality and accuracy of maternalmortality data. However, short of reviewing every death certicate and medicalrecord after the death of a woman aged 1549 years, the study further highlights

the diffi culty associated with considering adjustments to account for this type ofmisclassication, or indeed any misclassication. Validation studies can identifypatterns of systematic or unbalanced misclassication, but the validity ofapplication of adjustment parameters derived by verbal autopsy data from onelocation to another, and application of factors from hospital-based studies topopulation-based data, can be problematic. 23

Obstructed labourDeaths that happen after obstructed labour and its consequences are hard tomeasure because they can be coded as uterine rupture, haemorrhage, or sepsis.This is especially problematic in settings where verbal autopsies are used toestablish cause of death. Verbal autopsy methods do not have consistent casedenitions, which creates confusion regarding hierarchical assignment ofcauses and subsequently affects the validity of the study data.24 A specic

mention is warranted to clarify the classication of obstructed labour in thisstudy, which is subsumed into direct causes of death, following guidance fromICD-MM.14 Although from a clinical perspective, obstructed labour is commonlyunderstood as a phenomenon by which a woman might die in labour, from anepidemiological and classication standpoint, it is inappropriate to identifyobstructed labour as a cause of death.

The ICD-10 aims to capture the initiating step most relevant to public health inthe sequence leading to death, because preventing this disorder would preventnot just the death, but all of the illness, complications, and disability thatpreceded it. In these cases of obstructed labour, death might be prevented byaccess to operative delivery. However, when the only available informationfrom a lay reporter suggests that the woman seemed to be in labour, or inpain, for a long time before death, little is actually known about the sequenceof events that leads to death, or about the progress of labour. These deathsmight be misattributed to obstructed labour, leading to overestimation of theproportion that could be prevented through operative delivery andunderestimation of the need for other services. In most settings, theimplementation of ICD-10 coding does not allow dual coding for cause ofdeatheg, obstructed labour and sepsis, or obstructed labour with uterinerupture and haemorrhage. Proposals for the ICD-11 revision link the disorders,thereby satisfying the need for clinicians to document obstructed labour whileascertaining the cause of death.

Indirect causes of death (excluding HIV/AIDS)The phenomenon of misattribution of indirect maternal causes of death,resulting in underestimation of 2090% of maternal deaths, has beendescribed in a number of settings.2530 In Austria, misclassication wassignicantly higher for indirect deaths (81%, 95% CI 6491) than directdeaths (28%, 2136),31 whereas in the UK, indirect deaths accounted for up to74% of under-reported maternal deaths during 200305.30

HIV/AIDSUnder-reporting and misclassication of indirect maternal deaths due to HIV/

AIDS are especially problematic. Although verbal autopsy might be able tomeasure AIDS mortality,43 hospital data-based validation studies might not beuseful in adjusting for the effect of misclassication error in the estimates ofcause-specic mortality fractions at the population level.23

When deaths happen in a facility, death certicate reporting might showonly HIV as a cause of death and not an obstetric complication such as sepsis.This situation highlights the need for specic review of deaths of womeninfected with HIV temporal to pregnancy. The woman might die from HIV orwith HIV while pregnant. Since 2010, this distinction is now possible fromthe standpoint of statistical tabulation as per ICD-10 coding. Our analysesprecede the changes in ICD-10 coding and so a decision was made toconsider cases where HIV was listed as a cause of maternal death, whether bydescription or use of a B code, as an indirect maternal death. As these dataare scarce, the proportion of indirect maternal deaths due to HIV is probablyunderestimated in this study. It is anticipated that as methods for globalmaternal death estimation evolve, evidence of the parameters needed toestimate indirect maternal HIV deaths and further clarication on the use ofICD-10 codes will standardise and improve our understanding of maternaland HIV death tallies.14


10/11

Articles


Declaration of interestsFive WHO staff members (LS, DC, T, AMG, and MT) are part of theteam that did the study. The ndings in this paper represent theconclusions of the authors. We declare that we have no furthercompeting interests.

AcknowledgmentsWe thank Andrew Howman and Rita Champaneria for their assistancewith the literature searching and data management; the Child HealthEpidemiology Reference Group (CHERG) for their technical review;Cristina Cuesta, Gilda Piaggio, Zoe Matthews, and Colin Mathers for theiradvice on the statistical methods; and Alexandra Furmston, Lixia Dou,Naomi Lee, Celia Liu, Joo Paulo Dias De Souza, Roderik F Viergever,Karmela Krlea-Jeri Irena Zakarija-Grkovic, and the Croatian Branch ofthe Italian Cochrane Centre for their assistance with translating reportsfor this study.

References 1 WHO, UNICEF, UNFPA, The World Bank. Trends in maternal

mortality: 1990 to 2010. WHO, UNICEF, UNFPA, and The WorldBank Estimates, 2012.

2 United Nations. United Nations Millennium Development Goals.2013. http://www.un.org/millenniumgoals/maternal.shtml(accessed Feb 28, 2014).

3 United Nations Secretary General. Global Strategy for Womensand Childrens Health. New York: United Nations, 2010.

4 WHO, UNICEF, UNFPA, The World Bank. Trends in maternalmortality: 1990 to 2008. WHO, UNICEF, UNFPA, and The WorldBank Estimates, 2010.

5 Hogan MC, Foreman KJ, Naghavi M, et al. Maternal mortality for181 countries, 19802008: a systematic analysis of progress towardsMillennium Development Goal 5.Lancet 2010;375: 160923.

6 Wilmoth J, Mathers C, Say L, Mills S. Maternal deaths drop byone-third from 1990 to 2008: a United Nations analysis.Bull World Health Organ 2010;88: 718.

7 WHO. International Classication of Diseases and Related HealthProblems. Geneva: World Health Organization, 1992.

8 Khan KS, Wojdyla D, Say L, Gulmezoglu AM, Van Look PF.WHO analysis of causes of maternal death: a systematic review.Lancet 2006;367: 106674.

9 Lozano R, Naghavi M, Foreman K, et al. Global and regionalmortality from 235 causes of death for 20 age groups in 1990 and2010: a systematic analysis for the Global Burden of Disease Study2010.Lancet 2012;380: 2095128.

10 United Nations. Millenium Development Indicators: world andregional groupings. http://mdgs.un.org/unsd/mdg/Host.aspx?Content=Data/RegionalGroupings.htm (accessed Feb 28, 2014).

11 WHO. WHO mortality database: tables. 2010. http://www.who.int/healthinfo/morttables.

12 Gulmezoglu AM, Say L, Betran AP, Villar J, Piaggio G.WHO systematic review of maternal mortality and morbidity:methodological issues and challenges.BMC Med Res Method 2004;4: 16.

13 WHO. ICD-10 translator. http://www.icd10data.com/Convert(accessed April 30, 2014).

14 WHO. The WHO Application of ICD-10 to deaths during pregnancy,childbirth and the puerperium: ICD-MM, 2012.

15 WHO, UNICEF, UNFPA, The World Bank. Trends in maternalmortality: 1990 to 2010. World Health Organization, UNICEF,UNFPA, and The World Bank, 2012.

16 Wilmoth JR, Mizoguchi N, Oestergaard MZ, et al. A new methodfor deriving global estimates of maternal mortality.Stat Politics Policy 2012;3: 1038.

17 R Development Core Team. R: A Language and Environment forStatistical Computing. Vienna, Austria: R Foundation for StatisticalComputing, 2011.

18 Plummer M. JAGS: A Program for Analysis of Bayesian GraphicalModels Using Gibbs Sampling. Proceedings of the 3rd InternationalWorkshop on Distributed Statistical Computing (DSC 2003); 2003;Vienna, Austria; 2003.

19 Abouzahr C, Royston E. Maternal mortality: a global factbook.Geneva: World Health Organization, 1991.

20 Ahman E, Shah IH. New estimates and trends regarding unsafe

abortion mortality. Int J Gynaecol Obstet 2011;115: 12126.

worldwide among all direct causes and the most

prominent cause in the Latin America and Caribbeanregion. This nding is despite the well establishedevidence that magnesium sulphate more than halves therisk of death from pre-eclampsia.3639 Although magnesiumsulphate is deemed an essential drug by WHO,40 theproblem is the extent to which it is available andappropriately used in most countries. A systematic reviewon the prevalence of pre-eclampsia and eclampsiadescribed the barriers to the use of magnesium sulphateas drug licensing and availability, inadequate and poorlyimplemented clinical guidelines, and insuffi cient politicalsupport for policy change.41 More recently, a WHO survey42 of delivery care in more than 300 health facilities in29 countries highlighted that, even if coverage ofmagnesium sulphate is high in cases where coverage isneeded, the overall mortality due to eclampsia was notreduced, highlighting the fact that more attention to otherelements of quality of care is also needed.

The large proportion of deaths attributed to indirectcauses cannot be ignored. As direct maternal deathsdecrease because of targeted interventions, efforts toreduce maternal mortality will have to be refocused onreduction of indirect causes. Although emphasis has beenplaced on linking maternal and HIV care, addressing theneeds of women with pre-existing comorbid disorderssuch as cardiac and endocrine disease in pregnancy willneed additional links between obstetric and other medicalspecialties. This situation is further complicated as theburden of non-communicable diseases is high indeveloping countries where health systems are poorlyequipped to coordinate specialised care. The four mainnon-communicable diseases are cardiovascular diseases,cancers, diabetes, and chronic respiratory diseases.43 Among these diseases, diabetes, cardiovascular,respiratory disorders, and cancers are of particular concernamong reproductive aged women for their potentialcontribution as indirect causes for both morbidity andmortality. Further focus on understanding the trueburden of these disorders in pregnancy, and on thechanging demographics of disease patterns, is warranted.

The gaps in coverage of effective interventions, for bothdirect and indirect causes of deaths, according to theirdistribution in various settings have large implications inview of the urgent need to accelerate the rate of decreasein maternal mortality to reach the MDG5 target andfurther to end all preventable maternal deaths.44 Therefore,accurate and routine information about causes of maternaldeaths is crucial in both implementation of interventionsand tracking and interpretation of the gaps in coverage.ContributorsLS and AMG conceived the review. All authors developed the methods.A-BM did the literature search and prepared the vital registration data.AB-M, DC, and JPD considered studies for inclusion and extracted thedata. LA developed the statistical model and did the statistical analysis. Allauthors provided input to the interpretation of the results. LS, OT, DC, andLA prepared the manuscript and supplementary material. All authorsprovided input to the manuscript and approved the nal version.


11/11

Articles

21 Gerdts C, Vohra D, Ahern J. Measuring unsafe abortion-relatedmortality: a systematic review of the existing methods.PloS one 2013;8: e53346.

22 Walker D, Campero L, Espinoza H, et al. Deaths fromcomplications of unsafe abortion: misclassied second trimesterdeaths. Reprod Health Matters 2004;12: 2738.

23 Chandramohan D, Setel P, Quigley M. Effect of misclassication ofcauses of death in verbal autopsy: can it be adjusted?Int J Epidemiol2001;30: 50914.

24 Leitao J, Chandramohan D, Byass P, et al. Revising the WHO verbalautopsy instrument to facilitate routine cause-of-death monitoring.Glob Health Action 2013;6: 21518.

25 Bouvier-Colle MH, Varnoux N, Costes P, Hatton F. Reasons for theunderreporting of maternal mortality in France, as indicated by asurvey of all deaths among women of childbearing age.Int J Epidemiol 1991;20: 71721.

26 Deneux-Tharaux C, Berg C, Bouvier-Colle MH, et al.Underreporting of pregnancy-related mortality in the United Statesand Europe. Obstet Gynecol2005;106: 68492.

27 Kao S, Chen LM, Shi L, Weinrich MC. Underreporting andmisclassication of maternal mortality in Taiwan.Acta Obstet Gynecol Scand1997;76: 62936.

28 Karimian-Teherani D, Haidinger G, Waldhoer T, Beck A, Vutuc C.Under-reporting of direct and indirect obstetrical deaths in Austria,1980-98.Acta Obstet Gynecol Scand2002;81: 32327.

29 Schuitemaker N, Van Roosmalen J, Dekker G, Van Dongen P,Van Geijn H, Gravenhorst JB. Underreporting of maternalmortality in The Netherlands.Obstet Gynecol1997;90: 7882.

30 Centre for Maternal and Child Enquiries. Condential enquiry intomaternal and child health. Saving mothers lives: reviewingmaternal deaths to make motherhood safer 20032005. TheSeventh Report of the Condential Enquiries into Maternal Deathsin the United Kingdom. Dec, 2007. http://www.publichealth.hscni.net/sites/default/les/Saving%20Mothers%20Lives%202003-05%20.pdf (accessed April 15, 2014).

31 Lopman B, Cook A, Smith J, et al. Verbal autopsy can consistentlymeasure AIDS mortality: a validation study in Tanzania and

Zimbabwe. J Epidemiol Community Health 2010;64: 33034. 32 Naghavi M, Makela S, Foreman K, OBrien J, Pourmalek F,Lozano R. Algorithms for enhancing public health utility ofnational causes-of-death data.Popul Health Met 2010;8: 9.

33 Vos T, Flaxman AD, Naghavi M, et al. Years lived with disability(YLDs) for 1160 sequelae of 289 diseases and injuries 19902010:a systematic analysis for the Global Burden of Disease Study 2010.Lancet 2012;380: 216396.

34 WHO. The International Classication of Diseases 11th Revision isdue by 2017. http://www.who.int/classications/icd/revision/en/(accessed Feb 28, 2014).

35 WHO. WHO recommendations for the prevention and treatment ofpostpartum haemorrhage. Geneva: World Health Organization, 2012.

36 Duley L, Gulmezoglu AM, Chou D. Magnesium sulphate versus lyticcocktail for eclampsia.Cochrane Database Syst Rev2010;9: CD002960.

37 Duley L, Gulmezoglu AM, Henderson-Smart DJ, Chou D.Magnesium sulphate and other anticonvulsants for women withpre-eclampsia.Cochrane Database Syst Rev2010;11: CD000025.

38 Duley L, Henderson-Smart DJ, Chou D. Magnesium sulphateversus phenytoin for eclampsia.Cochrane Database Syst Rev2010;10: CD000128.

39 Duley L, Henderson-Smart DJ, Walker GJ, Chou D. Magnesiumsulphate versus diazepam for eclampsia.Cochrane Database Syst Rev

2010;12: CD000127. 40 WHO. WHO Model List of Essential Medicines, 2013. http://www.who.int/medicines/publications/essentialmedicines/en/ (accessedApril 29, 2014).

41 Aaserud M, Lewin S, Innvaer S, et al. Translating research into policyand practice in developing countries: a case study of magnesiumsulphate for pre-eclampsia.BMC Health Serv Res 2005;1: 68.

42 Souza JP, Gulmezoglu AM, Vogel J, et al. Moving beyond essentialinterventions for reduction of maternal mortality (the WHOMulticountry Survey on Maternal and Newborn Health):a cross-sectional study.Lancet 2013;381: 174755.

43 WHO. Global status report on noncommunicable diseases 2010,2011. http://www.who.int/nmh/publications/ncd_report2010/en/(accessed April 30, 2014).

44 Bustreo F, Say L, Koblinsky M, Pullum TW, Temmerman M,Pablos-Mendez A. Ending preventable maternal deaths: the time isnow. Lancet Glob Health 2013;1: 17677.

Documents

Casuas Globales de Muerte Materna