Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎

Contents lists available at SciVerse ScienceDirect

Midwifery

0266-61http://d

n CorrE-m

Pleasprov

journal homepage: www.elsevier.com/midw

Estimating maternal mortality and causes in South Africa:National and provincial levels

Eric O. Udjo, PhD (Research Director, Demographic Research Division)a,n,Pinky Lalthapersad-Pillay, PhD (Professor of Economics)b

a Bureau of Market Research, University of South Africa, P.O. Box 392, UNISA 0003, Pretoria, South Africab Department of Economics, University of South Africa, P.O. Box 392, UNISA 0003, Pretoria, South Africa

a r t i c l e i n f o

Article history:Received 16 October 2012Received in revised form20 May 2013Accepted 27 May 2013

Keywords:Pregnancy-related deathsMaternal mortality ratioCauses of deathSouth Africa

38/$ - see front matter & 2013 Elsevier Ltd. Ax.doi.org/10.1016/j.midw.2013.05.011

esponding author.ail addresses: udjoe@unisa.ac.za, bororue@yah

e cite this article as: Udjo, E.O., Laltincial levels. Midwifery (2013), http:

a b s t r a c t

Objectives: maternal mortality estimates for South Africa have methodological weaknesses. This studyuses the Growth Balance Method to adjust reported household female deaths and pregnancy-relateddeaths and the relational Gompertz model to adjust reported number of live births and estimatematernal mortality in South Africa at national and provincial level; examines the potential impact ofHIV/AIDS prevalence; and investigates the recorded direct causes of maternal mortality.Design: data from the 2001 Census, 2007 Community Survey and death registrations were utilised.Information on household deaths, including pregnancy-related deaths was collected from the afore-mentioned census and survey.Setting: enumerated households in the 2001 Census and a nationally representative sample of 250,348households in the 2007 Community Survey.Participants: information about members of households who died in the preceding 12 months wascollected, and of these deaths whether there were women aged 15–49 who died while pregnant orwithin 42 days after childbirth.Findings: maternal mortality ratio of 764 per 100,000 live births in 2007, ranging from 102 per 100,000 livebirths in theWestern Cape province to 1639 in the Eastern Cape. Maternal infections and parasitic diseases aswell as other maternal diseases complicating pregnancy, childbirth and the puerperium are the major causes.The study found a weak correlation between provincial HIVprevalence and maternal mortality ratio.Conclusion: despite strategies to improve maternal and child health, maternal mortality remains high in SouthAfrica and it is unlikely that the Millennnium Developmemnt Goal of reducing maternal will be achieved.

& 2013 Elsevier Ltd. All rights reserved.

Introduction

Maternal mortality is specifically an indicator of reproductivehealth and socio-economic development in general. The Safe Mother-hood Initiative was partly to reduce maternal mortality (Lilijestrandand Pathmanathan, 2004). Globally maternal deaths decreased by 47%between 1990 and 2010 (World Health Organization (WHO), 2012).Shah and Say (2004) have provided the following maternal mortalityratios (per 100,000 live births): globally in 2005, 400; Sub-SaharanAfrica: 920 in 1990 and 900 in 2005; South East Asia: 450 in 1990 and300 in 2005; developed regions: 11 in 1990 and 9 in 2005. Despite thedecreasing levels of maternal mortality globally and in Africa ingeneral, they remain relatively unchanged in Southern African coun-tries (Botswana, Lesotho, Namibia, South Africa and Swaziland). Trendin maternal mortality ratio (250 in 1990 and 360 in 2005) in SouthernAfrica (World Health Organization (WHO), 2012) is in adverse contrastto international trends. The issue of maternal mortality has been

ll rights reserved.

oo.com (E.O. Udjo).

hapersad-Pillay, P., Estimati//dx.doi.org/10.1016/j.midw.

thrust again to the fore by its inclusion in the Millennium Develop-ment Goals (MDGs) and has provided further impetus to studies onmaternal mortality in recent years (see for example, Human RightsWatch, 2011; Horton, 2012; Hsu et al., 2012).

Changes in health legislation, health policy and delivery of healthservices in post-apartheid South Africa have led to reforms inreproductive health (National Committee for the ConfidentialEnquiry into Maternal Deaths, 1998; Cooper et al., 2004). Despitethese reforms, the high rate of maternal mortality is one of thecountry's major population concerns (Department of Welfare, 1998).

Rationale for the study

The rationale for this study is as follows:

(1)

ng201

There is dearth of reliable estimates for monitoring maternalmortality in South Africa. Although several studies haveprovided estimates, the studies have weaknesses. Garenneet al.'s study (2008) used the general pattern of the UN modellife table system in estimating and assessing the plausibility of

maternal mortality and causes in South Africa: National and3.05.011i

E.O. Udjo, P. Lalthapersad-Pillay / Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎2

Plpr

their estimates of maternal mortality for South Africa. This lifetable is inappropriate for South Africa as it disregards the effects ofthe HIV/AIDS epidemic (see Udjo (2008)). Furthermore, theysuggest that mortality in the preceding 12 months to the 2001Census (Statistics South Africa, 2003) was overestimated whereasthe application of the Growth Balance Method to the data showsthat deaths were underreported. Also, they noted that the numberof births they projected backwards appeared to be low and arguedthat the number of women who delivered in the preceding12 months appeared to be high, which compelled them to makeadjustments. The study onwhich this article is based shows that itwas not about too many women giving birth in the preceding12 months but rather that some births that occurred in thereference period were not reported. A follow-up study byGarenne et al. (2009) has similar limitations. Although theyprovided confidence intervals for their estimates, the confidenceintervals do not resolve the biases in their estimates.One of the sources utilised in the Hogan et al. (2010) study wasvital registration data. Their estimates for South Africa are biasedas they did not adjust for incomplete registration that is often thecase in vital registration data (see Brass (1971), Hill (1987) andUdjo (2006)). Secondly, an examination of the UN PopulationDivision data base used in their estimates indicated that, in thecase of South Africa, the denominator in their estimate for 2008was the number of live births obtained from the 2007 CommunitySurvey (Statistics South Africa, 2007b). Besides the fact that theuniverse of the estimates (2008 death registration) and reportedlive births in the 2007 Community Survey are different, thenumber of live births (in the preceding 12 months) was notadjusted for reference period error.WHO, UNICEF, UNFPA and the World Bank (2010) estimates usedglobal adjustment factors for a group of countries (including SouthAfrica) they considered lacking good vital registration, to adjust formis-classification and incomplete registration. Such global adjust-ments may produce estimates that are either too low or too highas some countries may lie towards the extreme end of the medianvalue. Also, the magnitude of underreporting of deaths tends toget smaller over time due to improvements in the registrationsystem (Udjo, 2006). In providing additional explanation on theWHO, UNICEF, UNFPA and the World Bank estimates, Wilmothet al. (2012) noted that the evidentiary basis underlying theassumptions in the estimates is fairly weak. They further notedthat the model underlying the WHO, UNICEF, UNFPA and WorldBank estimates is clearly an enormous simplification of reality.In view of these limitations, a different approach is needed todetermine maternal mortality levels in South Africa to impart abetter understanding of their magnitudes.

(2)

Both national and provincial estimates of maternal mortality areneeded to ensure that suitable interventions are appropriatelytargeted. Most studies provide estimates at national level.

(3)

There are challenges when using currently available data, such aschanges to provincial boundaries (in 2005, 2008 and 2011) andthe possibility that the high prevalence of HIV/AIDS may maskother issues contributing to maternal mortality. HIV/AIDS is likelyto be an indirect cause of maternal death rather than a directcause. It is important to distinguish between direct and indirectcauses of maternal death.

Study objectives

The objectives of this study therefore are (1) to use the GrowthBalance Method to adjust reported household female deaths andpregnancy-related deaths, and the relational Gompertz model toadjust reported number of live births so as to provide estimates ofmaternal mortality in South Africa at national and provincial

ease cite this article as: Udjo, E.O., Lalthapersad-Pillay, P., Estimatiovincial levels. Midwifery (2013), http://dx.doi.org/10.1016/j.midw.

levels; (2) to examine the potential impact of HIV/AIDS prevalenceon maternal mortality at provincial levels in South Africa; and(3) to examine the recorded direct causes of maternal mortality inSouth Africa.

Methods

Data and subjects

Censuses, sample surveys and vital registration are the primarysources of nationally and provincially representative mortality data.Censuses and surveys usually do not provide information on perinatalmortality but they may be obtained from vital registration. Censuses,sample surveys and vital registration in South Africa have severalweaknesses as is the case in many other countries. Besides coverageerrors in censuses, these sources of data have content errors and, inthe context of maternal mortality, include underreporting of deaths aswell as errors in the number of live births. The District HealthInformation System (DHIS) in South Africa collects mortality databut these are hospital-based and, therefore, cannot be used to generatenationally and provincially representative mortality estimates. Further-more, the quality of the DHIS data varies from one province to another.Thus, estimates of maternal mortality in South Africa rely heavily oncensuses and surveys and to a lesser extent, on vital registration.

This study therefore utilised the 2001 South African Census, 2007Community Survey and Death registrations for 1997, 2001 and 2007(Statistics South Africa, 2007a). The 69 questions in the 2001 Censuscovered demographic and socio-economic profiles of the populationand households. The overall undercount in the 2001 Census was 18%.The 2007 Community Survey was the largest sample survey everconducted by Statistics South Africa. The objectives of the survey were(a) to provide information at lower geographical levels; (b) to buildhuman, management and logistical capacities towards the 2011Census; and (c) to facilitate the linkage between the establishmentof the national address system and database of dwelling units. A two-stage stratified cluster sampling method, comprising 947,331 indivi-duals from 250,348 households, was used. Institutions were excludedfrom the sampling. There were 88 questions in the 2007 CommunitySurvey and like the 2001 Census, it covered demographic and socio-economic profiles of the population and households. The overallresponse rate in the 2007 Community Survey was 93.9%. (StatisticsSouth Africa, the government department responsible for officialstatistics in South Africa, instituted the 2001 Census and the 2007Community Survey. Statistics South Africa makes census and surveydata available to the public through various media soon after theresults have been officially released).

Regarding mortality, and aside from the orphanhood questions,the 2001 Census and the 2007 Community Survey included questionsabout the number of deaths in the household in the preceding12 months, sex of the deceased, age of the deceased and whether thecause of death was natural or unnatural. If the deceased was a femaleaged 12–50 the question was posed as to whether she was pregnantat the time of death or whether death occurred within six weeksafter childbirth. The date of the last live birth was also asked offemales aged 12–50 years. These questions constituted the basis forthe computation of maternal mortality ratios.

The analysis of the direct causes of maternal deaths was basedon the death records for 1997, 2001 and 2007. Death recordsin South Africa are obtained through vital registration. The medicalcertification includes the immediate and underlying causes ofdeath. The certification forms are processed by Statistics SouthAfrica using ICD-10 coding. An inherent problem in vital registra-tion data in less developed countries hinges on completeness ofregistration. The adjustments carried out on the data are describedin another section of this article.

ng maternal mortality and causes in South Africa: National and2013.05.011i

Box 1. Estimating adjusted number of household andmaternal deaths, D´ using the Growth Balance Method

The Growth Balance Method by Brass (1971) was designed toestimate the completeness of, and hence adjust for incom-plete or overreporting of deaths in households during acensus or survey. This is based on the linear relationshipbetween deaths and age distributions expressed as

NðxÞ=NðxþÞ¼ r þ DðxþÞ=NðxþÞ ð1ÞwhereN(x) is the number of persons at exact age x, N(x+) is thetotal number of persons above age x, D(x+) is the total numberof deaths occurring to persons aged x and over and r is thegrowth rate. Since there would be some error patterns in thecompleteness of death distribution, the equation may berewritten as

NðxÞ=NðxþÞ¼ r þ kðDðxþÞ=NðxþÞÞ ð2Þwhere r (the intercept of the straight line fitted to the data) is anestimate of the growth rate, k (the slope of the fitted line) is acoefficient that is used to adjust for the completeness of deathreporting. The completeness of death reporting c, is estimated as

c ¼ 1=k ð3Þ

E.O. Udjo, P. Lalthapersad-Pillay / Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 3

Definition of maternal mortality

WHO defines a maternal death ‘as the death of a woman whilepregnant or within 42 days of termination of pregnancy, irrespectiveof the duration and the site of the pregnancy, from any cause relatedto or aggravated by the pregnancy or its management, but not fromaccidental or incidental causes’ (Graham, 1991: 102). The questions inthe 2001 Census and 2007 Community Survey were phrased asfollows: 2001 Census: ‘Has any member of this household died in thepast 12 months, i.e. between 10 October 2000 and 10 October 2001’‘If the deceased was a woman under 50 years, did (the person) diewhile pregnant or within six weeks after delivery’ and 2007 Com-munity Survey: ‘Has any member of this household passed away inthe last 12 months between February 2006 and March 2007?’ Theestimates of maternal mortality in this study were based on thepregnancy-related deaths computed from the above questionsadjusted for a 42 day puerperium. The magnitude of the directcauses of maternal mortality advanced by the authors of this articlewas based on the ICD-10 coding as provided by Statistics South Africaon registered deaths.

Estimating maternal mortality

If the data are perfect, unadjusted maternal mortality ratio,MMR is computed as

MMR¼ DBk ð1Þ

where MMR is the period maternal mortality ratio, D is thenumber of maternal deaths in the period, B is the number of livebirths in the period whereas k is a constant, usually 100,000. Butbecause the data are not perfect, adjusted maternal mortality ratioMMR′ is estimated as

MMR′¼ D′B′

k ð2Þ

where D′ is the adjusted number of pregnancy-related deaths inthe period, and B′ is the adjusted number of live births in theperiod. Thus, the steps involved in estimating the maternalmortality ratios in this study are as follows.

Regarding the numerator of the MMR′, firstly, the numbers ofreported household female deaths and pregnancy-related deathsin the reference period were tabulated by five-year age groupnationally and provincially (based on the 2001 provincial bound-aries) from the 2001 Census and 2007 Community Survey data.

Completeness of reporting of household female deaths nation-ally and provincially was then assessed using the Growth BalanceMethod so as to derive an adjustment factor for the number ofpregnancy-related deaths (see Box 1 for details).

Regarding the denominator of the MMR′, firstly, the numbers ofwomen and reported live births in the preceding 12 months weretabulated by reproductive five-year age group nationally and provin-cially from the 2001 Census and 2007 Community Survey data. Next,the relational Gompertz model was fitted to the data to assess theaccuracy and adjust the reported number of live births (see Box 2 fordetails). Lastly, MMR′ was computed based on the adjusted numbers.

c is for all causes of deaths and was used to adjust the reportednumber of pregnancy-related deaths. The procedure involved thefollowing: (a) N(x)/N(x+) and D(x+)/N(x+) values were com-puted from the tabulated age distribution of the population offemales and reported female deaths, (b) the computed N(x)/N(x+) values were plotted against the computed D(x+)/N(x+)values, (c) a straight line was fitted to the ‘best’ points(i.e. ignoring outliers) of the plotted values; (d) r and k weredetermined from the fitted line using the least squares method;(e) the reciprocal of k was then used to adjust the reportednumber of pregnancy-related deaths.

Findings

Estimated completeness of reporting of deaths in households usingthe Growth Balance Method

Fig. 1 illustrates the result of the application of the Growth BalanceMethod to the reports on female deaths at national level in 2001. Thescatter plot shows deviation from a straight line at some ages,indicating errors in the completeness of reporting of female deaths.

Please cite this article as: Udjo, E.O., Lalthapersad-Pillay, P., Estimatiprovincial levels. Midwifery (2013), http://dx.doi.org/10.1016/j.midw.

A similar pattern was observed at provincial levels. The estimatedcompleteness of reporting is summarised in Table 1. The estimates forthe Free State, Northern Cape and the Western Cape should be treatedwith scepticism as they imply large adjustments of maternal mortalityratios upward in the three provinces. The estimates indicate that thecompleteness of reporting of female deaths in households was higherin the 2001 Census compared to the 2007 Community Surveynationally and provincially except in the Northern Cape and WesternCape provinces where the reverse was the case. The differences in thecompleteness of reporting of female deaths in the 2001 Census andthe 2007 Community Survey were statistically significant nationallyand provincially (po0.05) except in the Northern Cape and NorthWest provinces (p40.05).

Assessment and adjustment of reported births in the preceding12 months

If there were no errors in the reporting of births in thepreceding 12 months, all the points in the scatter plot in Fig. 2(depicting the national level analysis) should lie in a straight linebut as seen in the graph, some of the points deviate from a straightline. The points to the left of the fitted line indicate omission ofsome births in the preceding 12 months whereas the pointscorresponding to the oldest reproductive age groups indicate ageerrors. A similar pattern was observed at provincial level. Theobserved and adjusted total fertility rates summarised in Table 2indicate that births in the preceding 12 months were under-reported during the 2001 Census and 2007 Community Survey(with the exception of the Eastern Cape in 2007). The magnitudeof the underreporting was greater in the 2001 Census comparedwith the 2007 Community Survey. Statistical tests of the

ng maternal mortality and causes in South Africa: National and2013.05.011i

0

0.02

0.04

0.06

0.08

0.1

0.12

0.00 0.02 0.04 0.06 0.08 0.10 0.12

N(x

)/N

(x+

)

D(x+)/N(x+)

Fig. 1. Plot of partial birth rates, N(x)/N(x+), against partial death rates, D(x+)/N(x+),2001: Females. .Source: Computed from 2001 Census data.

Table 1Estimated percentage completeness of reporting of female deaths in household byprovince, 2001 and 2007.Source: Computed from 2001 Census data.

2001Census

2007Community

Survey

Statistical test ofdifference incompleteness

of reporting between2001 and 2007: z value

South Africa 74.5 73.1 16.9n

Eastern Cape 75.0 71.0 18.4n

Free State 60.2 58.1 5.3n

Gauteng 72.4 71.7 3.71n

KwaZulu-Natal 73.2 70.0 17.4n

Limpopo 75.3 72.8 9.6n

Mpumalanga 74.7 67.5 23.2n

Northern Cape 59.5 60.1 −1.4North West 70.9 70.4 1.5Western Cape 61.6 69.6 −29.2n

n Statistically significant (po0.05): standard normal table value at 95% level ofconfidence¼1.96.

-3

-2

-1

0

1

2

3

4

5

6

-4 -2 0 2 4 6

z(x

)-(e

x)

g(x)

2001 Census 2007 CS 2007 CS Fitted

Fig. 2. Fitting the relational Gompertz model to current fertility, 2001 and 2007:National .Source: Computed from 2001 census and 2007 Community Survey (2007 CS).

Box 2. Estimating adjusted number of live births, B´ using therelational Gomperzt model

The number of live births to women aged 15–49 from theinformation on the reported date the last live birth child wasborn, may be under- or overreported (reference period error:see Brass (1971)). The relational Gompertz model wasdesigned to detect and adjust for such errors. The model isexpressed as

F ðxÞ ¼ F ⋅e−e−½aþbYsðxÞ� ð1Þwhere F(x) is the cumulated age-specific fertility rate up to agex, and F is the total fertility rate. Ys(x) is defined as –ln[–ln Fs(x)/F] where Fs(x) is a standard cumulative fertility rate up toage x (Brass, 1981). The and parameters measure the locationand spread of the fertility distribution (Brass, 1981). In the usualapplication, F is unknown but can be separated from theestimation of and with ¼1 using a linear transformation (Zaba,1981) from the following equation:

zðxÞ−eðxÞ ¼ aþ 0:48ðb−1Þ2þ bgðxÞ ð2Þwhere

zðxÞ ¼−ln½−lnF ðxÞ=Fðx þ 5Þ� ð3Þwhere F(x) is the cumulated ASFR up to age x, e(x) and g(x) aretabulated standard values; 0.48 is a constant. Model ASFRs wereestimated as

f ðxÞ′¼ F ⋅F ðxÞ′ ð4Þwhere f(x)′ is the decumulated model age-specific fertility ratefor women aged x and F. F(x)′ is the model cumulative fertilityup to age x. The adjusted number of live births for a specificperiod was estimated as

B′¼ ∑49

x ¼ 15

ðwðxÞ⋅f ðxÞ′Þ ð5Þ

where B′ is the adjusted number of live births and w(x) is thenumber of women aged x in a five-year reproductive age group.The procedure involved in adjusting the reported number oflive births therefore was as follows: (a) −ln[−ln F(x)/F(x+5)]values were computed from the distribution of reportednumber of live births in the last 12 months by age of womento obtain observed z values; (b) the observed z values wereplotted against standard z values; (c) a straight line was fitted tothe points using the group average method; (d) and werecomputed algebraically from the fitted line and applied to thestandard series of z's to obtain a single estimate of F (byaveraging the series of F's that were of consistent level);(e) model age-specific fertility rates were computed using theestimated , and and applied to the standard series of z's, and(f) the model age-specific fertility rates were applied to thedistribution of the number of women to obtain adjustednumber of live births.

E.O. Udjo, P. Lalthapersad-Pillay / Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎4

differences indicate significantly lower levels of reported fertilitythan estimated at national and provincial levels (po0.05) exceptin the Eastern Cape in 2007, where the reported level of fertilitywas significantly higher (po0.05) than the estimated level.

Estimated maternal mortality ratios

Table 3 shows the estimated maternal mortality ratios afteradjustments for errors and suggest that maternal mortality ratioincreased from about 473 per 100,000 live births in 2001 to about764 per 100,000 live births in 2007. This indicates an annualincrease of about 11% between 2001 and 2007. At provincial level,the results indicate that maternal mortality increased between

Please cite this article as: Udjo, E.O., Lalthapersad-Pillay, P., Estimatiprovincial levels. Midwifery (2013), http://dx.doi.org/10.1016/j.midw.

2001 and 2007 in all provinces except in the Western Cape. In2001, the province with the highest level of maternal mortalitywas the Free State but in 2007, it had shifted to the Eastern Cape.In both years, the Western Cape had the lowest maternal mortalityratio and was the only province that experienced a declinebetween 2001 and 2007.

HIV/AIDS and maternal mortality in South Africa

If HIV/AIDS was responsible for the increasing level of maternalmortality in South Africa, then the provincial distribution of HIVprevalence should be similar to the provincial distribution ofmaternal mortality ratio and result in a high correlation (r≥0.5)

ng maternal mortality and causes in South Africa: National and2013.05.011i

Table 2Reported and estimated total fertility rate by province, 2001 census and 2007 Community Survey.Source: Computed from 2001 census and 2007 community survey.

2001 Census 2007 Community Survey

Reported Estimated Statistical test of difference: z value Reported Estimated Statistical test of difference: z value

South Africa 2.2 3.0 −715.4n 2.5 2.7 −34.8Eastern Cape 2.2 3.3 −369.6n 2.9 2.8 3.2Free State 2.1 3.3 −272.8n 2.5 2.6 −2.50Gauteng 1.9 2.6 −272.4n 2.1 2.5 −28.3KwaZulu-Natal 2.2 3.8 −653.6n 2.5 3.0 −37.1Limpopo 2.6 4.0 −450.6n 2.8 3.1 −15.9Mpumalanga 2.4 3.4 −252.7n 2.6 2.9 −12.1Northern Cape 2.3 2.8 −69.1n 2.6 2.8 −5.3North West 2.4 3.4 −245.5n 2.8 3.2 −15.5Western Cape 2.1 2.7 −180.4n 2.2 2.4 10.5

n Statistically significant (po0.05): standard normal table value at 95% level of confidence¼1.96.

Table 3Provincial levels of maternal mortality ratios per 100,000 live births, 2001 and2007 (2001 Provincial boundaries).Source: Computed from 2001 census and 2007 Community Survey.

2001 Census 2007 Community Survey

South Africa 463 764Eastern Cape 539 1639Free State 619 1080Gauteng 372 484KwaZulu-Natal 579 969Limpopo 223 587Mpumalanga 468 477Northern Cape 396 610North West 545 698Western Cape 179 102

y = 1.7626x + 301.3R² = 0.1128

0

200

400

600

800

1000

1200

1400

0 50 100 150 200 250 300Mate

rnal d

eath

s p

er

100,0

00

live b

irth

s

HIV prevalence per thousand population aged 15-49

Fig. 3. Relationship of maternal mortality ratio (2007) and HIV prevalence (2008)in South Africa's provinces*. Source of HIV prevalence: Shisana et al. (2008).*It would have been more appropriate to use HIV prevalence for women aged 15–49in the general population instead of women and men combined in the generalpopulation, but the information is not available at provincial level from thepublished figures. HIV prevalence among females aged 15–49 in the generalpopulation was estimated as 20.2% and for males of the same age as 16.2 in thegeneral population (Shisana et al, 2005).

E.O. Udjo, P. Lalthapersad-Pillay / Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 5

and high R2 (≥0.5) between HIV prevalence and maternal mortalityratio in a scatter plot of the two variables by province. The scatterplot of the 2008 HIV prevalence and 2007 maternal mortalityratio, however, indicated a weak correlation (r¼−0.34). The R2

value of 0.1128 suggests that only about 11% of the differences inmaternal mortality ratio in South Africa's provinces in 2007 wereexplained by differences in HIV prevalence in the provinces (seeFig. 3).

Causes of maternal deaths

Of the direct causes, hypertensive disorder is the highestcontributor to maternal deaths in South Africa, though its magni-tude as a direct cause appears to be declining. The data suggest

Please cite this article as: Udjo, E.O., Lalthapersad-Pillay, P., Estimatiprovincial levels. Midwifery (2013), http://dx.doi.org/10.1016/j.midw.

that in 1997, it accounted for about 16% of the total immediatecauses of maternal death whereas in 2007 it accounted for about8% (Table 4). On the other hand, ‘other maternal conditions’account for over 50% of the total immediate causes of maternaldeath and the trend is upward. Of these ‘other maternal condi-tions’, maternal infections and parasitic diseases as well as othermaternal diseases complicating pregnancy, childbirth and thepuerperium are the major contributors. In 2007, these groups ofconditions accounted for about 62% of the maternal deaths(Table 4).

As seen in Table 5, the magnitude of the direct causes of deathexhibits differences at provincial level. In 2007, the relativecontribution of maternal haemorrhage, maternal sepsis and abor-tion to maternal deaths was highest in the Northern Cape andlowest in the Western Cape (Table 5); the latter also exhibited thelowest maternal mortality ratio in 2007. However, the relativecontribution of ‘other maternal conditions’ was highest in theWestern Cape compared with other provinces. Due to smallnumbers, further disaggregation of the ‘other maternal conditions’could not be carried out at provincial level.

Discussion

Maternal mortality is a public health issue and one of theMillennium Development Goals. Estimates of maternal mortalityfor South Africa have been bandied around, both in national andinternational publications. There is a need to critically examine theestimates to attest to the veracity of the problem in South Africa.

This study's examination of the methodologies of the estimatesindicates that there are a number of issues in the numerator anddenominator of the estimates that were inadequately addressed,which may have resulted in biases in the estimates. This studyattempted to improve on the estimates. Its national estimate ofmaternal mortality ratio of 463 per 100,000 live births in 2001 islower than the estimate by Garenne et al. (2008) (542 per 100,000live births) for the same period. However, its estimate for 2001 ishigher than that of Hogan et al. (2010) (237 per 100,000 livebirths) for the period 1980–2008. Regarding 2007, this study'snational estimate for the period (764 per 100,000 live births) ishigher than that of Garenne et al. (2009) (700 per 100,000 livebirths) for the period but lower than the WHO, UNICEF, UNFPA andthe World Bank (2010) estimate (425 per 100,000 live births) forthe same period. South Africa's Health Data Advisory and Co-ordination Committee (National Department of Health, 2012)recommended to the National Department of Health that, becausepregnancy-related mortality reported by households in censuses istwice as high as that from vital registration, cause of death data

ng maternal mortality and causes in South Africa: National and2013.05.011i

Table 4Contribution of direct immediate causes and other maternal conditions to maternal death 1997–2007: South Africa.Source: Computed from registered deaths 1997, 2001, 2007 based on ICD-10 codes.

Per cent contribution to maternal deaths

1997 2001 2007

Direct causes of deathn: South AfricaMaternal haemorrage 10.4 10.0 6.5Maternal sepsis 8.0 11.5 5.9Hypertensive disorder 15.8 13.4 8.3Obstructed labour 0.2 0.4 0.05Abortion 9.5 10.5 5.0Other maternal conditions 55.9 53.2 74.3

Other maternal conditions:Disorders related to pregnancy 3.1 3.0 5.0Maternal care related to fetus, amniotic cavity and childbirth problem 0.7 0.6 0.4Complications of labour and childbirth 4.6 4.1 2.7Complications related to puerperium 8.0 6.2 3.5Other obstetric conditions 42.1 40.4 63.4

Other obstetric conditions:Obstetric death of unspecified cause 0.3 0.8 1.0Death from obstetric cause 442 days ando1 year after childbirth 0.2 0.2 0.1Maternal infections and parasitic diseases complicating pregnancy, childbirth and the puerperium 4.9 7.1 18.0Other obstetric conditions 36.6 32.3 44.3

n HIV/AIDS is an indirect cause of maternal death and so is not included in the table.

Table 5Per cent contribution of direct immediate causes and other maternal conditions to maternal death by Province, 2007.Source: Computed from registered deaths, 2007 based on ICD-10 codes.

Direct causes of death EC FS GT KZN LP MP NC NW WC

Maternal haemorrage 9.0 5.6 6.9 4.7 9.0 5.9 10.8 6.3 3.6Maternal sepsis 9.0 2.5 4.5 6.6 6.4 10.8 5.4 4.9 0.0Hypertensive disorder 7.2 8.8 7.9 9.6 7.1 8.1 10.8 7.0 5.5Obstructed labour 0.0 0.0 0.0 0.2 0.0 0.0 0.0 0.0 0.0Abortion 4.5 3.8 4.1 5.9 5.8 5.4 10.8 3.5 1.8Other maternal conditions 70.1 79.4 76.6 73.0 71.8 69.9 62.2 78.3 89.1

EC¼Eastern Cape; FS¼Free State; GT¼Gauteng; KZN¼KwaZulu-Natal; LP¼Limpopo; MP¼Mpumalanga; NC¼Northern Cape; NW¼North West; WC¼Western Cape.

E.O. Udjo, P. Lalthapersad-Pillay / Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎6

from vital registration with adjustments be used to monitormaternal mortality. Accordingly, the Committee adopted a mater-nal mortality estimate of 310 per 100,000 live births for SouthAfrica for 2010. The scientific logic of this recommendation isbaffling, given that a number of studies (Groenewald andBradshaw, 2005; Groenewald et al., 2005; McKerrow andMulaudzi, 2010; Birnbaum et al., 2011) have expressed concernabout the accuracy of cause of death information from vitalregistration in South Africa. One study in particular, noted that‘in terms of monitoring the health status of the nation andunderstanding the burden of disease, the extent of ill-definedcauses together with 9.0% of deaths being due to garbage codes,highlights the urgent need to improve the quality of cause of deathcertification’ (Bradshaw et al., 2010: 27).

This study also focused on the direct causes of maternalmortality as well as provincial differentials and came to the sameconclusion as other studies that maternal mortality increasedduring the period 2001–2007 in South Africa and may have begunin the late 1990s. The 2011 South African Census data becameavailable as this study was nearing completion. Using the methodsdescribed above, the study estimated maternal mortality ratio tobe 692 per 100,000 live births in 2011 from the data, whichsuggests that maternal mortality declined during the period 2007–2011. It has been argued that the increase in maternal mortalitylevels in the 1990s–2007 in South Africa is due to HIV/AIDS(Garenne et al., 2009). The study on which this article is based

Please cite this article as: Udjo, E.O., Lalthapersad-Pillay, P., Estimatiprovincial levels. Midwifery (2013), http://dx.doi.org/10.1016/j.midw.

indicates a weak correlation between provincial HIV prevalenceand maternal mortality ratio. Moszynski (2011) attributes SouthAfrica's rising maternal mortality to health system failures.

Direct causes of maternal deaths are haemorrhage, sepsis,hypertensive disorder, obstructed labour and complications ofpoorly performed abortions (Winikoff et al., 1991). Analysis ofthe death records from vital registration (in the study underdiscussion) suggest that over 50% of maternal deaths in SouthAfrica are currently due to ‘other maternal conditions’. Thesecomprise mainly maternal infections and parasitic diseases as wellas other maternal diseases complicating pregnancy, childbirth andthe puerperium. This requires further investigation as this studydoes not proffer a strict conclusive answer as to the high level ofmaternal mortality due to other maternal conditions: did some ofthese women give birth in hospital or at home? This cannot beinvestigated from the vital registration data. Also, the data inves-tigating possible confounding factors are limited. It is suggestedthat, whereas interventions should continue to focus on reducingmaternal mortality due to hypertensive disorder, maternal hae-morrhage, maternal sepsis and obstructed labour, the thrust ofintervention should focus on reducing maternal infections andparasitic diseases as well as other maternal diseases that compli-cate pregnancy, childbirth and the puerperium.

This study's provincial level estimates suggest that the EasternCape currently has the highest maternal mortality ratio. Eight ofthe nine provinces showed a rising trend in maternal mortality

ng maternal mortality and causes in South Africa: National and2013.05.011i

E.O. Udjo, P. Lalthapersad-Pillay / Midwifery ∎ (∎∎∎∎) ∎∎∎–∎∎∎ 7

ratio between 2001 and 2007. The Western Cape had the lowestmaternal mortality ratio and was the only province that experi-enced a decline between 2001 and 2007. In all provinces, ‘othermaternal conditions’ are currently the largest contributors tomaternal deaths. Although further disaggregation of these condi-tions could not be carried out at provincial level due to smallnumbers, the same intervention measures should be applied atprovincial level.

The main limitation of this study is that, in adjusting forunderreporting of maternal deaths, it was assumed that under-reporting/overreporting of deaths in households was constant forall causes of death. As there is no methodology comparable to theGrowth Balance Method for estimating the completeness of deathreporting by specific causes, this assumption provided perhaps aminimum adjustment factor for the reported pregnancy-relateddeaths. Other possible limitations have to do with stillbirths andmultiple births. Regarding stillbirths, although the 2001 Censusand 2007 Community Survey fertility questions related only to livebirths, some women may have reported a still birth as a live birth.But some women may also have reported a live birth as a stillbirthif the child died a short moment after birth. The resulting biaseswould be in the opposite direction though they may not necessa-rily cancel out each other. Regarding multiple births, births in thepreceding 12 months were derived from date of birth of last childso multiple births would be present in the data (though likely tobe relatively small). Multiple births were not given a specific codein the data. Theoretically, stillbirths and multiple births wouldhave a potential impact on the denominator of the estimate.However, this study argues that any possible bias arising fromthis is most likely to be negligible because of how the denominatorwas derived: stillbirths and multiple births in the data are unlikelyto produce significant shifts in the α and β parameters of theGompertz model and hence unlikely to produce significant impacton the model age-specific fertility rates derived from the α and βparameters. Despite these limitations, the results in this studysuggest that, despite strategies by the National Department ofHealth to improve maternal and child health, maternal mortalityremains a public health challenge in South Africa. Whilst theaccuracy of maternal mortality estimates for South Africa might bedebatable, one irrefutable conclusion from different studies is thatmaternal mortality increased between 2001 and 2007. Althoughthe 2011 South African Census data indicate that maternal mor-tality declined between 2007 and 2011, the level remains high andSouth Africa is unlikely to achieve the MDG of reducing maternalmortality by three quarters by 2015. This should be a focus for therelevant government departments.

Conflict of interest statement

We hereby affirm that there is no conflict of interest that couldinappropriately influence this study. We did not receive anyfinancial support from any individual or organisation for thisstudy.

Acknowledgement

We wish to thank Statistics South Africa for providing access toits data. The views expressed in this paper are, however, those ofthe authors.

References

Birnbaum, J.K., Murray, C.J., Lozano, R., 2011. Exposing misclassified HIV/AIDSdeaths in South Africa. Bulletin of the World Health Organization 89, 278–285.

Please cite this article as: Udjo, E.O., Lalthapersad-Pillay, P., Estimatiprovincial levels. Midwifery (2013), http://dx.doi.org/10.1016/j.midw.

Bradshaw, D., Pillay-Van Wyk, V., Laubscher, R., 2010. Cause of Death Statistics forSouth Africa: Challenges and Possibilities for Improvement. Burden of DiseaseResearch Unit, Medical Research Council, Cape Town.

Brass, W., 1981. The use of the Gompertz relational model to estimate fertility.Paper presented at the International Union for the Scientific Study of Popula-tion Conference, Manila, 3, 45-361.

Brass, W., 1971. Methods for Estimating Fertility and Mortality from Limited andDefective Data. The University of North Carolina, Chapel Hill.

Cooper, D., Morroni, C., Orner, P., et al., 2004. Ten years of democracy in SouthAfrica: documenting transformation in reproductive health policy and status.Reproductive Health Matter 12, 70–85.

Department of Welfare, 1998. Population Policy for South Africa. Ministry forWelfare and Population Development, Pretoria.

Garenne, M., McCaa, R., Nacro, K., 2008. Maternal mortality in South Africa in 2001:from demographic census to epidemiological investigation. Population HealthMetrics 6, 1–13.

Garenne, M., McCaa, R., Nacro, K., 2009. Maternal mortality in South Africa: anupdate from the 2007 Community Survey. Paper prepared for the 26th IUSSPInternational Conference, Marrakech, 27 September–2 October.

Graham, W.J., 1991. Maternal mortality: levels, trends and data deficiencies. In:Feachem, R.G., Jamieson, D.T. (Eds.), Disease and Mortality in Sub-SaharanAfrica. Oxford University Press, London, pp. 425–445.

Groenewald, P., Bradshaw, D., 2005. Official death statistics under-represent AIDS asa cause in South Africa. AIDS Analysis Africa, April–May, pp. 2–4.

Groenewald, P., Nannan, N., Bourne, D., Laubscher, R., Bradshaw, D., 2005.Identifying deaths from AIDS in South Africa. AIDS 19, 193–201.

Hill, K., 1987. Estimating census and death registration completeness. Asian andPacific Population Forum 1, 8–13.

Hogan, M.C., Foreman, K.J., Naghavi, M., et al., 2010. Maternal mortality for 181countries, 1980–2008: a systematic analysis of progress towards MillenniumDevelopment Goal 5. The Lancet 375, 1609–1623.

Horton, R, 2012. Women's and children's health: no time for complacency. TheLancet 380, 1123–1125.

Hsu, J., Pitt, C., Greco, G., Berman, P., Mills, A., 2012. Count down to 2015: changes inofficial development assistance to maternal, newborn, and child health in2009–10, and assessment of progress since 2003. The Lancet 380, 1157–1168.

Human Rights Watch, 2011. Stop making excuses: accountability for maternalhealth care in South Africa. ⟨http://www.hrw.org/norde/100757⟩.

Lilijestrand, J., Pathmanathan, I., 2004. Reducing maternal mortality: can we derivepolicy guidance from developing country experiences? Critical elements inreducing maternal mortality. Journal of Public Health Policy 25, 299–314.

McKerrow, N., Mulaudzi, M., 2010. Child mortality in South Africa: using existingdata. South African Health Review, (Part 5), 59–72.

Moszynski, P., 2011. South Africa's rising maternal mortality is due to health systemfailures. British Medical Journal 343, d5089.

National Committee for the Confidential Enquiry into Maternal Deaths, 1998.Interim Report on the Confidential Enquiry into Maternal Deaths in SouthAfrica. Pretoria.

National Department of Health, 2012. Health Data Advisory and Co-OrdinationCommittee Report. Pretoria.

Shah, H.I., Say, L., 2004. Maternal mortality and maternity care from 1990 to 2005:uneven but important gains. Reproductive Health Matters 15, 17–27.

Shisana, O., Rehle, T., Simbayi, L., et al., 2005. South African National HIV Prevalence,HIV Incidence, Behaviour and Communication Survey, 2005. HSRC Press, CapeTown.

Shisana, O., Rehle, T., Simbayi, L., et al., 2008. South African National HIV Prevalence,HIV Incidence, Behaviour and Communication Survey, 2008: A Turning TideAmong Teenagers. HSRC Press, Cape Town.

Statistics South Africa, 2003. 2001 Census: Census in Brief. Pretoria.Statistics South Africa, 2007a. Causes of Death in South Africa 1997–2007. Pretoria.Statistics South Africa, 2007b. Community Survey 2007: Methodology, Processes

and Highlights of Key Results. Pretoria.Udjo, E.O., 2006. Estimation of mortality from vital registrations in South Africa.

Current HIV Research 4, 469–474.Udjo, E.O., 2008. A re-look at recent statistics on mortality in the context of HIV/

AIDS with particular reference to South Africa. Current HIV Research 6,143–151.

World Health Organization, 2012. Maternal mortality. ⟨http://www.who.int/gho/maternal_health/mortality/maternal/en/⟩.

World Health Organization (WHO), UNICEF, UNFPA and the World Bank, 2010.Trends in Maternal Mortality: 1990 to 2010. World Health Organization,Geneva.

Wilmoth, J.R., Mizoguchi, N., Oestergaard, M.Z., et al., 2012. A new method forderiving global estimates of maternal mortality. Statistics, Politics and Policy 3,1–38.

Winikoff, B., Carignan, C., Bernardik, E., Semeraro, P. 1991. Medical Services to SaveMothers' Lives: Feasible Approaches to Reducing Maternal Mortality. Popula-tion Council, New York: Programs Division Working Paper 4, March.

Zaba, B., 1981. Use of the Relational Gompertz Model in Analysing Fertility DataCollected in Retrospective Surveys. Centre for Population Studies WorkingPaper 81-2, London School of Hygiene and Tropical Medicine.

ng maternal mortality and causes in South Africa: National and2013.05.011i