Maternal mortality estimates – Authors' reply

Abstract

www.thelancet.com Vol 384 December 20/27, 2014 2211 Nicholas Kassebaum and colleagues noted that maternal deaths have decreased less rapidly worldwide between 1990 and 2010 than reported by the UN agencies. They attribute this difference partly to higher UN estimates of all-cause reproductive-age mortality, especially in west Africa. They state that the UN estimates almost exclusively uses child mortality to predict adult mortality in west Africa, whereas their estimates are based on actual data. However, this is not the case. UN adult mortality estimates for all 16 west African countries use all empirical evidence available, as publicly documented. Kassebaum and colleagues’ estimates of all-cause reproductive-age mortality (Global Burden of Disease [GBD] 2013) are not publicly available, but previous GBD 2010 estimates for sub-Saharan Africa show unexpectedly low ratios of adult-to-child mortality. This is shown in the fi gure for Nigeria, which contains roughly half the population of west Africa. In Nigeria, GBD 2010 estimates for 1970–90 suggest an adult-to-child mortality ratio well below those documented in demographic surveillance sites (DSS) in other west African countries (eg, Ghana and Senegal) where high-quality data are available. This finding suggests that GBD underestimated the baseline level of mortality in women of repro ductive age in 1990. However, the UN’s World Population Prospects (WPP) estimates for 1970–90 are consistent with the documented DSS data. Both GBD 2010 and the WPP estimates suggest sharp increases in adult mortality in the 1990s, but large diff erences between estimates remain up to 2010. GBD 2010 estimates for 1990–2010 match the adult-tochild mortality ratios observed in DSS located in Senegal and Gambia (with under-5 mortality between 125 and 200 and female adult mortality between 250 and 300), whereas the WPP ratios are slightly higher than those observed in the Ghana DSS (with under-5 mortality between 150 and 200 and female adult mortality around 350). Since the HIV For the WHO, UNICEF, UN Population Fund, World Bank estimates see http://reliefweb. int/sites/reliefweb.int/fi les/ resources/Full_Report_3984.pdf Figure: Association between under-5 mortality and women’s mortality in Nigeria and west African demographic surveillance sites The dashed lines are a prediction of mortality of women aged 15–60 years (45q15) on the basis of the information about under-5 mortality (5q0), as shown in model life tables used to synthesise the history of countries with good death registration data conforming, respectively, to northern and southern European historical age patterns of mortality. Estimates of the probabilities 5q0 and 45q15 are the two key variables that model the mortality envelope (distribution of deaths by age for each country and year) in both WPP and GBD. DSS estimates refer to a period of 7·5 years on average. DSS which have existed for many years appear more than once. GBD=Global Burden of Disease (2010). WPP=World Population Prospects (2012). DHS=demographic and health survey. DSS=demographic surveillance sites (Nouna, Ouagadougou, and Oubritenga in Burkina Faso; Bandafassi, Mlomp, and Niakhar in Senegal; Farafenni in The Gambia; Cape Coast, Navrongo, and Kintampo in Ghana). MLT=model life table. epidemic (and possibly other causes of adult mortality) is more severe in Nigeria than in other west African countries, we should expect adult-to-child mortality ratios to be higher in Nigeria too. UN estimates are in line with data for recent household deaths reported in the 2008 Nigeria demographic and health survey, whereas GBD 2010 closely match estimates from sibling histories collected in the same demographic and health survey (figure). Although sibling histories have provided invaluable retro spective mortality data in low-income countries, they are aff ected by substantial under-reporting of adult deaths in west African settings. In a validation study of sibling histories in Senegal, nearly 25% of female respondents did not report deaths of one of their adult sisters. The methods Kassebaum and colleagues use to correct this under-reporting make strong assumptions and do not account for several reasons why deaths are under-reported in sibling histories. In countries with scarce vital registration, diff erent data sources often yield very diff erent estimates of adult mortality rates. This information must therefore be triangulated and validated against high-quality independent datasets to adjust implausibly low mortality rates. Without such eff orts, we might consistently misrepresent progress towards the fi fth Millennium Development Goal.