In his commentary on our paper ‘Mortality estimates for South East Asia, and INDEPTH mortality surveillance: necessary, but not sufficient,’1 Rao rightly points out the limitations of using mortality estimates based either on: (i) statistical modelling from a fraction of the population with inaccurate assumptions; or (ii) on small, unrepresentative population clusters with surveillance systems and accurate measures. When used for describing disease burden, driving evidence-based policy decisions, evaluating the targets of the Millennium Development Goals project, or making international comparisons, mortality estimates should be adequately qualified with uncertainty intervals and a discussion of data limitations. By contrasting the data-collection systems, statistical assumptions, and modelling used in two studies described in the June issue of IJE,1,2 and contrasting these studies’ representativeness, Rao highlights the inherent strengths and weaknesses of different approaches to disseminating population-level estimates in countries that may have incomplete or may lack registration systems for vital data or census or sample registration data. With the example of Vietnam, he points out the variation in estimates of cause-specific mortality based on these approaches, and the misrepresentation for countries that do not mirror similar disease patterns as their reference populations. The strength of each of the two approaches named above for estimating mortality drives its utility. In the first approach, the utility lies in comparisons over time and across regions of the world, for purposes such as prioritizing health needs, designing programmes, and financing health systems. Our paper is part of a series on the status of epidemiology in six regions of the world defined by the World Health Organization (WHO),3 and hence the need for comparable data. In our overview of the South East Asia region, we described the limitations of the mortality data, and in Table 2 of our paper we provided both official estimates by member states and WHO-based best estimates from available evidence. However, more specific details on modelling assumptions could have been provided, in addition to uncertainty intervals (UIs) to capture the uncertainty based on insufficient data and resulting from the specifications given for a model.4 Table 1 of our paper compares male and female life expectancy estimates from WHO5 and from the Global Burden of Disease Study series published in Lancet in 2010, with uncertainty intervals.6 The differences vary, and as indicated by WHO are subject to considerable uncertainty, particularly for countries with health-information systems that are weak, unreliable. or of low quality,5 underlying the need for uncertainty estimates6 or qualifiers, as suggested by Rao. Table 1 Life expectancy and healthy life expectancy estimates from the WHO World Health Statistics 2011 and the 2010 Global Burden of Disease series (with uncertainty intervals) The utility of the second approach and efforts of estimates based on statistical modelling should not be underestimated. The recent Global Burden of Disease projections in 187 countries are based on such estimates, utilizing methods that have been developed over the past 60 years (including a modification of previous methods that used under-5 mortality rates to predict adult mortality, as pointed out by Rao) and involving more than 500 researchers from 300 institutions in 50 countries. Margaret Chan, the director of WHO, states that such efforts also highlight ‘the need to close the data gaps’ and understand the discrepancies in estimates, particularly in low- and middle-income countries.7 This ‘unprecedented’ exercise that drives researchers towards more comprehensive and more accurate use of methods, both in primary data collection and in statistical modelling, is in stark contrast to the ‘potential to remain complacent’ suggested by Rao.7 Lastly, we are grateful for Rao’s commentary for informing an important debate, and are in complete agreement that capacity building in the collection and utilization of high-quality local mortality data needs to be strengthened, in conjunction with work done by government offices, to provide nationally representative estimates of mortality with better accuracy.
Read full abstract