Who Needs Cause-of-Death Data?

Perspectives M ore than half of the world's deaths pass by undocumented as to cause [1]. Whilst the appropriate focus of health services may well be the care of the living, consistent and reliable cause-of-death data also constitute a crucial and major resource for health planning and prioritisation, and their lack in many settings is a major concern. Two new papers from Christopher Murray and colleagues in this issue of PLoS Medicine [2,3] report important methodological advances which should go some way towards filling these data gaps. To paraphrase George Orwell, all deaths are equal, but some are more equal than others. In particular, the chance of a death being registered and documented as to cause depends strongly on the socioeconomic status of the community and nation in which it occurs, and this is a major obstacle in coming to a meaningful global overview of mortality patterns. Whilst richer settings have traditionally aggregated physician death certificates and autopsy data as the basis for public health reporting, in poorer circumstances alternative approaches have to be used. Over the last 25 years, these strategies have often involved so-called " verbal autopsy " (VA)—interviewing relatives and witnesses of deaths and interpreting the interview material to arrive at cause(s) of death [4,5]. Much VA interpretation has been undertaken by physicians (physician-coded verbal autopsy, PCVA), but this approach makes large demands on limited resources and can be inconsistent over time and place. Much work on VA methodology has concentrated on emulating individual physician death certification, often glossing over the considerable variability and imprecision with which death certificates, the supposed " gold standard, " are sometimes completed [6]. Newer approaches using computer models for interpreting VA data are now tending to supersede PCVA, both for populations in general [7,8] and for specific subgroups [9,10], putting more emphasis on cause-specific mortality fractions (CSMFs) than on individual causes. Methodological advances in cause-of-death determination have not always been explicit about which gaps in the global data they seek to fill, and this has sometimes led to a confused overall picture. There are different levels at which data on mortality patterns are needed (i.e., from the local to the global) and various ways of meeting these needs, as shown in Table 1. Murray and colleagues' new approach for estimating population CSMFs [3] within countries that have existing data on hospital deaths and partial vital registration is a big step forward …