On causal decomposition of summary measures of population health

Measures of Population Health (SMPH) are useful simply to follow trends and reveal patterns in the health of members of a population. For example, an SMPH like health-or disability-adjusted life expectancy (HALE or DALE), will hopefully increase over time for a given population. If this measure increases more rapidly than life expectancy (LE), then we may conclude that not only are we " adding years to life " , but also " adding life to years " (Rochon Commission 1987), or in Fries' (1980) sense, we are witnessing a " compression of morbidity ". Moreover, given the ways this kind of SMPH is typically constructed for a country, such as the Sullivan method, increment-decrement life tables, and the Global Burden of Disease (GBD) approach; and given sufficient sample size in the underlying data systems, we should also be able to produce estimates of the measure by sex, geographic area, socio-demographic group and age. These " straightforward " disaggregations rely on the fact that HALE or DALE measures can be produced for constituent subpopu-lations, and are additive over age groups that together cover the full life cycle (Wolfson 1996). Comparing these disaggregated SMPH and examining their trends over time can help indicate groups or areas for priority health policy attention. However, expectations for SMPH are much higher than this. It is better if historical changes in SMPH can be ascribed to one or another cause, and the impacts of prospective changes from interventions or other causes can be estimated. SMPH will be most relevant for public policy if their movements or variations can be attributed to causes, and their likely responses to health policy interventions can be explicitly quantified. For these reasons, there is a desire that SMPH also be disaggregated along another dimension, " causes ". The (naive) assumption that this kind of disaggregation is reasonable stems from widely used breakdowns of mortality rates by " cause of death " (i.e. the ICD-defined disease recorded on death certificates). In this case, the sum of all " causes " of mortality 304 Summary Measures of Population Health equals the overall mortality rate, and the causal attribution of deaths to diseases is additively decomposable. Similarly, the entire burden of health problems measured in DALYs can be exhaustively partitioned among a set of explicit causes (e.g. Table 4 in WHO 2000). This is a convenience, especially when communicating to non-expert journalists and politicians. …