Is the use of self-rated health measures to assess health inequalities misleading?

In three studies published in IJE, authors examined whether the association between self-rated health (SRH) and mortality varied by socioeconomic status (SES). They then used the empirical results to comment on the potential discordance in the magnitude of social inequalities in health when estimated using SRH as opposed to mortality. The three studies, however, arrived at different conclusions about the potential misestimation of health inequalities when using SRH instead of mortality. Huisman and colleagues concluded that a weak interaction between SRH and education whilst predicting mortality is unlikely to result in over-estimation of educational differences in health status. Meanwhile, SinghManoux and colleagues concluded that use of SRH might under-estimate health inequalities by income and occupation. Dowd and colleagues reported that ‘self-rated health does not predict mortality as well at lower levels of SES’, but do not speculate on whether this will result in overor under-estimation of social inequalities in health. Testing whether the association between SRH and mortality varies by SES can at best provide indirect clues as to whether the use of SRH (instead of mortality) overor under-estimates health inequalities. Furthermore, interpretations of interactions in generalized linear models are not straightforward. It is, therefore, surprising that the authors did not directly test whether social inequalities in health are misestimated, even though this seems possible with the datasets they used for their analyses. Using the Established Populations for the Epidemiologic Study of the Elderly (EPESE) dataset—the same dataset used by Idler and colleagues in their classic study showing that SRH predicts mortality— we conducted a direct test of the difference in social inequalities in health when predicting SRH compared with mortality. We used education and income collected in 1982 to predict poor SRH in 1994 and mortality before 1994 in the New Haven EPESE study population. The mean (SD) age of subjects in 1982 was 74.2 (6.8). As shown in the Table 1, compared with those with a college degree or more, the odds ratio (OR) for reporting fair/poor/bad health for those with less than a high school education was 2.28 (95% CI 1.22–4.25); the corresponding OR for mortality was 2.04 (95% CI 1.41–2.94). Concordance in the ORs were observed across other educational categories as well as by income (Table 1). We also tested for the presence of an interaction between SRH and each indicator of SES (education and income) when predicting mortality. In our sample, we did not find any evidence for these interactions (results available upon request). We think it is important for readers to note that in the recent articles that appeared in IJE, the authors did not directly test whether social inequalities in health are misestimated with the use of SRH. Instead, they tested whether the SRH-mortality association varied by SES. While interesting in itself, such an approach to ascertaining the magnitude of social inequalities of health is circuitous and potentially misleading. At least in the EPESE dataset, when we tested for this directly, there is no evidence that we are overor underestimating social inequalities in health by using SRH, a subjective measure, as opposed to the ‘objective’ measure of mortality. It would be interesting to know the results of similar direct assessments in the datasets used in the IJE studies and whether substantive conclusions would be altered.