KIVIMÄKI ET AL. RESPOND

Abstract

We welcome Ljung and Hallqvist’s evidence from the Stockholm Heart Epidemiology Program (SHEEP) case-control study1 and confirm that they picked up a typographical error in our Table 2 (they are correct that the number should be 213 and not 283), the analyses being based on correct figures.2 Studies suggest a greater number of risk factors for coronary heart disease (CHD) are present in low than in high socioeconomic groups, but a reasonably similar clustering of risk factors within these groups.1–3 However, evidence for the latter, although consistent, is still scarce. Why bother to study risk clustering? This is an attempt to solve the decades-old “puzzle” of what explains the social gradient in CHD. The so-called relative approach that compares relative risks between low and high socioeconomic groups before and after adjustments suggests that conventional risk factors explain only a part of the “puzzle”. However, if there were socioeconomic differences in clustering, synergistic effects between risk factors could lead to an underestimation of the contribution of risk factors to the social gradient in CHD if this within-group clustering were not appropriately taken into account. In light of the existing evidence from a previously published study,3 our recent publication,2 and the replication of our findings by Ljung and Hallqvist in the SHEEP study,1 socioeconomic differences in clustering seem unlikely, but prospective analyses are still lacking. A recent study of Finnish men explored the “puzzle” using an absolute approach comparing absolute socioeconomic differences in CHD between the total sample and a group that was free of all measured risk factors.4 This illustrated what would happen to the social gradient in CHD if the risk factors were removed from the population. According to that simulation, the majority of the gradient is explained by the risk factors. However, corresponding analysis in the SHEEP data provided inconsistent findings1 and, in our study, risk factors did not explain a large part of either the relative or absolute socioeconomic gradient in CHD, although our risk factor data are limited by the lack of biological measures. Further research is needed, particularly in statistically powerful databases containing the full range of behavioral and biological risk factor measures. The absolute approach is illustrative of the hypothetical basis of population-level differences in risk. However, alleviation of all risk factors may not be a realistic scenario, and comparisons between different populations (i.e., the total sample and a risk factor-free group) will generate contrasts that differ by more than the measured risk factors. Future analyses should consider different simulations (e.g., by modeling what would happen if the risk profiles were similar among low and high socioeconomic groups or if, in both groups, risk factor prevalences were reduced by, for example, 20%).