Background Health is affected by the joint impact of lifestyle factors, living environment, and social contexts. Lower socio-economic position may associate with exposure to several hazardous risks and susceptibility to poor health; higher socio-economic position may afford better protection from ill-health via resources acting as buffering factors. This study aimed to investigate associations in the study population between several health determinants and self-rated health (SRH - a measure of global health status, and independent predictor of mortality), and conduct an exploratory analysis of the relative importance of determinants in income-based groups, using classification trees.

Methods Using cross-sectional data from year 15 of the American CARDIA longitudinal study (Coronary Artery Risk Development in Young Adults), the sample of 3649 men and women (mean 40.2 years) was split into 5 income-based groups. Factors associated with SRH in each group were analysed using classification trees; parametric multivariate regression is limited in capturing complex combinations of multilevel factors in a simple model. SRH responses were categorised as ‘good’/ ‘poor’. Predictor variables represented health determinants based on the ecological model of health: age, sex, hereditary factors; individual lifestyle factors / medical history; social / community influences; living / working conditions.

Results Income and SRH were positively associated (p < 0.05): proportion of ‘good’ SRH increased from 37.5% (n = 217): < $25,000; to 77.1% (n = 615): $100,000+. Data suggested a socio-economic gradient for lifestyle and social factors, and living / working conditions. Ranking of health determinants (normalised importance) in relation to SRH differed for each income-based group. Physical activity and chronic burden from serious personal ill-health were associated with SRH for all income groups; other determinants varied. In lower income groups, additional indicators of chronic burden were associated with SRH. Social support, control over life, optimism, and resources for paying for basics, medical care and health insurance were greater (%) in higher income groups.

Conclusion SRH does not simply reflect disease. Classification trees are a useful tool in contextualising risk factors, highlighting population subgroups with relatively homogenous risks of an outcome, and identifying the relative importance of associated risk and protective factors for further inquiry. Ranking of predictor variables were not identical for each income-based group, suggesting a potential contribution of these different factors to the socio-economic gradient in SRH. Findings imply that as well as differences in the intensity of public health action across the socio-economic gradient, differences in the type of interventions to improve SRH may also be important.