Creation and validation of a polysocial score for mortality among community-dwelling older adults in the USA: the health and retirement study

Abstract

Abstract Background the interrelatedness between social determinants of health impedes researchers to identify important social factors for health investment. A new approach is needed to quantify the aggregate effect of social factors and develop person- centred social interventions. Methods participants ([n = 7,383], 54.5% female) were aged 65 years or above who complete an additional psychosocial questionnaire in the health and retirement study in 2006 or 2008. Social determinants of health encompassed five social domains: economic stability, neighbourhood and physical environment, education, community and social context, and healthcare system. We used the forward stepwise logistic regression to derive a polysocial score model for 5-year mortality. Indices of goodness-of-fit, discrimination and reclassification were used to assess model performance. We used logistic regression to identify the association between polysocial score and mortality. Subgroup analyses were conducted to examine sex- and race-specific association. Results polysocial score was created using 14 social determinants of health. In the training cohort, the C-statistic was 0.71 for the reference model (only age, sex and race/ethnicity) and increased to 0.75 for the continuous and categorical polysocial score. Compared with the reference model, the integrated discrimination index for adding the continuous or categorical polysocial score was both 0.03 (P values < 0.001). Participants with an intermediate (odds ratio [OR] = 0.69; 95% confidence interval [CI], 0.51–0.82) or high (OR = 0.48; 95% CI, 0.38–0.60) polysocial score had lower odds of death than those in the low category in the fully adjusted model, respectively. Conclusions the polysocial approach may offer possible solutions to monitor social environments and suggestions for older people to improve their social status for specific health outcomes.