Correlation is not prediction: Self-assessed overall health as a predictor of mortality

Abstract

I quantify the degree to which people’s self-assessed overall health (SAOH) predicts their subsequent mortality. Prior research has justified using this measure in substantive research by claiming it has high predictive validity; many authors subsequently suggest that the measure has clinical utility because of its validity in predicting mortality. I computed risk of death within 1, 3, 5, and 10 years as a function of responses at baseline to a question about whether people’s health is generally excellent, very good, good, fair, or poor. Data come from the 1986 to 2018 U.S. National Health Interview Surveys and the 2014/2015 High School and Beyond study—each linked to mortality records. People offering worse evaluations of their health were substantially more likely to die within 1, 3, 5, and 10 years. Nevertheless, SAOH has low sensitivity and a high false positive rate (regardless of how it is dichotomized). Results are consistent across samples and sociodemographic groups. SAOH is an important measure of well-being and should continue to be used in epidemiological and other research. However, on its own it is not predictive of individual-level mortality; claims about its clinical utility for targeting interventions at people at highest risk of death are not supported.