Haves and have-nots: socioeconomic position improves accuracy of machine learning algorithms for predicting high-impact chronic pain.

Abstract

Lower socioeconomic position (SEP) is associated with increased risk of developing chronic pain, experiencing more severe pain, and suffering greater pain-related disability. However, SEP is a multidimensional construct; there is a dearth of research on which SEP features are most strongly associated with high-impact chronic pain, the relative importance of SEP predictive features compared to established chronic pain correlates, and whether the relative importance of SEP predictive features differs by race and sex. This study used 3 machine learning algorithms to address these questions among adults in the 2019 National Health Interview Survey. Gradient boosting decision trees achieved the highest accuracy and discriminatory power for high-impact chronic pain. Results suggest that distinct SEP dimensions, including material resources (eg, ratio of family income to poverty threshold) and employment (ie, working in the past week, number of working adults in the family), are highly relevant predictors of high-impact chronic pain. Subgroup analyses compared the relative importance of predictive features of high-impact chronic pain in non-Hispanic Black vs White adults and men vs women. Whereas the relative importance of body mass index and owning/renting a residence was higher for non-Hispanic Black adults, the relative importance of working adults in the family and housing stability was higher for non-Hispanic White adults. Anxiety symptom severity, body mass index, and cigarette smoking had higher relevance for women, while housing stability and frequency of anxiety and depression had higher relevance for men. Results highlight the potential for machine learning algorithms to advance health equity research.