ML study reveals persistent healthcare financial vulnerability after COVID-19

A new population-level machine learning analysis published on arXiv examines healthcare financial vulnerability before and after COVID-19 using Medical Expenditure Panel Survey (MEPS) data from 2019 and 2021. The study, led by Alexey Kresin and eight co-authors, defines high financial burden as out-of-pocket healthcare spending exceeding 10% of family income. Survey-weighted subgroup analyses were complemented by interpretable logistic regression and two machine learning models—random forest and gradient boosting—to assess predictive performance and temporal generalization (i.e., whether pre-pandemic models could forecast post-pandemic vulnerability).

The results show that financial vulnerability is strongly tied to poverty status, insurance coverage, and prescription drug spending. Subgroup analyses found persistent disparities across demographic and socioeconomic groups, with some evidence of increased burden among vulnerable populations in 2021. Notably, models trained on 2019 data exhibited only modest reductions in accuracy when applied to 2021 data, suggesting that the core predictors of healthcare financial risk remained stable despite the pandemic's disruptions. This stability implies that policy interventions targeting these persistent factors could be effective even in crisis periods.

The study demonstrates the value of combining classical statistical modeling with machine learning for population health surveillance. By using interpretable models alongside high-performance algorithms, the researchers provide actionable insights for risk stratification and healthcare policy. The findings could help identify at-risk populations and inform strategies to reduce financial barriers to care, especially during public health emergencies.