Managing existing pain: Predicting pain progression and improvement using machine learning approaches.

PURPOSE: Chronic pain (CP) is highly prevalent in the U.S. and imposes substantial health and economic burdens. Understanding which biopsychosocial and treatment-related factors drive pain progression and improvement is essential for developing targeted prevention strategies.

METHODS

Using the 2019-2020 National Health Interview Survey Longitudinal Cohort data, we applied Random Forest machine learning algorithms with permutation-based feature importance to predict four one-year pain transitions: nonchronic-to-chronic pain progression, CP-to-high-impact chronic pain (HICP) progression, complete CP improvement (CP to no pain), and partial CP improvement (CP to nonchronic pain). Analyses were stratified by age, sex, race/ethnicity, education, and urban-rural residence.

RESULTS

Comorbidity count was a universal predictor across all four outcomes and nearly every subgroup. Opioid use ranked first for HICP progression and predicted higher odds of progression. Socioeconomic factors, particularly employment status, ranked highest among older, male, non-Hispanic White, and less-educated adults, whereas social relations ranked highest among racial and ethnic minorities. Predictors shifted from health conditions for CP progression toward pain treatment use and social and economic factors for HICP progression.

CONCLUSIONS

Predictors of pain progression and improvement differ by transition stage and across populations, highlighting the need for equity-informed pain management strategies tailored to population-specific risk profiles.