Abstract / Summary
Importance: Pain varies in its intensity and course, yet much of our understanding of pain is based on cross-sectional assessments or longitudinal studies of selected clinical populations. Routinely collected electronic health record data may enable characterization of longitudinal pain patterns at scale. Objective: To identify longitudinal pain profiles among U.S. Veterans and characterize their clinical and polygenic correlates. Design: Cohort study using electronic health record and genomic data from the Million Veteran Program, with up to 16 years of pain assessments collected between 2000 to 2025. Setting: U.S. Department of Veterans Affairs health care system. Participants: Million Veteran Program participants assigned to African-like, Admixed American-like, East Asian-like, European-like, or South Asian-like genetically inferred ancestry groups who have at least three years of outpatient numeric rating scale pain records. Exposures: Outpatient pain ratings and polygenic scores for psychiatric, sleep, and substance use traits. Main Outcomes and Measures: Pain profiles were identified using Gaussian finite mixture modeling of pain intensity, directional change, and within-person variability. Profiles were characterized by pain, psychiatric and substance use diagnoses, opioid exposure, body mass index, and polygenic scores. Results: Among 624,996 participants (Mean [SD] age at first pain rating=53.4 [13.5] years; 568,833 [91.0%] male), 51,731,567 outpatient pain ratings were analyzed. Three profiles were identified: stable no pain, stable mild pain, and mild-to-moderate increasing pain. Compared with the stable no pain profile, the increasing pain profile had higher odds of chronic pain (OR=21.14 [95% CI: 18.96-23.57]), opioid use disorder (OR=9.32 [95% CI: 8.31-10.45]), opioid exposure (OR=8.61 [95% CI: 8.41-8.83]), major depressive disorder (OR=3.77 [95% CI: 3.67-3.87]), and posttraumatic stress disorder (OR=3.72 [95% CI: 3.61-3.84]). Associations were attenuated but remained after adjustment for mean pain severity. Polygenic scores for depression, insomnia, attention-deficit/hyperactivity disorder, and tobacco use disorder were the most consistent predictors of increasing pain across the three largest genetically inferred ancestry groups. Conclusions and Relevance: Longitudinal modeling of routinely collected pain scores identified pain profiles associated with clinical burden and polygenic liability. These profiles captured clinically meaningful information beyond mean pain severity alone, highlighting the value of longitudinal pain phenotyping in electronic health record data.