PAIN TO OPIOID USE DISORDER PATTERN THROUGH SEQUENCE (POPS): GENETIC, BEHAVIORAL, AND ENVIRONMENTAL ROOTS

Background While changes in policy have attempted to stem opioid use disorder (OUD) caused by prescription opioids, imprecise policy actions often create compromise at the cost of chronic pain patient treatment. Programs like the existing “Prescription Drug Monitoring Programs” (PDMP) in the United States of America are widespread, but do not have known clinical evaluation and create a single data point that often reduces patient experience to a single data point. While this approach led to initial success, later waves of the U.S. opioid epidemic have shown that such imprecise thresholding may cause later increases in illicit opioid use. Instead, substance use disorders are developmental, and require evaluation of treatment and symptomology longitudinally to adequately capture risk. We leveraged the AI infrastructure of the All of Us program to generate a longitudinal predictive relationship between pain, opioid analgesic prescribing, and later diagnosis of OUD. We call our predictive score the “Pain to Opioid use disorder Pattern through Sequence” or POPS. Method We developed a longitudinal long and short-term memory system (LSTM) model using a sample of 24,200 (N case=12,100) patients matched on electronic health record data, ethnicity, age, and sex in the All of Us sample. We then created a longitudinal probability of OUD for the entirety of the All of Us EHR Sample (N=358,677). We developed a POPS score in split half samples, ensuring no data leakage. We related the POPS to common correlates of OUD, including psychiatric diagnoses, substance use and social determinants of health. Finally, we used the genetic sequence data to conduct a multi-ancestry genome-wide association study (GWAS), where the rank normalized probability was used as the outcome. Results The model showed stronger prediction than most clinical predictors of OUD (AUC =.829, OR = 5.75 for each quantile increase in the probability of OUD, P < 2e1-6) and greatly out-predicted a chronic pain (OR = 2.280) and long-term prescription of opioid analgesics (OR = 1.822) phenotype. The POPS significantly and weakly related to common correlates of OUD. The POPS also remained a significant predictor of OUD, accounting for chronic pain and common correlates, suggesting the longitudinal pattern is a more sensitive predictor than singular clinical correlates of OUD. The cross-ancestry genome-wide data found 7 associated loci, which included loci in neurologically relevant pathways. Conclusion The POPS more effectively captures OUD risk than chronic opioid prescription or other correlates of opioid use disorder. The POPs shows some genetic origin. Future work will expand the genome-wide analyses to include drug repurposing to target the neurological pathways discovered in the GWAS.