Additionally, while recent studies22 have shown poor positive predictive values of ICD codes for OUD, our analysis was conducted on treatment-seeking individuals with a primary diagnosis of OUD who were likely to have a high severity of substance use and its concomitant impairment,23 rendering it more likely that our sample was correctly capturing individuals with active OUD. We also conducted sensitivity analyses that eliminated individuals with cooccurring alcohol use disorder and limited observations to those with active OUD claims, illustrating robustness of our main effects. Our study is also strengthened by use of repeatable-event case-crossover design, mitigating bias from time-invariant confounding, as each participant acted as his or her own control.
