This study has some limitations, including unmeasured confounding and measurement error, as prescription coverage and fills data do not always reflect actual consumption, as well as lower rates of naltrexone receipt compared with buprenorphine. Future research is needed to assess time using and not using medication (ie, treatment discontinuation risk). Although we adjusted for both calendar-time and restricted individuals to 2-year observation periods surrounding the index drug-related poisoning to reduce heterogeneity in observation time, unmeasured time-varying confounding is possible, such as engagement with the health care system, although we found no significant protective associations with statins (negative control). Bias resulting from control period selection as a function of event time cannot be ruled out, although our design benefited from bidirectional time sampling, with past results in this study cohort being robust to the choice of control period window.16 Our results are also limited by lack of race and ethnicity data in commercial insurance claims, limiting our ability to study the associations of structural racism in shaping outcomes in individuals with cooccurring SUD. These findings also predate the surge of synthetic opioid use in the late 2010s, further external validity.
