(A-207) Epidemiologic Challenges in Using Claims Data to Study Opioid Use Disorder Treatment Trajectories

Background: The ability to accurately capture and analyze complex treatment patterns for opioid use disorder (OUD) using claims data is challenging given the nature of OUD treatment.

Objectives: Leveraging case examples from our observational longitudinal claims-based studies, we elucidate epidemiologic challenges encountered in OUD research and explore the implications of different study design decisions on findings.

Methods: This descriptive retrospective cohort study of commercially-insured adults (≥18 years) with OUD from 2010-2019 uses IBM MarketScan® claims data. All beneficiaries were required to be continuously enrolled during baseline (6 months before index date) and follow-up (3 months after index date). Our principal analysis defines index date by first OUD diagnosis claim. In the secondary analysis, we use an alternative method to capture index date utilizing either exposure to medications for OUD (MOUD) or OUD diagnosis claim (whichever comes first). We describe how treatment patterns for MOUD (buprenorphine, naltrexone, methadone) and behavioral therapy vary based on each analysis in the follow-up period. We also describe challenges to exposure definition.

Results: Our principal study sample included 372,614 beneficiaries and our secondary sample included 356,126 beneficiaries. In the principal analysis, behavioral therapy alone predominated followed by behavioral plus buprenorphine. Overall patterns of treatment trajectories in the secondary analysis were similar with behavioral therapy predominating followed closely by no treatment. While the secondary method for defining index date allowed for greater capture of beneficiaries based specifically on MOUD exposure, additional challenges arose due to greater sample attrition likely as a result of continuous enrollment requirements and uncertainty regarding MOUD indication. Notably, in both analyses, we encountered several challenges in measuring treatment exposure, including: assignment of days’ supply for procedure claims; under-coding of methadone; differentiating remission from no treatment; ascertaining buprenorphine and naltrexone indications; and inadequate capture of social interventions, such as harm reduction.

Conclusions: By not incorporating MOUD exposure in index date definition, researchers may potentially underestimate, exclude, or inaccurately represent treatment patterns among those with OUD. However, the benefit of incorporating MOUD exposure may be offset by sample loss and introduction of heterogeneity of indication and OUD severity mixing. These findings thus have implications for study design decisions in OUD research and highlight the importance of prioritizing data linkages to better capture the totality of experiences of people with OUD.