(171) Quantifying the impact of different look-back period criteria to identify new users when medication adherence is poor: A study of chronic medications for opioid use disorder

Background: The new user study design is subject to misclassification of prevalent users as new users when using shorter look-back periods. Previous studies focused on intermittent medications and chronic medications with at least moderate adherence.

Objectives: We quantified the impact of different look-back period criteria to identify new users of medications for opioid use disorder (MOUD), a chronic therapy to treat opioid use disorder (OUD), with known poor adherence (e.g., frequent interruption and reinitiation).

Methods: Using the Merative™ MarketScan® Multi-State Medicaid Database, we constructed eight yearly cohorts (2014-2021) of adults 18–64 years coded for OUD with an MOUD prescription (≥28 days’ supply) from Jan 1 to Dec 31. For each yearly cohort, we created new user cohorts with increasingly restrictive look-back periods without MOUD exposure (6-, 12-, 24-, 36-months) and implemented three definitions for inclusion: (1) continuous enrollment; (2) continuous enrollment + washout for MOUD of same agent (no MOUD ≥28 days; less restrictive); (3) continuous enrollment + washout for any MOUD (no MOUD ≥1 day; most restrictive). We quantified changes in cohort sample size and proportion of prevalent users misclassified as incident MOUD users (gold standard: 36-month lookback period). We stratified results by agent and year. We averaged proportions across years, accounting for patients in multiple yearly cohorts.

Results: We identified 209,046 eligible individuals contributing 526,435 unique index patient-years (index MOUD: 68% buprenorphine, 18% methadone, 13% naltrexone). For the 6-, 12-, 24-, and 36-month lookback periods, an increasing proportion of new users were excluded with increasingly restrictive criteria: (1) enrollment [range, 12% (95% CI, 11%-14%) for 6 months to 53% (95% CI, 44%-63%)] for 36 months; (2) enrollment + less restrictive washout [range, 66% (95% CI, 62%-70%) to 90% (95% CI, 88%-92%)]; and (3) enrollment + most restrictive washout [range, 73% (95% CI, 69%-78%) to 94% (95% CI, 92%-95%)]. The proportion of prevalent users misclassified as incident users increased steadily with decreasing duration of washout periods (24-, 12-, and 6-months); we observed misclassification among 9% (95% CI, 7%-10%) to 34% (95% CI, 31%-38%) of patients (less restrictive definition), and 14% (95% CI, 12%-17%) to 50% (95% CI, 46%-53%) of patients (most restrictive definition). Patterns differed slightly by year and agent (e.g., naltrexone users less likely to be misclassified).

Conclusions: The choice of look-back period criteria presents tradeoffs between cohort sample size and misclassification of incident use. To study chronic medications characterized by poor adherence, stricter inclusion criteria may be warranted to mitigate prevalent user bias.