(P111) CLINICALLY SIGNIFICANT ADHERENCE THRESHOLDS FOR ORAL ANTICOAGULANTS IN PATIENTS WITH ATRIAL FIBRILLATION

Background: Oral anticoagulant (OAC) nonadherence in patients with atrial fibrillation (AF) is common and associated with higher risks of stroke and death. The conventional 80% threshold used to discriminate adherence from nonadherence has shown poor predictive accuracy for clinical outcomes in the few conditions in which it has been studied. Because OAC adherence thresholds significantly associated with clinical outcomes (“optimal thresholds”) have not been studied in AF, our objective was to identify them.

METHODS AND RESULTS: This retrospective observational study used physician billing, hospitalization, and prescription records of all British Columbians during 1996-2020 to identify incident cases of AF with incident OAC prescriptions. The adherence exposure was proportion of days covered (PDC). Outcomes were time to first stroke or systemic embolism (SSE), transient ischemic attack (TIA), or death, time to first SSE, and several secondary clinical outcomes. Effects of changes in PDC on relative hazard of outcomes were explored using flexible spline Cox regression models. We found PDC optimal thresholds by fitting regularized Cox models on dichotomous PDCs at all potential cut-points followed by lasso estimated coefficient paths to detect OAC-specific optimal thresholds per outcome. Cox models with categorical PDC based on optimal thresholds were fitted and hazard ratios were then compared between PDC categories.

44,172 patients were included (mean age 70.6, 44% female) with median follow-up of 6.7 years. Associations between PDC and all outcomes were nonlinear. Two-threshold models had the lowest AICs for the primary outcomes, and all performed better than the conventional 0.8 threshold. Optimal PDC thresholds were 0.8 and 0.9 for DOAC for SEE, TIA, or death. The corresponding optimal thresholds for VKA were 0.2 and 0.3 (Figure). Thresholds for all outcomes are shown in the Table. For stroke outcomes with DOAC, there was a significant hazard reduction for high and middle adherence (PDC≥0.8, ≥0.9) compared to low (PDC < 0.8). For VKA, the same was true for some, but not all stroke outcomes, and for some outcomes higher PDC was associated with higher risk of events.

Conclusion: The association between OAC adherence and AF clinical outcomes is nonlinear, and optimal adherence thresholds are different for VKA and DOACs. Two-threshold models represent these relationships best, being superior to the conventional single threshold of 0.8. For stroke outcomes with DOACs, there are optimal thresholds at PDC 0.8 and 0.9, and for VKA thresholds vary between outcomes.