OBJECTIVES: To assess heterogeneity in adherence to medications and examine the degree to which health system characteristics can explain observed heterogeneity in comparative effectiveness research (CER) studies. METHODS: We analyzed data from two cohorts of patients >18 years old enrolled in UnitedHealth and initiating 1) a statin during 2005-2012 or 2) a new oral anticoagulant (NOAC) or warfarin from October 2010 through December 2012. We explored variation in adherence to high vs low intensity statin and to NOAC vs warfarin across prescribers, pharmacies, providers, and states, and we evaluated to what extent the variation could be explained by patient, therapy, and novel healthcare characteristics not ordinarily accounted for in CER studies. Adherence was defined as the proportion of days covered (PDC) with medication during 365 days of follow-up. Hierarchical linear models with different combinations of fixed and random effects for each health system level were built to explain variation in adherence, measured as cross-validated RRESULTS: Adherence was higher among NOAC initiators (mean PDC= 55%) versus warfarin initiators (48%), but similar between high and low-intensity statin users (58%). The healthcare system was responsible for 18.4% of the variation in adherence in the anticoagulant cohort, and patient, healthcare, and provider characteristics explained much of that (13.0%). In the statin cohort, only 7.9% of adherence variation was related to the healthcare system, with patient characteristics explaining 4.7%. Within the anticoagulant cohort, the healthcare system had little influence on adherence among patients with prior kidney disease or renal impairment. Within the statin cohort, adherence heterogeneity was slightly higher among secondary prevention patients. CONCLUSIONS: Adherence to study medications varied across the healthcare system, but the amount and type of variation depended on the clinical context. Investigators planning a CER study should consider the role of health system factors and their potential to influence medication adherence and effectiveness.