Predicting 1-Year Statin Adherence Among Prevalent Users: A Retrospective Cohort Study

Area of Science:

• Pharmacoeconomics and Outcomes Research
• Health Services Research
• Clinical Pharmacy

Background:

• Predicting medication nonadherence typically focuses on therapy initiation, overlooking established users who form the majority of patients.
• Developing methods to predict nonadherence in prevalent medication users is crucial for improving evidence-based therapy use.

Purpose of the Study:

• To evaluate prediction approaches for statin nonadherence among prevalent users.
• To compare short-term prescription filling behavior, investigator-specified predictors, and empirically selected variables for adherence prediction.

Main Methods:

• Utilized claims data from a large national health insurer to form a cohort of statin users.
• Defined six groups of adherence predictors and tested ten models, including stratification by prescription days supply.
• Employed high-dimensional propensity score for empirical variable selection.

Main Results:

• The study included 93,777 individuals, with 58.4% adherence to statins.
• Three pre-index adherence predictors achieved a c-statistic of 0.70; investigator-specified and empirical variables performed worse (0.57-0.60).
• Post-index adherence indicators (c-statistic 0.74) and the addition of pre-index predictors (c-statistic 0.78) improved prediction accuracy.

Conclusions:

• Adherence prediction in established medication users is feasible using filling behavior.
• Refill patterns before and immediately after an index prescription are effective predictors of future adherence.