OBJECTIVES: Primary medication nonadherence (PMN) occurs when patients prescribed a medication by a physician fail to fill their prescription. The study objective was to identify patient characteristics associated with a high risk of PMN for lipid lowering therapies (LLT).

METHODS: PMN for LLT was estimated by identifying patients with new prescription orders for statin and/or ezetimibe prescriptions in a large U.S. electronic medical records database between 7/1/2013 - 7/31/2015, and linking them to a database of filled prescription claims to determine if the prescription order was filled within 30 days. Logistic regression analysis and two non-parametric predictive analytic methods- random forests and boosted trees, were used to identify patient characteristics associated with high risk of PMN.

RESULTS: PMN was observed in 38.6% of the 69,227 patients who met all study criteria. A comparison of the three methods for predicting PMN showed that the boosted trees method had the best performance gauged by area under the curve (88.5%) and precision at 10% recall (98.6%). Lower baseline low-density lipoprotein cholesterol (LDL-C), not having a fill for an antihypertensive and older age were associated with higher risk of PMN. The interaction of lower LDL-C and no fills for an antihypertensive medication made the highest contribution to model performance (33.4%), followed by pre-index LDL-C (11.5%), and other predictors including age. Results were consistent across all analytic methods. Also, among the 30% of patients with the highest predicted probabilities of PMN, 85.7% were classified as PMN and accounted for 66.5% of all PMN patients.

CONCLUSIONS: These results show that information available to both the prescribing physician (LDL-C and age) and the patient's pharmacy benefit coordinator (adherence to pre-index medications) could be used to identify patients at highest risk for PMN and to target programs to improve adherence to LLT.