Applying Mixture Cure Survival Modeling to Medication Persistence Analysis

Author(s)

Chao Cai, College of Pharmacy, University of South Carolina, Department of Clinical Pharmacy and Outcomes Sciences, Columbia, South Carolina, USA.
Bryan L. Love, College of Pharmacy, University of South Carolina, Department of Clinical Pharmacy and Outcomes Sciences, Columbia, South Carolina, USA.
Ismaeel Yunusa
Claiborne E. Reeder, College of Pharmacy, University of South Carolina, Department of Clinical Pharmacy and Outcomes Sciences, Columbia, South Carolina, USA.

Document Type

Article

Subject Area(s)

Bias; Cohort Studies; Humans; Hydroxymethylglutaryl-CoA Reductase Inhibitors (therapeutic use); Medication Adherence; Middle Aged; Proportional Hazards Models; Survival Analysis

Abstract

PURPOSE: Standard survival models are often used in a medication persistence analysis. These methods implicitly assume that all patients will experience the event (medication discontinuation), which may bias the estimation of persistence if long-term medication persistent patients rate is expected in the population. We aimed to introduce a mixture cure model in the medication persistence analysis to describe the characteristics of long-term and short-term persistent patients, and demonstrate its application using a real-world data analysis. METHODS: A cohort of new users of statins was used to demonstrate the differences between the standard survival model and the mixture cure model in the medication persistence analysis. The mixture cure model estimated effects of variables, reported as odds ratios (OR) associated with likelihood of being long-term persistent and effects of variables, reported as hazard ratios (HR) associated with time to medication discontinuation among short-term persistent patients. RESULTS: Long-term persistent rate was estimated as 17% for statin users aged between 45 and 55 versus 10% for age less than 45 versus 4% for age greater than 55 via the mixture cure model. The HR of covariates estimated by the standard survival model (HR = 1.41, 95% CI = [1.35, 1.48]) were higher than those estimated by the mixture cure model (HR = 1.32, 95% CI = [1.25, 1.39]) when comparing patients with age greater than 55 to those between 45 and 55. CONCLUSIONS: Compared with standard survival modeling, a mixture cure model can improve the estimation of medication persistence when long-term persistent patients are expected in the population.

Digital Object Identifier (DOI)

https://doi.org/10.1002/pds.5441

Publication Info

Pharmacoepidemiology and Drug Safety, Issue 7, 2022, pages 788-795.