Group-based Trajectory Models

A New Approach to Classifying and Predicting Long-Term Medication Adherence

Abstract

Background: 

Classifying medication adherence is important for efficiently targeting adherence improvement interventions. The purpose of this study was to evaluate the use of a novel method, group-based trajectory models, for classifying patients by their long-term adherence.

Research Design: 

We identified patients who initiated a statin between June 1, 2006 and May 30, 2007 in prescription claims from CVS Caremark and evaluated adherence over the subsequent 15 months. We compared several adherence summary measures, including proportion of days covered (PDC) and trajectory models with 2–6 groups, with the observed adherence pattern, defined by monthly indicators of full adherence (defined as having ≥24 d covered of 30). We also compared the accuracy of adherence prediction based on patient characteristics when adherence was defined by either a trajectory model or PDC.

Results: 

In 264,789 statin initiators, the 6-group trajectory model summarized long-term adherence best (C=0.938), whereas PDC summarized less well (C=0.881). The accuracy of adherence predictions was similar whether adherence was classified by PDC or by trajectory model.

Conclusions: 

Trajectory models summarized adherence patterns better than traditional approaches and were similarly predicted by covariates. Group-based trajectory models may facilitate targeting of interventions and may be useful to adjust for confounding by health-seeking behavior.