The purpose of this study is to provide a theoretical framework for the analysis of medication adherence based on longitudinal data from electronic medication monitors and to suggest methods for unbiased estimation of the effect of time and covariates on adherence.
After defining the statistical summaries involved in adherence analyses and the assumptions necessary for their estimation, we address the issue of bias encountered when adherence is estimated on censored data. We suggest 2 unbiased methods to estimate adherence: (i) indirect combining implementation and persistence; and (ii) based on weights, allowing estimation of the effect of time and covariates on adherence via generalized estimating equations models.
We applied the proposed methods to investigate the effect of sex on adherence in a sample of 43 oncology patients followed 1 year. Implementation was higher for men than for women at baseline (98.8 vs. 97.5%, odds ratio [OR] 2.08, 95% confidence interval [CI]: 1.00-4.35), whereas the relationship was reversed at 1 year (94.5 vs. 96.4%, OR 0.65, 95%CI: 0.28-1.52). Adherence declined faster in men, with year-end values of 46.3% for men and 92.2% for women (OR 0.07, 95%CI: 0.02-0.26).
Estimation of adherence is a complex statistical issue with longitudinal and duration data, possibly censored, interleaving. This study provides a theoretical framework and suggests methods for unbiased estimation of adherence as a function of time and covariates. This allows the effect of an intervention to be estimated in clinical trials, and helps healthcare providers reframe adherence programmes to address covariates such as sex.