Abstract
Extensive research has been devoted to issues related to noncompliance in long-term clinical trials. Lack of compliance to the prescribed treatment by the participants in the trial can cause a loss in power of the study to detect the treatment difference. Traditional methods for evaluating the compliance of subjects include self-reported questionnaires and pharmacologic assays of drug levels in randomly-drawn blood samples, but each of these has important limitations. This paper proposes a method for assessing noncompliance based on longitudinal laboratory marker data that are affected by the drug. The problem of estimating the specific times of noncompliance of an individual is considered first. Next, combining these estimates across individuals to estimate the prevalence, over time, of noncompliance in a population is considered. The method will estimate the times of noncompliance of individual subjects with the subjects estimated to be compliant regarded as right-censored observations. Thus, the distribution of times of noncompliance in a population can be estimated by combining the estimated times of noncompliance using the Kaplan-Meier procedure, a methodology commonly used in failure time data. Since time to noncompliance is not observed, however, but rather needs to be estimated by the proposed method, the properties of the Kaplan-Meier estimators based on the estimated times are investigated. The results are illustrated with data from a recent AIDS clinical trial.
Get full access to this article
View all access options for this article.