Abstract:

Participant noncompliance, in which participants do not follow their assigned treatment protocol, has long complicated the interpretation of RCTs. Because participant noncompliance is often unobserved, no gold standard exists for ascertainment. In some trials, biomarker data may have a relationship with participant noncompliance. We present a method that uses longitudinal biomarker data to model compliance across multiple time points when compliance is unobserved. We use random effects models to express the biomarker distribution as a longitudinal mixture density with parameters estimated by an EM algorithm. Conditional on biomarker data, we extract three probability estimators from the mixture density: 1) the probability of compliance at the final time point of the trial; 2) the probability of compliance at all time points; and 3) the prediction probability of compliance at the final time point. We evaluate our method by simulating a trial of n = 1000 participants followed over 5 time points. We compare probability estimators 1) and 2) to those that ignore the withinparticipant correlation in the data. As 3) does not have a naive comparator, we plotted its calibration lines.
