Detecting participant noncompliance across multiple time points by modeling a longitudinal biomarker

Area of Science:

• Biostatistics
• Clinical Trial Methodology
• Pharmacokinetics

Background:

• Participant noncompliance complicates randomized clinical trial interpretation.
• Existing methods for detecting noncompliance are limited to single time points and single biomarker measurements.
• Biomarkers can objectively indicate exposure to non-study treatments.

Purpose of the Study:

• To develop a novel method for detecting participant noncompliance using longitudinal biomarker data.
• To estimate the probability of compliance at single, all, or future time points.
• To leverage full biomarker histories for more accurate noncompliance detection.

Main Methods:

• Modeling biomarker data as a mixture density with latent compliance components.
• Fitting mixed effects models for compliance and biomarker data.
• Deriving compliance probabilities conditional on longitudinal biomarker data.
• Evaluating the method via simulation and application to a smoking cessation trial.

Main Results:

• Conditioning on longitudinal biomarker data uniformly improved the area under the receiver operating characteristic curve for compliance detection.
• The novel method demonstrated improved accuracy in identifying noncompliant participants compared to existing methods.
• The method accurately predicted compliance at future time points in simulations.

Conclusions:

• The proposed method enhances the accuracy of noncompliance detection in clinical trials by utilizing longitudinal biomarker data.
• This approach offers a significant improvement over methods relying on single time point measurements.
• The ability to predict future compliance is a key advancement for trial monitoring and interpretation.