Randomized in error in pragmatic clinical trials.

Area of Science:

• Clinical Trials
• Biostatistics
• Health Informatics

Background:

• Pragmatic trials integrating electronic health records and patient-reported data face selection bias from differential post-randomization exclusion of "randomized-in-error" participants.
• Incomplete pre-randomization data can lead to participants being deemed ineligible after randomization, particularly in the intervention arm, complicating analysis and validity.

Purpose of the Study:

• To develop a statistical approach to mitigate selection bias in pragmatic trials arising from differential post-randomization exclusion.
• To accurately estimate the average treatment effect among participants not randomized in error.

Main Methods:

• A Bayesian model was developed within the potential outcomes framework to simultaneously identify "randomized-in-error" status and estimate the average treatment effect.
• Simulation studies evaluated the model's performance with 5%-15% randomization error rates, considering both measured and unmeasured outcomes for erroneous participants.

Main Results:

• The proposed Bayesian model demonstrated satisfactory performance, yielding low bias (<1%) and high coverage (approx. 95%) for estimated average treatment effects.
• Alternative methods, such as intention-to-treat and covariate-adjusted estimators, exhibited notable biases and lower coverage in simulations.

Conclusions:

• Differential exclusion post-randomization is a significant source of selection bias in pragmatic clinical trials.
• Bayesian methods offer a robust solution for identifying "randomized-in-error" participants and estimating intervention effects accurately, enhancing the reliability of trial findings.