Estimating treatment effects in randomised controlled trials with non-compliance: a simulation study.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Randomized controlled trials (RCTs) are the gold standard for evaluating health interventions.
• Patient non-compliance with assigned treatment can compromise the integrity of RCTs and bias results.
• Understanding the impact of non-compliance on treatment effect estimation is crucial.

Purpose of the Study:

• To compare the performance of common analytical approaches for handling non-compliant data in simulated RCTs.
• To evaluate bias and precision under various non-compliance scenarios.
• To provide guidance on selecting appropriate analysis methods.

Main Methods:

• Simulated hypothetical RCTs based on a real study, manipulating non-compliance factors (type, randomness, degree).
• Compared intention-to-treat (ITT), as-treated (AT), per-protocol (PP), instrumental variable (IV), and complier average causal effect (CACE) analyses.
• Assessed bias, mean square error (MSE), and 95% coverage of true treatment effects (large, moderate, null).

Main Results:

• Intention-to-treat (ITT) analysis showed considerable bias for moderate to large treatment effects.
• As-treated (AT), per-protocol (PP), instrumental variable (IV), and CACE estimates were unbiased with random non-compliance.
• IV was robust to symmetrically dependent non-compliance; PP was reliable when controls lacked intervention access.
• ITT was less biased only when the intervention had no effect.

Conclusions:

• Standard ITT analysis is unreliable for estimating moderate to large treatment effects in the presence of non-compliance.
• Alternative methods (AT, PP, IV, CACE) can yield unbiased or less biased estimates.
• The choice of analysis method should be tailored to the specific non-compliance patterns observed.