Evaluation of biases present in the cohort multiple randomised controlled trial design: a simulation study

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Health Services Research

Background:

• Cohort multiple randomized controlled trials (cmRCTs) integrate randomization into clinical practice, enhancing cost-effectiveness and external validity.
• A key challenge in cmRCTs is treatment refusal, which is typically confined to the intervention arm, potentially introducing bias and reducing statistical power.

Purpose of the Study:

• To evaluate the impact of treatment refusal on bias and statistical power in cmRCTs.
• To compare the performance of different analytical methods, including intention-to-treat (ITT), per-protocol (PP), and instrumental variable (IV) approaches, under various refusal scenarios.

Main Methods:

• Simulation studies were conducted to model cmRCTs with time-to-event endpoints.
• The study assessed bias and statistical power using ITT, PP, and two IV methods (two-stage predictor substitution and two-stage residual inclusion).
• Various refusal scenarios, including random and event-risk-related refusal, were simulated.

Main Results:

• Instrumental variable (IV) methods demonstrated reduced bias in estimating causal effects when treatment refusal was present in the intervention arm.
• The two-stage residual inclusion IV method exhibited the best performance, balancing bias reduction and statistical power.
• Effective IV analysis necessitates sample size adjustments based on anticipated and observed refusal rates.

Conclusions:

• The study recommends employing both IV and ITT analyses in individually randomized cmRCTs to capture a range of effect sizes.
• The two-stage residual inclusion method is identified as the optimal IV approach for minimizing bias and maximizing power.
• Adaptive power calculations, updated during trial recruitment based on actual refusal rates, are crucial for ensuring sufficient power for IV analysis.