Correcting for non-compliance in randomized trials: an application to the ATBC Study

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Epidemiology

Background:

• Non-compliance in longitudinal studies complicates treatment effect estimation.
• The Alpha-Tocopherol, Beta-Carotene Cancer Prevention (ATBC) Study provides a relevant example with significant participant dropout.
• Intention-to-treat analysis in ATBC suggested increased mortality risk with beta-carotene, necessitating evaluation of treatment actually received.

Purpose of the Study:

• To discuss and compare different methods for estimating treatment effects in longitudinal placebo-controlled trials with non-compliance.
• To evaluate the 'intention-to-treat', 'as-treated', and g-estimation approaches.
• To apply these methods to the ATBC Study data to estimate the effect of beta-carotene actually received.

Main Methods:

• A simple model for treatment action was employed.
• Comparison of 'intention-to-treat', 'as-treated', and g-estimation methods.
• Simulation study to assess method performance under various non-compliance scenarios.
• Analysis of ATBC Study data using the discussed estimation methods.

Main Results:

• The study outlines methodologies for handling non-compliance in treatment effect estimation.
• Simulation results provide insights into the behavior of different estimation approaches.
• Application to ATBC Study data allows for a nuanced interpretation of beta-carotene's effect.

Conclusions:

• Accurate estimation of treatment effects in the presence of non-compliance requires careful methodological consideration.
• The choice of estimation method can influence the observed treatment effect, particularly in longitudinal trials.
• Further analysis using methods beyond intention-to-treat is valuable for understanding the impact of treatments actually received.