Sense and sensitivity when correcting for observed exposures in randomized clinical trials

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Intent-to-treat analyses in randomized clinical trials (RCTs) can be biased due to patient non-compliance with assigned treatments.
• Accurate estimation of treatment efficacy and toxicity is crucial for reliable clinical decision-making.

Purpose of the Study:

• To develop and demonstrate a practical sensitivity analysis framework for RCTs with partial non-compliance.
• To quantify uncertainty in treatment effect estimates arising from non-compliance.

Main Methods:

• Utilized a sensitivity analysis approach, extending methods from missing data contexts (Molenberghs, Kenward, Goetghebeur).
• Introduced Honestly Estimated Ignorance Regions (HEIRs) and Estimated Uncertainty Regions (EUROs) to assess uncertainty.
• Applied the methods to estimate the causal effect of exposure on blood pressure reduction in an RCT with non-compliance.

Main Results:

• The proposed sensitivity analysis effectively quantifies the impact of non-compliance on treatment effect estimates.
• HEIRs and EUROs provide distinct measures of uncertainty stemming from structural assumptions and sampling variation.
• Causal effect estimation in the presence of non-compliance was demonstrated using real-world trial data.

Conclusions:

• Sensitivity analysis is essential for robust interpretation of RCTs involving non-compliance.
• The HEIRs and EUROs framework offers a valuable tool for assessing the plausibility of treatment effects under various compliance scenarios.
• This approach enhances the reliability of causal inference in clinical research where perfect compliance is rare.