Addressing missing data in randomized clinical trials: A causal inference perspective.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Causal Inference

Background:

• Randomization in clinical trials is key to preventing selection bias.
• Selective attrition (participant dropout) can reintroduce bias.
• This study examines missing outcome data under various scenarios: missing completely at random (MCAR), missing at random (MAR), and missing not at random (MNAR).

Purpose of the Study:

• To adopt a causal inference perspective for analyzing missing outcome data in clinical trials.
• To provide empirical strategies for estimating average treatment effects (ATE) and improving estimator precision.
• To introduce and evaluate Random Forest Lee Bounds (RFLB) for addressing selective attrition and refining ATE intervals.

Main Methods:

• Utilizing a causal inference framework to analyze selective attrition.
• Proposing Random Forest Lee Bounds (RFLB) as an empirical strategy.
• Employing simulated attrition data to test the RFLB method.

Main Results:

• Identifying assumptions for causal inference are difficult to verify empirically when assuming MCAR or MAR.
• Missing outcome data in clinical trials should be treated as potentially MNAR.
• RFLB significantly tightens average treatment effect intervals by using continuous and discrete attrition predictors.

Conclusions:

• Bounding approaches are essential to account for selective attrition in randomized clinical trials.
• RFLB estimates provide more informative intervals than point estimates derived from MCAR or MAR assumptions.
• Acknowledging uncertainty due to selective attrition is crucial for valid causal inference.