Comparing Performance of Methods to Deal With Differential Attrition in Randomized Experimental Evaluations.

Area of Science:

• Biostatistics
• Epidemiology
• Clinical Trials

Background:

• Differential attrition in randomized controlled trials (RCTs) can compromise causal inference.
• Existing methods like inverse probability weighting and bounding are used to address this bias.
• Lottery-based randomization was used to generate study datasets.

Purpose of the Study:

• To compare the performance of various attrition correction methods.
• To evaluate methods in datasets with varying levels of differential attrition.
• To assess the impact of assumption violations on bounding approaches.

Main Methods:

• Performance assessment of correction methods in a dataset with significant differential attrition.
• Conducting simulation analyses to evaluate method robustness.
• Comparing results from a problematic dataset and an augmented dataset.

Main Results:

• Attrition correction methods demonstrated poor performance in the dataset with considerable differential attrition.
• Simulation analyses revealed that violated assumptions negatively impacted the performance of bounding methods.
• The effectiveness of correction methods is highly dependent on the degree of attrition and assumption adherence.

Conclusions:

• Verification of underlying assumptions for attrition correction methods is crucial.
• When assumption verification is not feasible, these methods should be applied with extreme caution.
• The reliability of causal inference in RCTs with differential attrition depends on careful method selection and validation.