Attrition adjustment in sample size calculation for Intention-to-treat (ITT)-based clinical trials: Evidence from a...

Attrition in randomized clinical trials (RCTs) threatens internal validity and reduces statistical power, challenging the integrity of analyses. Different methods to handle missing data such as last observation carried forward (LOCF), multiple imputation (MI) and per-protocol (PP) vary in bias and impact. Whether to inflate sample size to offset dropout remains debated, necessitating systematic investigation to evaluate effects of missing data strategies and sample size adjustments on trial outcomes.

A Monte Carlo simulation modelled parallel-arm RCTs with correlated baseline and longitudinal continuous outcomes. Four missingness mechanisms, that is, missing completely at random (MCAR), missing at random (MAR), missing not at random (MNAR) and differential dropout, were studied with dropout rates up to 20%. Analytical approaches included PP, LOCF and MI. Scenarios were replicated 1000 times. Key metrics were power, bias, Type I error and sensitivity assessed by tipping point and scenario analyses.

At 20% dropout, MI preserved power at approximately 72%-73% under MCAR, MAR and MNAR mechanisms, surpassing PP (~68%-70%) and LOCF (~59%-61%). Attrition-adjusted sample size restored power to approximately 79%-82% across mechanisms. Differential dropout produced the greatest bias and power reduction for all methods and was the only mechanism showing meaningful Type I error inflation. Tipping-point analyses confirmed that conclusions remain robust to deviations from the MAR assumption.

Combining prospective sample size inflation with MI enhances internal validity and power in RCTs facing attrition. These findings have important implications for clinical trial guidelines, especially in sample size estimation and missing data handling, promoting evidence-based standards to maintain trial integrity.