Statistical inference for time-to-event data in non-randomized cohorts with selective attrition

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Survival Analysis

Background:

• Multi-season clinical trials with a "randomize-once" strategy face challenges with selective patient attrition.
• Unbiased estimation of survival functions and hazard ratios is crucial for accurate treatment effect assessment.

Purpose of the Study:

• To develop and validate a statistical method for unbiased survival analysis in multi-season clinical trials.
• To address selective attrition in non-randomized cohorts using advanced weighting techniques.

Main Methods:

• Development of an inverse probability of treatment weighting (IPTW) method utilizing season-specific propensity scores.
• Application of bootstrap variance estimators to handle weight randomness and within-patient correlations.
• Validation through simulation studies and analysis of the INVESTED trial data.

Main Results:

• The proposed IPTW method with bootstrap variance estimators yields unbiased estimates of survival functions and hazard ratios.
• Simulation studies confirm the validity of inferences in Kaplan-Meier estimates and Cox proportional hazard models.
• The method effectively accounts for selective attrition in multi-season trial designs.

Conclusions:

• The developed inverse probability of treatment weighting method provides a robust approach for analyzing multi-season clinical trial data.
• This methodology enhances the reliability of survival analyses, particularly when dealing with patient attrition.
• The approach is applicable to various survival analysis models, including Kaplan-Meier and Cox models.