Sequential modeling for a class of reference-based imputation methods in clinical trials with quantitative or binary...

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Missing data in longitudinal clinical trials complicates treatment effect estimation.
• Standard methods often assume data are missing at random, necessitating sensitivity analyses for missing not at random scenarios.
• Existing reference-based imputation methods like Jump to Reference (J2R) and Copy Reference (CR) have limitations.

Purpose of the Study:

• To propose a novel, comprehensive framework of reference-based imputation methods for sensitivity analysis in clinical trials.
• To extend existing methods by considering a spectrum of potential treatment carry-over effects after treatment deviation.
• To provide a flexible approach applicable to both quantitative and categorical longitudinal outcomes.

Main Methods:

• Developed a wide class of reference-based imputation methods, encompassing J2R and CR as boundary cases.
• The framework models the potential carried-over effect of an investigative treatment after deviation.
• Demonstrated implementation through sequential modeling, suitable for diverse outcome types.

Main Results:

• The proposed methods provide a unified framework for sensitivity analysis.
• The sequential modeling approach allows for practical application in various clinical trial settings.
• Causal-inference arguments and numerical examples validate the performance of the new methods.

Conclusions:

• The novel reference-based imputation methods offer a robust approach to handling missing data in longitudinal clinical trials.
• This framework enhances the reliability of treatment effect estimation under missing not at random assumptions.
• The methods are versatile, applicable to both quantitative and categorical outcomes, and implementable via sequential modeling.