Efficient Multiple Imputation for Sensitivity Analysis of Recurrent Events Data with Informative Censoring.

Area of Science:

• Biostatistics
• Clinical Trials Methodology
• Survival Analysis

Background:

• Missing data are prevalent in clinical trials, particularly affecting recurrent events analysis.
• Patient dropout can lead to undercounting of events, biasing study results.
• The 'missing at random' assumption is often untestable, necessitating robust sensitivity analyses.

Purpose of the Study:

• To develop and validate a control-based multiple imputation method for recurrent events data.
• To address missing data in clinical trials where patients drop out.
• To improve the accuracy of recurrent event outcome analysis under non-random missingness.

Main Methods:

• A control-based multiple imputation technique is proposed, assuming dropouts follow the control group's response profile.
• The 'copy reference' and 'jump to reference' approaches are considered for imputation.
• Data are modeled using a semiparametric proportional intensity frailty model with an unspecified baseline hazard.
• Nonparametric maximum likelihood estimation and inference procedures are developed.

Main Results:

• The proposed multiple imputation method effectively handles missing recurrent events data.
• Simulation studies confirm the method's good performance in practical clinical trial settings.
• The technique was successfully applied to analyze data from two distinct clinical trials.

Conclusions:

• The developed control-based multiple imputation method offers a reliable approach for analyzing recurrent events data with missing observations.
• This method enhances the robustness of statistical inference in clinical trials impacted by patient dropout.
• The findings provide valuable tools for biostatisticians and researchers working with complex longitudinal data.