Sensitivity analysis for calibrated inverse probability-of-censoring weighted estimators under non-ignorable dropout

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Causal Inference

Background:

• Inverse probability-of-censoring weighting (IPCWE) is common for handling dropout in longitudinal studies.
• Maximum likelihood estimation for IPCWEs can lead to inefficiency and instability.
• Existing calibrated IPCWEs rely on unverifiable assumptions and lack sensitivity analysis for non-ignorable dropout.

Purpose of the Study:

• To develop a sensitivity analysis approach for calibrated IPCWEs under non-ignorable dropout.
• To provide methods for assessing the impact of unobserved dropout mechanisms.
• To enhance the robustness of statistical inference in longitudinal studies with missing data.

Main Methods:

• Developed a novel sensitivity analysis framework for calibrated IPCWEs.
• Proposed an accelerated computation technique for bootstrap and jackknife confidence intervals.
• Utilized simulation studies to evaluate finite-sample performance.
• Applied methods to an international systemic lupus erythematosus cohort study.

Main Results:

• The proposed sensitivity analysis effectively evaluates the impact of non-ignorable dropout on calibrated IPCWEs.
• The computational speed-up facilitates practical implementation of sensitivity analyses.
• Simulations demonstrated the reliability and performance of the developed methods.
• The approach was successfully applied to real-world cohort data.

Conclusions:

• The study provides a crucial tool for sensitivity analysis in calibrated IPCWEs, addressing limitations of existing methods.
• The developed approach enhances the reliability of findings from longitudinal studies with potential non-ignorable dropout.
• This work offers practical guidance and computational tools for researchers in biostatistics and related fields.