Direct likelihood inference and sensitivity analysis for competing risks regression with missing causes of failure.

Area of Science:

• Biostatistics
• Survival Analysis
• Epidemiology

Background:

• Competing risks analysis is crucial when multiple failure causes exist.
• Complete cause of failure data is often unavailable in studies.
• Existing semi-parametric models for missing competing risks data under the missing at random (MAR) assumption have limitations.

Purpose of the Study:

• To propose a direct likelihood approach for fully-parametric regression modeling of cause-specific hazards (CSH) and cumulative incidence functions (CIF) under MAR.
• To develop a sensitivity analysis method to assess the robustness of inferences to departures from the MAR assumption.
• To provide a framework applicable to various competing risks regression models.

Main Methods:

• Developed a "direct likelihood" approach for fitting fully-parametric regression models for CSH and CIF under MAR.
• Proposed a sensitivity analysis using pattern-mixture models to evaluate robustness to MAR assumption violations.
• Evaluated the methods through a simulation study and illustrated with a breast cancer clinical trial dataset.

Main Results:

• The direct likelihood approach enables fully-parametric modeling of competing risks with missing data under MAR.
• The sensitivity analysis provides a means to assess the impact of potential MAR assumption departures.
• The methods were successfully demonstrated on real-world clinical trial data.

Conclusions:

• The proposed direct likelihood and sensitivity analysis methods offer a robust framework for analyzing competing risks data with missing failure causes.
• These methods enhance prediction and understanding of competing risks mechanisms, even when data is incomplete.
• The approach is versatile, applicable to both fully-parametric and semi-parametric models for CSH and CIF.