Competing Risk Modeling with Bivariate Varying Coefficients to Understand the Dynamic Impact of COVID-19

Area of Science:

• Biostatistics
• Epidemiology
• Public Health

Background:

• The COVID-19 pandemic severely affected patients with end-stage renal disease requiring dialysis.
• Preliminary analyses showed time-varying impacts of COVID-19 on dialysis patient outcomes.
• Existing models were insufficient to capture these complex dynamics.

Purpose of the Study:

• To develop and validate a novel statistical model for analyzing competing risks in dialysis patients during the COVID-19 pandemic.
• To quantify the dynamic effects of COVID-19 on hospital readmissions and deaths in end-stage renal disease patients.
• To assess the temporal variations in COVID-19's impact relative to postdischarge time and pandemic onset.

Main Methods:

• Proposed a bivariate varying coefficient model for competing risks using tensor-product B-splines.
• Developed a proximal Newton algorithm for efficient model fitting on large Medicare dialysis patient datasets.
• Implemented difference-based anisotropic penalization and cross-validation for model stability and parameter selection.

Main Results:

• The proposed model effectively captures the complex, time-varying effects of COVID-19 on dialysis patient outcomes.
• Hypothesis testing confirmed significant variations in COVID-19 effects with postdischarge time and pandemic duration.
• Model performance was validated through applications to Medicare dialysis data and simulation studies.

Conclusions:

• The developed bivariate varying coefficient model provides a robust framework for analyzing time-dependent risks in large patient populations.
• This methodology offers crucial insights into the nuanced impact of pandemics on vulnerable patient groups like those on dialysis.
• The findings underscore the need for dynamic statistical approaches to understand and manage health crises in chronic disease populations.