An Expectation-Maximization Algorithm for Including Oncological COVID-19 Deaths in Survival Analysis

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• COVID-19 deaths in oncology trials complicate standard survival analysis.
• Cancer patients face higher COVID-19 mortality risk, making these deaths informative.
• Traditional survival estimators require non-informative censoring, which COVID-19 deaths violate.

Purpose of the Study:

• To develop a method for consistently incorporating COVID-19 deaths into survival estimates for oncology trials.
• To propose an algorithmic approach that accounts for the informative nature of COVID-19 deaths in cancer patients.
• To provide a statistically sound method that enhances the accuracy of survival function estimation.

Main Methods:

• Developed the Covid-Death Mean-Imputation (CoDMI) algorithm, based on Expectation-Maximization (EM).
• Utilized a counterfactual approach to impute virtual lifetimes for COVID-19 deaths.
• Extended Greenwood's formula to calculate adjusted confidence intervals for survival estimates.

Main Results:

• The CoDMI algorithm generates a completed dataset for survival analysis.
• CoDMI provides more accurate survival estimates compared to naive approaches (censoring or disease death).
• Simulation studies show CoDMI is largely unbiased and outperforms alternative methods.

Conclusions:

• The CoDMI algorithm offers a robust solution for handling COVID-19 deaths in oncology survival analysis.
• CoDMI enhances the reliability of survival estimates by treating COVID-19 deaths as informative.
• An R implementation of CoDMI is available for practical application in clinical research.