Informative Censoring-A Cause of Bias in Estimating COVID-19 Mortality Using Hospital Data

Area of Science:

• Epidemiology
• Biostatistics
• Infectious Diseases

Background:

• Retrospective observational studies are crucial for analyzing COVID-19 treatment outcomes.
• Convalescent plasma therapy has been investigated as a potential treatment for hospitalized COVID-19 patients.
• Informative censoring can introduce bias in survival analyses of treatment effectiveness.

Purpose of the Study:

• To evaluate the impact of informative censoring on COVID-19 treatment outcome analyses.
• To assess the performance of Inverse Probability of Censoring Weighting (IPCW) in correcting for bias in observational studies.
• To compare IPCW-adjusted results with traditional survival analysis methods.

Main Methods:

• Utilized a database of hospitalized COVID-19 patients, comparing those who received convalescent plasma with a control group.
• Applied Inverse Probability of Censoring Weighting (IPCW) to adjust for informative censoring.
• Employed unadjusted Kaplan-Meier curves and Cox proportional hazard models for comparative analysis.

Main Results:

• Unadjusted Kaplan-Meier curves overestimated overall mortality compared to IPCW analysis.
• Cox proportional hazard models underestimated hazard ratios for older age groups versus the youngest when not using IPCW.
• IPCW analysis demonstrated stabilizing weights, improving the reliability of estimates based on hospital admission data.

Conclusions:

• IPCW is a valuable statistical method for correcting bias due to informative censoring in COVID-19 treatment outcome studies.
• Traditional survival analyses without appropriate bias correction can lead to inaccurate estimations of mortality and risk.
• Accurate statistical methods are essential for reliable interpretation of treatment effectiveness in clinical research.