Nonparametric inference of general while-alive estimands for recurrent events

Area of Science:

• Biostatistics
• Clinical Trials
• Epidemiology

Background:

• Recurrent events (e.g., hospitalization) in the presence of death pose challenges for traditional statistical analysis.
• Existing methods can unfairly penalize longer survival, as extended survival often correlates with more adverse events.

Purpose of the Study:

• To develop a novel statistical framework for measuring treatment effects on recurrent events, accounting for patient survival time.
• To introduce the "while-alive" event rate as a causal estimand that adjusts for length of survival.

Main Methods:

• Defined a general class of estimands based on user-specified loss functions, focusing on average loss per unit time alive.
• Developed a nonparametric estimator for the loss rate using cumulative loss and restricted mean survival time.
• Derived the influence function for variance estimation and hypothesis testing.

Main Results:

• The "while-alive" approach corrects for the underestimation of treatment effects caused by increased patient survival.
• Simulations and real-world heart failure trial data demonstrated increased statistical power using the proposed method.
• The method provides a more accurate and interpretable measure of treatment efficacy in the presence of competing risks.

Conclusions:

• The "while-alive" event rate offers a statistically robust and causally interpretable method for analyzing recurrent events in clinical research.
• The proposed methodology enhances the power to detect true treatment effects, particularly in studies where treatments improve survival.
• An R package (WA) is available for implementing these advanced statistical techniques.