Semiparametric marginal methods for clustered data adjusting for informative cluster size with nonignorable zeros

Area of Science:

• Biostatistics
• Clinical Epidemiology
• Longitudinal Data Analysis

Background:

• Clustered and longitudinal data are common in clinical research, often collected via real-time monitoring.
• Cluster size can be informative, potentially correlating with disease status.
• Existing methods often exclude nonignorable zero-sized clusters, losing valuable data.

Purpose of the Study:

• To develop statistical methods that incorporate informative nonignorable zero-sized clusters.
• To improve the analysis of clustered data by utilizing information from both event-free and event-occurring participants.
• To address limitations in current statistical approaches for longitudinal and clustered health data.

Main Methods:

• Proposed weighted within-cluster resampling (WWCR) method.
• Introduced asymptotically equivalent dual-weighted generalized estimating equations (WWGEE).
• Employed inverse probability weighting techniques to handle informative cluster sizes.

Main Results:

• Theoretically established asymptotic properties of the proposed methods.
• Demonstrated the finite-sample behavior through extensive simulations.
• Showcased advantageous performance compared to existing approaches in the ASSESS-AKI study.

Conclusions:

• The WWCR and WWGEE methods effectively utilize data from informative zero-sized clusters.
• These novel methods offer improved statistical power and accuracy in analyzing clustered health data.
• The findings have significant implications for clinical trials and observational studies with complex data structures.