RANDOMIZATION INFERENCE FOR CLUSTER-RANDOMIZED TEST-NEGATIVE DESIGNS WITH APPLICATION TO DENGUE STUDIES: UNBIASED...

Area of Science:

• Epidemiology and Biostatistics
• Infectious Disease Control

Background:

• Dengue is a major global health threat identified by the WHO.
• The Applying Wolbachia to Eliminate Dengue (AWED) study used a novel cluster-randomized test-negative design (CR-TND) for dengue control.
• CR-TND offers cost-efficiency through passive surveillance compared to traditional cluster-randomized trials.

Purpose of the Study:

• To investigate the statistical assumptions and properties of CR-TND under a robust randomization inference framework.
• To address potential biases and inflated type I errors in CR-TND analysis when healthcare-seeking behavior differs across clusters.
• To propose and validate a novel statistical method for accurate inference in CR-TND.

Main Methods:

• Analysis of CR-TND using a randomization inference framework.
• Development and application of a log-contrast estimator to adjust for covariates and varying healthcare-seeking behavior.
• Extension of methods to accommodate partial intervention compliance and stepped-wedge designs.
• Validation through simulation studies and reanalysis of the AWED study data.

Main Results:

• Identified bias and inflated type I error in current CR-TND analysis methods when differential healthcare-seeking behavior varies across clusters.
• The proposed log-contrast estimator effectively eliminates bias and improves precision.
• The extended methods successfully handle partial compliance and stepped-wedge designs.

Conclusions:

• The proposed log-contrast estimator enhances the validity and reliability of CR-TND for dengue surveillance.
• The developed statistical framework provides robust inference for complex cluster-randomized trial designs.
• These advancements contribute to more effective and cost-efficient strategies for controlling dengue and similar infectious diseases.