Advancing Interpretable Regression Analysis for Binary Data: A Novel Distributed Algorithm Approach.

Area of Science:

• Biostatistics
• Epidemiology
• Distributed Learning

Background:

• Sparse data bias is a significant challenge in analyzing rare binary outcomes.
• Existing two-step meta-analysis methods can reduce but not eliminate bias in effect estimation.

Purpose of the Study:

• To propose a novel one-shot distributed algorithm, ODAP-B, for unbiased relative risk estimation in binary data analysis.
• To evaluate the performance of ODAP-B against traditional meta-analysis using simulations and real-world data.

Main Methods:

• ODAP-B employs a modified Poisson regression for binary data within a distributed learning framework.
• The algorithm is communication-efficient and privacy-preserving, utilizing aggregated data.
• A robust variance estimator is incorporated for reliable inference.

Main Results:

• ODAP-B provided more accurate relative risk estimates compared to the two-step meta-analysis method across various outcomes.
• Simulations and case studies, including post-acute sequelae of SARS-CoV-2 infection in children, demonstrated ODAP-B's effectiveness.
• The method proved superior in mitigating sparse data bias for rare binary outcomes.

Conclusions:

• ODAP-B is an effective distributed learning algorithm for Poisson regression, particularly for rare binary outcomes.
• The algorithm offers a communication-efficient and privacy-preserving solution for unbiased effect estimation.
• ODAP-B enhances the analysis of sparse datasets in epidemiological and biostatistical research.