A statistical framework for comparing epidemic forests

Area of Science:

• Epidemiology
• Computational Biology
• Biostatistics

Background:

• Inferring transmission pathways during outbreaks is crucial for public health.
• Limited data and complex interactions make transmission inference challenging, often resulting in multiple plausible epidemic forests.
• Currently, no statistical methods exist to formally compare these different epidemic forests.

Purpose of the Study:

• To develop and evaluate a statistical framework for comparing epidemic forests.
• To determine if differences between epidemic forests generated by various methods are statistically significant.

Main Methods:

• Proposed a framework utilizing chi-square tests and permutational multivariate analysis of variance (PERMANOVA).
• Assessed the methods' ability to distinguish simulated epidemic forests based on different offspring distributions.
• Implemented the framework in the R package mixtree.

Main Results:

• Both chi-square tests and PERMANOVA demonstrated perfect specificity in distinguishing forests with 100+ trees.
• PERMANOVA consistently showed higher sensitivity than the chi-square test across various epidemic and forest sizes.
• The study provides the first robust statistical framework for comparing epidemic forests.

Conclusions:

• The proposed framework offers a statistically rigorous approach to compare epidemic forests.
• PERMANOVA is a powerful tool for assessing differences in transmission dynamics inference.
• This work enhances the ability to characterize outbreak transmission and guide interventions.