# Generalizing an outbreak cluster detection method for two groups: an application to rabies

## TL;DR

This paper improves a method for identifying disease clusters by considering differences between groups, and applies it to rabies in Tanzania to better understand transmission patterns.

## Contribution

The paper extends a graph-based cluster detection method to account for group-specific reporting and transmission differences in multi-host systems.

## Key findings

- Domestic animals had higher reporting probabilities than wildlife in rabies cases.
- There was no significant difference in mean transmission distance between groups.
- Inter-species transmission occurred frequently in the rabies data.

## Abstract

Identifying linked cases of an infectious disease can improve our understanding of its epidemiology by distinguishing sustained local transmission from frequent introductions with little onward transmission. This evidence can, in turn, inform decisions on interventions. Knowledge of epidemiological distributions and reporting probabilities is key in identifying linked cases. However, with multi-host pathogens quantitative differences between hosts may need consideration. In this study, an existing graph-based approach to detecting outbreak clusters was extended to allow for group-specific reporting probabilities and epidemiological distributions and to assess the level and importance of assortative mixing. This method was applied to data on animal rabies cases in Tanzania. Group-specific differences in reporting probabilities and epidemiological distributions and the level of assortative mixing had a marked impact on the size and composition of clusters. Results of the rabies cases analysis supported higher reporting probabilities in domestic animals than wildlife, no difference in mean transmission distance between groups, and frequent inter-species transmission. The method described here could be applied to other multi-host or multi-group systems in which heterogeneities in reporting probabilities, distributional parameters and/or levels of mixing exist between groups. This would allow more accurate characterization of transmission dynamics and thus facilitate implementation of more effective interventions.