Mining relationships between transmission clusters from contact tracing data: An application for investigating COVID-19 outbreak.

Objective Contact tracing of reported infections could enable close contacts to be identified, tested, and quarantined for controlling further spread. This strategy has been well demonstrated in the surveillance and control of COVID-19 (coronavirus disease 2019) epidemics. This study aims to leverage contact tracing data to investigate the degree of spread and the formation of transmission cascades composing of multiple clusters. Materials and methods An algorithm on mining relationships between clusters for network analysis is proposed with 3 steps: horizontal edge creation, vertical edge consolidation, and graph reduction. The constructed network was then analyzed with information diffusion metrics and exponential-family random graph modeling. With categorization of clusters by exposure setting, the metrics were compared among cascades to identify associations between exposure settings and their network positions within the cascade using Mann-Whitney U test. Results Experimental results illustrated that transmission cascades containing or seeded by daily activity clusters spread faster while those containing social activity clusters propagated farther. Cascades involving work or study environments consisted of more clusters, which had a higher transmission range and scale. Social activity clusters were more likely to be connected, whereas both residence and healthcare clusters did not preferentially link to clusters belonging to the same exposure setting. Conclusions The proposed algorithm could contribute to in-depth epidemiologic investigation of infectious disease transmission to support targeted nonpharmaceutical intervention policies for COVID-19 epidemic control.