Reconstructing the temporal dynamics of clustering from cluster surveillance of COVID-19
Background: Clusters, defined as a group of cases that share common place of transmission, have characterized local dynamics of COVID-19 transmission, and Japan is one of a few countries that continued to survey clustering events classified by social settings. Japanese government referred to cluster data to design and implement public health and social measures(PHSM), but little has been clarified with respect to the transmission mechanisms of epidemic that propagates with certain types of clusters. Here we aimed to analyze the temporal dynamics of clustering using the cluster surveillance data in Japan.
Methods: Clusters were classified into healthcare facility, welfare facility, school, working place, recreation event, and eating and drinking establishment. The number of cases who were unlinked was simultaneously recorded. We employed the vector autoregression linear non-gaussian acyclic model (VAR-LiNGAM) to describe the time-series causal mechanisms among different types of clusters as a function of time. The reliability of VAR-LiNGAM was assessed by bootstrapping method.
Results: Eating and drinking establishment and workplace were identified as constituting “upper stream” of the temporal order of clustering, and healthcare and welfare facilities tended to be dead-end. The upper-stream clusters predicted subsequent clusters for about 5 weeks ahead. We reconstructed the chain of those clustering patterns. The effectiveness of PHSM that focused on eating and drinking establishment at nights was assessed.
Conclusion: The present study has shown that empirical data based on cluster surveillance greatly helped identify forthcoming epidemic by exploring the type of clusters.