Clustering of COVID-19 testing and incidence time series: a method to understand epidemic dynamics at the territorial scale.

INTRODUCTION: The COVID-19 pandemic highlighted socioeconomic disparities in France. Health measures were used to control the epidemic at national and local levels. We studied the temporal patterns of COVID-19 data at the sub-municipal level in Marseille. METHODS: We obtained socioeconomic and COVID-19 data (testing and incidence rates) at the IRIS level (sub-municipal areas <5000 inhabitants) in Marseille. We built socioeconomic profiles using principal component analysis and clustering. We smoothed COVID-19 testing and incidence time series using functional data transformation, then grouped them by combining functional principal component analysis (FPCA) with clustering. We compared these clusters with the socioeconomic profiles, and examined testing in deprived IRIS with and without health mediation interventions. RESULTS: We identified four socioeconomic profiles (deprived, intermediate-employee-family, intermediate-active-city-center, and privileged) and grouped the IRIS into nine testing clusters and eight incidence clusters. Two testing clusters, covering 50% of deprived IRIS and 38.1% of intermediate-employee-family IRIS, showed higher testing between June and October 2021 (health pass period). One incidence cluster, covering 52.9% of deprived IRIS, had high incidence between June and October 2021, then low incidence from March 2022 (no health measures). Deprived IRIS with health mediation interventions (31.4% of deprived IRIS) had higher testing rates from June 2021 onward than those without. CONCLUSION: This study showed that testing and incidence patterns coincided with public health measures (including health mediation interventions) and with socioeconomic profiles, suggesting a differential association across areas. This method is a useful tool to identify priority areas for targeted public health interventions during epidemics.