Usefulness of ecological mobility and socio-economic indicators in SARS-CoV-2 infection modelling: a French case study

IntroductionFollowing its emergence in January 2020, SARS-CoV-2 diffusion occurred for a year with only non-pharmaceutical interventions (NPIs) available as mitigation tools. We aimed to assess the predictive capability of census-based indicators on the infection risk by SARS-CoV-2 in the French Auvergne-Rhone-Alpes region to assist NPIs allocation at the neighbourhood level. MethodsWe aggregated all counts of biologically confirmed cases of SARS-CoV-2 infection at the neighbourhood level between May 2020 and February 2021. 10 census-based ecological covariates were evaluated as predictors of case incidence using a Poisson regression with conditional autoregressive (CAR) spatial effects. Benefits of CAR effects and covariates on model fit were evaluated using pseudo-R{superscript 2} and Morans I statistics. Results438,992 infection cases over 5,410 neighbourhoods among 7,917,997 inhabitants were analysed. The association between covariates and case incidence was inconstant across time and space. Spatial correlation was estimated at high levels. Spatial CAR effects were necessary to improve on the pseudo-R2 and the Morans I statistics compared to the null model (intercept only). ConclusionThe ecological covariates assessed were insufficient to adequately model the distribution of cases at the neighbourhood level. Excess incidence was found mainly in metropolitan areas before the epidemic wave peak.