ABSTRACT
The exposure-lag response of air temperature on daily COVID-19 incidence is unclear and there have been concerns regarding the robustness of previous studies. Here we present an analysis of high spatial and temporal resolution using the distributed lag non-linear modelling (DLNM) framework. Utilising nearly two years' worth of data, we fit statistical models to twelve Italian cities to quantify the delayed effect of air temperature on daily COVID-19 incidence, accounting for several categories of potential confounders (meteorological, air quality and non-pharmaceutical interventions). Coefficients and covariance matrices for the temperature term were then synthesised using random effects meta-analysis to yield pooled estimates of the exposure-lag response with effects presented as the relative risk (RR) and cumulative RR (RRcum). The cumulative exposure response curve was non-linear, with peak risk at 15.1 °C and declining risk at progressively lower and higher temperatures. The lowest RRcum at 0.2 °C is 0.72 [0.56,0.91] times that of the highest risk. Due to this non-linearity, the shape of the lag response curve necessarily varied by temperature. This work suggests that on a given day, air temperature approximately 15 °C maximises the incidence of COVID-19, with the effects distributed in the subsequent ten days or more.