The association between air temperature and covid-19 incidence is unclear, particularly regarding lag effects. Here we address this research gap using high resolution data from Italy.
We obtained daily covid-19 cases, populations at risk, and mean daily air temperature from 97 Italian cities for the period 24 February through 21 September 2020. We fitted a mixed-effects distributed lag non-linear model, presenting the effects as relative risks (RR) and cumulative relative risks (RRcum).
Negative increments in mean daily temperature produced approximately inverted U- shaped lag-responses, though for large positive increments in temperature, the peak RR occurred at the maximal lag of 14 days. The temperature exposure response curves generally showed an increased RR with increasing temperature, though the shape varied according to the lag period. Positive and negative increments in temperature caused increases and decreases in the RRcum respectively, though the plateau effect for negative increments was not observed above small positive increments in temperature.
We postulate that latent variables correlated with temperature, such as frequency and duration of social activities, are the underlying cause of our observed trends. Nonetheless, our statistical model can be utilised to forecast cumulative covid-19 incidence rates assuming specified air temperature increments at the city level.