|Abstract:|| In the talk, we briefly discuss the main epidemiological features of SARS-CoV-2 one year into the pandemic, giving also a short account of the public data available for Italy and of the main limits of lay analyses.
We then discuss an accurate method for short-term forecasting ICU occupancy at local level. Our approach is based on an optimal ensemble of two simple methods: a generalized linear mixed regression model which pools information over different areas, and an area-specific non-stationary integer autoregressive methodology. Optimal weights are estimated using a leave-last-out rationale.
Daily predictions between February 24th and November, 27th 2020 have a median error of 3 beds (third quartile: 8) at regional level, with coverage of 99% prediction intervals that exceeds the nominal one.
Finally we present a different method based on a modified non-linear GLM for each indicator, including the potential effect of exogenous variables, based on appropriate distributional assumptions and a logistic-type growth curve. This allows us to accurately predict important characteristics of the epidemic (e.g., peak time and height).
Based on joint works with Pierfrancesco Alaimo di Loro, Fabio Divino, Giovanna Jona Lasinio, Gianfranco Lovison, Antonello Maruotti, Marco Mingione|