An ensemble approach to short‐term forecast of COVID‐19 intensive care occupancy in Italian regions

Abstract

The availability of intensive care beds during the COVID‐19 epidemic is crucial to guarantee the best possible treatment to severely affected patients. In this work we show a simple strategy for short‐term prediction of COVID‐19 intensive care unit (ICU) beds, that has proved very effective during the Italian outbreak in February to May 2020. 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 nonstationary integer autoregressive methodology. Optimal weights are estimated using a leave‐last‐out rationale. The approach has been set up and validated during the first epidemic wave in Italy. A report of its performance for predicting ICU occupancy at regional level is included.

A spatio-temporal model based on discrete latent variables for the analysis of COVID-19 incidence

“…Due to very limited pre-existing immunity, SARS-CoV-2 has the potential to be highly infectious. Simultaneously, pathogenicity is high enough to generate a proportion of severe cases( Buss et al, 2021 , DelSole et al, 2020 ) that can overwhelm health systems( Grasselli et al, 2020 , Farcomeni et al, 2021a ) when prevalence is high. Monitoring the epidemics is therefore a priority for policy, planning, and resource allocation.…”

Real-time forecasting of COVID-19 bed occupancy in wards and Intensive Care Units

“…In order to estimate the expected values at day s + t in (6), the average of the number of patients on day s + t at both departments is taken over all simulations runs. Next, in order to estimate the boundaries in (7) for a given day s + t, the respective empirical quantiles are taken over the simulated occupancy on that day for both departments. Furthermore, in order to determine the expressions in ( 9) and ( 10), the maximum occupancy is determined for both departments from the forecast day (time s) until the end of the forecasting horizon (time s + t).…”

“…With data on COVID-19 patients becoming more and more available, prediction of the infection rate a few days ahead of time [10,31], and of the LoS [22] is possible. Predictions of the number of hospitalised COVID-19 patients based on regression methods are, e.g., reported in [7,8,18]. The LoS distribution of COVID-19 ICU patients in the United Kingdom is fitted to probability distributions in [28].…”

“…Forecasting the number of hospitalised COVID-19 patients is difficult [12]. As stated above, various approaches exist based on regression methods [7,8,10,18,31], and on estimation of the LoS [22,28]. The key to the accuracy of our forecast is that admissions are predicted according to a Richards' curve, which is proven to be an accurate predictor for the number of COVID-19 infections [16,31], while the joint effect of the arrival rates and the LoS is taken into account via the underlying network of infinite server queues.…”

Safely opening schools: artificial intelligence techniques to control transmission of COVID-19

“…This approach has been taken with 80% of American counties; however, further work can be done to precisely include more variables in the model and specify the said angles. 20 Testing is one such variable that is often not included in models and is a major factor: Comprehensive testing as early as 31 December, 2019, in Vietnam and South Korea along with quarantine laws, flight bans and limited proximity on flights have suppressed their COVID-19 spread by providing more targeted treatment, despite raising detected cases and prevalence. Bayesian analysis utilizing total testing variability as an impacting factor on the total increases in prevalence showed that there was a decrease in infectivity in Orange County following a stay at home order.…”