Confronting COVID-19: Surging critical care capacity in Italy

Jose Manuel Rodriguez Llanes; Rafael Castro Delgado; Morten Gram Pedersen; Pedro Arcos Gonzalez; Matteo Meneghini

We then examined the robustness of this measure over time using provided data by the Italian Civil Protection (17) . Additionally, we investigated by means of visual plots the national rate of daily increments in ICU patients by the daily increment in active confirmed cases, expressed as a percentage (Figure 2 A) . Figure 2 B shows early surging of ICU patients on the first days of the epidemic 24-25 February, and a second surge in ICU use peaking around 6 March. Additional regional analyses (not shown) confirmed that case surges in neighboring regions of Lombardy such as Piemonte and Liguria were contributors for the national rise in ICU cases 10 days later. From March 12, a steady decrease is noted and this is further accentuated after March 18 (Figure 2 B ). As previously reported (15) these rates were mostly within the range of 9% and 12% of active confirmed cases. Based on these observations, we delineated four relevant periods (table 1) . We based our choice of period to calculate the Daily ICU rate on three aspects. First, the two early periods up to 6 March are likely more representative of ICU admission criteria based on clinical realities. Secondly, a higher value will support a more conservative estimate of ICU bed requirement (9.9%). Third, high value consistency across these the first two periods. Subsequently we focused on the modelling strategy to forecast the expected number of active cases. Specifically, we decided not to model the spread of the disease by the conventional SIR (susceptible, infectious, recovered) model, since Italy was the first Western country to adopt immediate quarantine to contrast the diffusion of the virus. For this, we used a SIQR model ( figure 3) , where we consider a sub-population of quarantined individuals (Q). Details on the model and on the related mathematical framework can be found elsewhere (19) . The quarantined patients, once identified, are immediately isolated (either in hospital or at home), thus no longer transmitting the disease. It is worth noticing that we decided not to adopt a SEIR approach (susceptible, infectious, exposed and recovered), since it is well known that undetected asymptomatic individuals can transmit the disease (19) . The model was fitted with data on total confirmed active COVID-19 cases up to 19 March, and its predictive capability was checked with the available data up to 24 March. Results were found to be within the 50 % confidence band (indicating less than 50% variation in R0) in the subsequent 5-days period, indicating that the model works reasonably for planning resources on the mid-term ( figure 4 ). Longer-term predictions make little sense in a period when new drastic measurements are introduced almost daily (the latest was introduced on March 22). On the other hand, thanks to its intrinsic robustness, our model can be used to compare long-term scenarios, thus allowing to visualize projections on the potential effectiveness of the latest measures implemented.