Robustly estimating the COVID19 epidemic curve in northern Italy using all-cause mortality

Abstract

Background: Northern Italy was one of the most impacted areas by COVID. It is now widely assumed that the virus was silently spreading for at least 2 weeks before the first patient was identified. During this silent phase, and in the following weeks when the hospital system was overburdened, data collection was not performed in an accurate enough way to estimate an epidemic curve. With the aim of assessing both the dynamics of the introduction of the virus and the effectiveness of containment measures introduced, we try to reconstruct the epidemic curve using all cause mortality data. Methods: we collected all cause mortality data stratified by age from the national institute of statistics, together with COVID-related deaths data released by other government structures. Using a SEIR model together with estimates of the exposure to death time distribution, we fitted the reproduction number in different phases of the spread at regional level. Results: We estimate a reproduction number of 2.6+/-0.1 before case 1 was identified. School closures in Lombardy lowered it to 1.3. Soft lockdown measures resulted in R<0.8 and no further reductions were observed when a hard lockdown was introduced (e.g. Emilia-Romagna soft lockdown 0.67 +/-0.07, hard lockdown 0.69+/-0.071). Reproduction number for the >75 age range during hard lockdown are consistently higher than for the rest of the population (e.g. 0.98 vs 0.71 in Milan province), suggesting outbreaks in retirement facilities. Reproduction numbers in Bergamo and Brescia provinces starting from March 7th are markedly lower than in other areas with the same strict lockdown measures (Nearby provinces: 0.73, Brescia: 0.52, Bergamo 0.43) supporting the hypothesis that in those provinces a large percentage of the population had already been infected by the beginning of March.