AI for predicting emergencies: the theoretical impact of TA Event Detector tool on Covid-19 pandemic

Abstract Background The COVID-19 pandemic highlighted the need of innovative methods for predicting the spread of communicable diseases and taking prompt containment measures. TrustAlert is a collaborative project aimed to develop AI tools to help policy makers and healthcare services in facing emergencies. Within the project, an integrated platform enabling the analysis of near real-time stream of data arising from healthcare structured databases and unstructured data was developed. The present study evaluates the theoretical impact of the anticipation of containment measures to the alert dates estimated by the Trust Alert Event Detector (ED) tool on the reduction of infections during COVID-19 pandemic in Piedmont region, Italy. Methods A compartmental model was used to evaluate the dynamics of the first COVID-19 wave in the region. The model was fit to actual data to assess its performance in predicting the real pandemic trends, then the hypothetical trends were simulated if the containment measures were adopted on the dates of the ED alerts. The model parameters were calculated from the data released daily by the Department of Civil Protection, and their modifications were estimated based on literature on effect of containment measures. Results Introducing lockdown measures and mandatory use of N95 masks on the first date of ED alert (24th February) would have led to a reduction of the infections of 85% and 99%, respectively. Introducing the measures on the second ED alert (2nd March) would have led to a reduction of 66% and 98.4%, respectively. Conclusions A technical platform able to predict real-time health emergencies may anticipate medical needs in healthcare crises. This kind of tool can be useful to guide response plans, pending up-to-date information and timely data elaboration. The availability of the ED tool at the time of COVID-19 pandemic if followed by prompt containment measures, would have had a relevant impact on the dynamics of the infection. Key messages • AI early warnings and forecasting tools based on near real-time structured and unstructured data can predict emergencies and guide policy makers and healthcare services to develop prompt responses. • During COVID-19 pandemic, both the adoption of lockdown and of mandatory use of N95 masks on alert dates as estimated by the ED tool would have had a relevant impact on the spread of the infections.