A mathematical dashboard for the analysis of Italian COVID‐19 epidemic data

Abstract

An analysis of the COVID‐19 epidemic is proposed on the basis of the epiMOX dashboard (publicly accessible at https://www.epimox.polimi.it ) that deals with data of the epidemic trends and outbreaks in Italy from late February 2020. Our analysis provides an immediate appreciation of the past epidemic development, together with its current trends by fostering a deeper interpretation of available data through several critical epidemic indicators. In addition, we complement the epiMOX dashboard with a predictive tool based on an epidemiological compartmental model, named SUIHTER, for the forecast on the near future epidemic evolution.

Modelling the COVID-19 epidemic and the vaccination campaign in Italy by the SUIHTER model

“…The model can be initialized at any time t 0 prior to the start of the vaccination campaign (December 27, 2020 in Italy) by using the data (for those compartments for which data are available, namely I , H , T , E ), null initial values for V 1 and V 2 , while, as proposed in ( Parolini et al, 2021b ), the Undetected and Recovered compartments are initialized as where IFR( t ) is the Infection Fatality Ratio , CFR( t ) is the time-dependent Case Fatality Ratio and d = 13 days denotes the confirmation-to-death delay. In ( Parolini et al, 2021b ), we considered a constant IFR computed as the weighted average of the age-specific IFR estimates weighted by the population age structure, under the assumption of equal attack rates across age-groups, as proposed in ( Brazeau et al, 2020 ).…”

“…The model can be initialized at any time t 0 prior to the start of the vaccination campaign (December 27, 2020 in Italy) by using the data (for those compartments for which data are available, namely I , H , T , E ), null initial values for V 1 and V 2 , while, as proposed in ( Parolini et al, 2021b ), the Undetected and Recovered compartments are initialized as where IFR( t ) is the Infection Fatality Ratio , CFR( t ) is the time-dependent Case Fatality Ratio and d = 13 days denotes the confirmation-to-death delay. In ( Parolini et al, 2021b ), we considered a constant IFR computed as the weighted average of the age-specific IFR estimates weighted by the population age structure, under the assumption of equal attack rates across age-groups, as proposed in ( Brazeau et al, 2020 ). A better estimate of the reference IFR( t ) at a specific time can be computed by considering the variable percentage of each age-group among the total infected over time (data available on ( Dati della Sorveglianza integrata COVID-19 in Italia, ) since December 8, 2020), namely: where IFR i denotes the infection fatality ratio for age-group i and q i ( t ) is the percentage of infected at time t belonging to age-group i .…”

“…SUIHTER is a homogeneous model and does not explicitly accounts for age or exposition context structure, thus the restriction coefficient is obtained from the restriction matrix by a weighted mean on the age groups with weights given by the amount of population in each group. Examples of scenario analyses using this approach to quantify the effect of different NPIs have already been presented and discussed in ( Parolini et al, 2021b ). The procedure for computing the restriction coefficient τ for a given NPI scenario is described Appendix A .…”

SUIHTER : a new mathematical model for COVID-19. Application to the analysis of the second epidemic outbreak in Italy

“…The curves show how the prediction in terms of the day of peak occurrence and peak value changes when an increasing number of data are used (the last data day is reported on the horizontal axis). To minimize the effect of daily data noise, the reference value (dashed line) is obtained by smoothing the data with a Savitzky-Golay polynomial smoothing filter of degree 3 [40].…”

“…The results of the simulation of the second wave carried out at the national and regional levels showed the capability of the model in predicting the time evolution accurately in a time frame of 15 days past the data used in the calibration. In longer term predictions the model should account for the possible changes in restriction rules that may occur in the future to supply analyses of different scenarios (as recently done in [40] based on the SUIHTER model).…”