As you know, we decided to build our own forecast model in order to give an estimate of Italy’s efforts in containing the spread of coronavirus pandemic.

The model results can be visualized in our Covid19 dashboard which is publicly accessible and updated on a daily basis.

Repaid efforts

As of April 16th Italy counts 168,841 total cases, of which 106,607 still active, 40,164 recovered and, unfortunately, 22,170 deaths. These figures sound much better than what we could have expected to happen 4 weeks ago (March) when, accordingly to our model, the forecast for April 16th showed 262,689 total cases and 45,081 deaths.

Thus we can say that lockdown helped in reducing the spread by 35,7% and deaths by more than 50%.

And it sounds as a great success.

Apr.16 Forecasts data as of:
Mar.19 Mar.26 Apr.2 Apr.9 Actual
Total Cases 262,689 228,891 177,591 167,149 168,941
Active Cases 153,065 143,544 109,348 103,345 106,607
Recovered 64,543 52,264 42,719 41,732 40,164
Deaths 45,081 33,083 25,524 22,072 22,170
Actual/forecast gap as of:
Mar.19 Mar.26 Apr.2 Apr.9
Total Cases -35.7% -26.2% -4.9% +1.1%
Active Cases -30.4% -25.7% -2.5% +3.2%
Recovered -37.8% -23.2% -6.0% -3.8%
Deaths -50.8% -33.0% -13.1% +0.4%

Approaching the peak

The forecasts made 4, 3 and 2 weeks before April 16th have over-estimated the number of cases and deaths. Therefore, since the model does not account for lockdown measures, we can intepret these differences as a positive effect of the lockdown itself.

The deviation from the forecast made one week before April 16th is in the opposite direction: actual data are worse than expected, even if only slightly. This is likely because the active case curve is flattening and, according to our model, will peak around April 22nd with approximately 109,000 cases. In this context, short-term deviations can be very low and of both signs.

We therefore expect - and hope - that spring finally arrives and everything starts to bloom again.

About the model

We took inspiration from the interesting paper from Yi-Cheng Chen et al. where a time-dependent SIR model has been defined in order to predict the evolution of the Wuhan crisis. The advantage of this approach is to use time-dependent parameters which are updated on a daily basis according to the last available data.

We then applied the SIR model to the Italian Covid data provided by the Civil Protection department of the Italian Government with the COVID-19 repository. Data and results have been encapsulated into a dashboard built using the flexdashboard R package.

From the dashboard it is possible to retrieve forecasts of the epidemiologic dynamics made at every point in time starting from March, 2nd. In this way it is possible to compare actual data with the respective predictions in order to have an idea of the lockdown effectiveness.