Modelling the COVID-19 epidemics in Brasil: Parametric identification and public health measures influence

Renato Machado Cotta; Carolina Palma Naveira-Cotta; pierre magal

The copyright holder for this preprint (which was not peer-reviewed) is . https://doi.org/10.1101/2020.03.31.20049130 doi: medRxiv preprint the disease itself, such as in [14] [15] [16] [17] . For this reason, we have also implemented a statistical inverse analysis with the full dataset of China, but now seeking the estimation of five parameters, so as to simultaneously estimate the average times (1/ν and 1/η). Both uniform and Gaussian distributions were adopted for the two new parameters, with initial guesses of 1/ν=7 days and 1/η=7 days, and N=29 days, as employed in [6] . Table 3 provides the estimated values and 95% confidence intervals for all five parameters, with Gaussian priors for the two average times with data obtained from [14, 17] . The most affected parameter in comparison with the previous estimates is the average time 1/η, which is also the one with widest confidence interval. This behaviour is also evident from the Markov chains for this parameter, now simultaneously estimated. Figure 5 compares the theoretical predictions with the model incorporating the five estimated parameters as in Table 3 , against the full CR(t) dataset for China, confirming the improved agreement.