Data-Driven Study of the the COVID-19 Pandemic via Age-Structured Modelling and Prediction of the Health System Failure in Brazil amid Diverse Intervention Strategies

Askery Canabarro; Elayne Tenorio; Renato Martins; Lais Martins; Samurai Brito; Rafael Chaves

The calibration of the model requires robust data so that the model parameters can be as realistic as possible. In the absence of such data for the Brazilian case, we used some data reported in studies with the largest number of individuals with an age-distributed fatality rate (γ i ) [1] and the percentage of persons undergoing critical intense care (c i ) [2] . For the recovery rate (α i ) we use the assumption that α i = (1 − γ i ) * r, where r is the overall fraction of recovery in the closed cases, known so far to be r = 0.82, meaning that 82% of those who did not succumb to the disease are now healed [7] . This does not imply a overall death ratio of 18%, since it accounts only for closed cases, in which one can compute statistics. Although this fraction could change over time, the statistics is reliable since the number of total closed cases is N closed = 191, 623, roughly 25% of the confirmed cases at the time of writing. The parameter β i describes the efficacy of the infection process and can be measure indirectly. Our first assumption is that this efficacy depends weakly on the age group, therefore β i = β is a constant vector. The effective value of β can be computed as R 0 = β αi+γ i , where x stands for the mean value of the variable x and R 0 reproduction number already defined and calculated to be in the range 1.5 − 6.0 in many countries [9, 11] . We will estimate the R 0 by performing a fit in our model with current data available. Given the current control measures in Brazil (a combination of CSU, HVQ and SD60+) our current g i is the concatenation of the corresponding columns in Table II , taking the lowest values.