Mathematical modeling of COVID-19 containment strategies with considerations for limited medical resources

Brydon Eastman; Cameron Meaney; Michelle Przedborski; Mohammad Kohandel

Note that χ then represents the mean squared errors of fitting (I(t), R(t), D(t)) to (I t − R t − D t , R t , D t ) and fitting (I(t) + R(t) + D(t), R(t), D(t)) to (I t , R t , D t ). We provide reasonable bounds for this fitting process by assuming that 0 ≤ E 0 , I 0 ≤ 20, 10 −1 ≤ γ ≤ 0.8, and 20 −1 ≤ α ≤ 0.3. In practice, the bounds we set for E 0 and I 0 are likely too large, however restricting the bounds did not change the optimizer discovered. The γ bound was chosen based on reports that the latent period of the disease has a median value of 5.1 days with a 95% Confidence Interval of 4.5 to 5.8 days [12] . We chose our range for γ −1 to be between 1.25 days and 10 days, in order to ensure that we capture the most likely parameter values. Similarly, the bounds on α were chosen based on reports that the mean infectious period is around 7-14 days (for instance, in [13] they suggest a quarantine time of 14 days from first symptom onset, in order to prevent further infections). Like with γ, we chose the bounds here to be larger than necessary in order to capture the most likely parameter value. Other reports provide ranges for the latent period and mean infectious period that fall within these bounds, see [14, 15, 16, 17] .