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

Brydon Eastman; Cameron Meaney; Michelle Przedborski; Mohammad Kohandel

Here we present the main findings of the study and the predictions of the standard SEIRD and Distancing-SEIRD models. First, we depict the short-term and longer-term fits to the country-wide data for Canada and China, as well as the fits to COVID-19 epidemiological data from Quebec and Ontario. We use the parameter fits to inform the Distancing-SEIRD model to ultimately make long-term predictions about the impact of social distancing strategies on the dynamics of the pandemic in Quebec and Ontario. The fits to the epidemiological data for China and Canada, obtained with the standard SEIRD model, are depicted in Figure 3 . The full fitting process is described in the Supplementary Information Section B.2. The process involves estimating key parameters from the raw epidemiological data and using machine learning techniques to optimize the remaining parameters. Figure 3 illustrates that the SEIRD model captured the overall trends in the number of infected individuals, the total number of deaths, and the number of recovered individuals. Importantly, with a suitable and realistic choice of parameters, the model captured the peak and the tail of the longer-term temporal data from China. Note that while the data fits for the infected curve for Canada (Figure 3 b) show quantitative disagreement, the qualitative dynamics of the curve are captured. Importantly, the deceased and recovered curves fit the data quite well. As discussed in Section 2, in the early dynamics of the epidemic at the national or provincial level the epidemic does not evolve as a single outbreak, but rather as the sum of multiple outbreaks in multiple cities with temporal delay. As a result, a model like Eq. (A.1) ought not to perfectly fit these early dynamics. In fact, if we did not set some of the parameters with epidemiological insights (as described in the Supplementary Information Section B.2) we could achieve stricter quantitative fits of the early dynamics of the epidemic at the cost of long term prediction. Parameter values for the Canadian provinces of Quebec and Ontario are provided in Table 1 .