Using the posterior predictive distribution to analyse epidemic models: COVID-19 in Mexico City

Epidemiological models usually contain a set of parameters that must be adjusted based on available observations. Once a model has been calibrated, it can be used as a forecasting tool to make predictions and to evaluate contingency plans. It is customary to employ only point estimators of model parameters for such predictions. However, some models may fit the same data reasonably well for a broad range of parameter values, and this flexibility means that predictions stemming from them will vary widely, depending on the particular values employed within the range that gives a good fit. When data are poor or incomplete, model uncertainty widens further. A way to circumvent this problem is to use Bayesian statistics to incorporate observations and use the full range of parameter estimates contained in the posterior distribution to adjust for uncertainties in model predictions. Specifically, given an epidemiological model and a probability distribution for observations, we use the posterior distribution of model parameters to generate all possible epidemic curves, whose information is encapsulated in posterior predictive distributions. From these, one can extract the worst-case scenario and study the impact of implementing contingency plans according to this assessment. We apply this approach to the evolution of COVID-19 in Mexico City and assess whether contingency plans are being successful and whether the epidemiological curve has flattened.