Modeling Covid-19 dynamics for real-time estimates and projections: an application to Albanian data

Erida Gjini

The advantage of this framework is that new data arriving in real-time can update model estimates and subsequent projections. In the case of Albanian data, adding the case observations for the next 3 days (March 29-31), narrows dramatically the credible intervals of the future course of the epidemic (see Figure 5 ) with little change in the posterior parameter estimates ( Figure S4 ). The model-predictions for the quantitative aspects of the epidemic are also narrowed substantially ( Figure S5 ). In particular, the peak of the outbreak is expected to occur between 5-15 April (Fig.S5C) , and not surpass 300 active confirmed cases (Fig.S5B) , if control continues with this trend. Comparing the model fits to data March 9-31 for the dynamic control and constant control model, we observe a smaller error for the dynamic control than the constant control model ( Figure 6 ). This is further confirmed from analysis of the deviance information criterion DIC (14) , where we find that DIC dynamic = 4447 for the dynamic model, and DIC constant = 27283 for the constant model. Typically models with smaller DIC are favoured. Thus, the data strongly suggest that the effect of control measures on transmission in Albania has been dynamic, increasing over time.