Qualitative and quantitative evaluation of COVID-19 outbreak severity with the use of meta-projections based on Richards' curve parameters

Researchers have shown that even simple empirical models stemming from biological growth modeling have the potential to provide useful information on the development and severity of ongoing epidemics since they can be employed as tools for carrying out projections on the size of the affected population, timing of turning points, as well as best- and worst-case scenarios. Nevertheless, they commonly exhibit considerable sensitivity to some input parameters' variance which results in large fluctuations in the generated projections, thus rendering predictions difficult and even risky. In this work we examine a novel meta-projections-based approach which allows us to evaluate the model's current trends and assess whether generated projections are at a transient or stable state. Meta-projections can be extracted from graphs of successive estimations of model's parameters and resulting projections, over a sequence of days being gradually added to the employed model. In other words, projections are carried out on truncated time series of cumulative numbers of confirmed cases with increased lengths at each successive evaluation. This allows us to trace the values of model parameters over a certain period of time and examine their trends which may converge to specific values for settled-growth cases or exhibit a changing or even an erratic behavior for cases that undergo epidemiological transitions and/or are inappropriately described by the current model instance(s). We have computed meta-projections and compared our findings for countries at different stages of the epidemic with stable or unstable behaviors and increasing or decreasing numbers of confirmed cases. Our results indicate that meta-projections can aid researchers in assessing the appropriateness of their relevant models and in effect decrease the uncertainty in their estimations of an epidemic's severity and development.