Modernization of regression models to predict the number of deaths from the new coronavirus infection

Abstract

The aim of the study was to modernize the existing prognostic regression models in the context of expanding knowledge about the new coronavirus infection. Materials and Methods. The modification of models and the increase in their predictive ability are based on collecting the available data from international and Russian databases. We calculated the traditional descriptive statistics and used the linear regression analysis for modeling. The work was performed using the IBM SPSS Statistics 26.0 and the R 3.6.0 (RStudio) software. Results. Manifestations of the COVID-19 epidemic process in several countries were studied; special attention was put to the number of deaths associated with the infection. A significant proportion of severe cases were noted among patients both in Russia and elsewhere. Considering that the disease incidence has reached its peak in China and Italy, we were able to improve the previously published (Sovremennye tehnologii v medicine 2020, Vol. 12, No.2) regression models and to compare their performance. The first modified model is based on the absolute increase in new cases of the infection: its regression coefficient is 0.16 (95% CI 0.137–0.181). In the extended version of the updated model, we additionally considered cases of aggravated COVID-19: the regression coefficients were 0.128 (95% CI 0.103–0.153) for model 2 and 0.053 (95% CI 0.029–0.077) for model 1.1; p=0.0001. Conclusion. Based on the most recent data (from January to May 2020) on the incidence of COVID-19 in the world, we have developed more specific versions of the basic and extended regression models of lethal outcomes. The resulting models are optimized and extrapolated to the current epidemiological situation; they will allow us to improve our analytical approach. For that purpose, data collection is currently ongoing.