Hierarchical Bayes and logistic growth: Predicting diagnosed COVID19 cases and the corresponding burden on health care systems

The generalized logistic growth model (GLGM) is a flexible growth curve model which has been successfully applied to stages of the COVID-19 outbreak. Here we demonstrate the effectiveness of such growth curve models in forecasting the number of diagnosed cases and hospital use due to COVID-19. In Iceland, this sort of top-down approach of using phenomenological growth models to predict future diagnosed cases and empirical data such as the age distribution of diagnosed cases to estimate hospital and ICU admissions was utilized to great success in predicting the date and magnitude of maximum diagnosed cases.

However, in spite of its flexibility, the GLGM gives poor fits to the highly right-skewed growth curves seen in many countries. To make this methodology applicable worldwide we therefore propose an inverted version of the GLGM that can fit arbitrarily right-skewed growth curves, and show results from a Bayesian hierarchical implementation to data from the European Center for Disease Prevention and Control on daily diagnosed COVID-19 cases in 68 countries worldwide.

The model is fit in a hierarchical fashion using Stan such that information on parameters within each country can be shared between countries. We also introduce our methodology for estimating the use of general hospital and ICU beds using predicted daily confirmed COVID-19 cases along with the age distributions of diagnosed cases from domestic data, and hospitalisation and ICU admittance rates from published research.

Code and workflow is available at: https://github.com/bgautijonsson/isa_trans_covid19

Presenter biography:
Brynjólfur Gauti Jónsson

Brynjólfur Gauti Jónsson is a PhD student of biostatistics at the Center of Public Health at the University of Iceland. He works as a researcher at the Icelandic Heart Association and as a teacher and statistical consultant at the University of Iceland.