Assessing the Accuracy of Early COVID-19 Case and Fatality Model Projections in Africa

Abstract

Abstract ObjectiveWe compared reported COVID-19 case, fatality, and peak date data for Africa Union (AU) member states with estimates and projections produced by various mathematical models to assess their accuracy in the context of an ongoing pandemic and identify key gaps to improve the utility of models in the future.MethodsWe conducted a systematic literature review to identify studies published in any language between January and December 2020 that reported results of COVID-19 modeling analyses for any AU member state. Reported COVID-19 case, fatality, peak date, and testing rate data were obtained. Descriptive, bivariate, and meta-analyses were conducted to compare reported data to model-generated estimates. FindingsFor included countries in the respective model simulation periods, model-predicted cumulative cases ranged from 2 to 76,213,155 while model-predicted cumulative deaths ranged from 8 to 700,000. The difference between reported and predicted cumulative COVID-19 cases was between -99.3 % to 1.44×106 % with most values being above 24.7%, and the difference between reported and predicted cumulative COVID-19 deaths for models reviewed was between -2.0 % to 2.73×105 % with most values being above 50.0%. The difference in the predicted and reported dates for the first epidemic wave peak was between -242 Days to 249 Days.ConclusionFor the first COVID-19 epidemic wave, epidemiological model results were observed to have high precision but low accuracy when compared to reported peak case date and cumulative cases and deaths indicating that these data were either under-reported or model-overestimated.