COVID-19 pandemic in the African continent: forecasts of cumulative cases, new infections, and mortality

Yohannes Kinfu; Uzma Alam; Tom Achoki

Thirdly, literature on the association between disease prevalence and population level characteristics that are known to be the staple of social and descriptive epidemiology, and the backbone of predictive modeling exercises are yet to emerge for COVID-19. (4-6, 8) Further, at the individual level, knowledge on the determinants of COVID-19 morbidity and mortality is limited to selected characteristics such as age and pre-existing conditions, and even then, they are drawn from scanty data. The association with community and population level characteristics is even more limited, hence largely speculative. In this context, the task of developing a multi-country covariate-based predictive model for COVID-19 in any country or region of the world, let alone Africa, is likely to be guided by what is feasible, and to the extent, the available data permits it. It also means that there is a need for constant revision of predictive models as new data becomes available. In our case, the work is further complicated because in Africa data on key covariates are either lacking or when they exist, they tend to be biased or derived from other global covariate-based modeling exercises. (21) (22) Despite this, we have made efforts to mitigate these limitations by curating data from various credible sources around the world and assessing them for consistency. In developing our model, we have used data sources from 193 countries globally and used statistical relationships to fill data gaps on covariates in the Africa region. Additionally, our model performance was assessed through rigorous out of sample calibration using k-fold cross-validation techniques and obtaining robust results. We have also restricted our forecast to the first few months (April-June) and refrained from a long-term projection exercise. Firstly, this is because we anticipate new and additional data to emerge in the short term that will lead to better and improved estimates. Secondly, where data is sparse, any longterm projection is more likely to be detached from reality and can easily become a wild guess and is therefore less useful for informing policy actions. Thirdly, if the countries in the region will not be . CC-BY-NC-ND 4.0 International license It is made available under a author/funder, who has granted medRxiv a license to display the preprint in perpetuity.