Real-time projections of epidemic transmission and estimation of vaccination impact during an Ebola virus disease outbreak in Northeastern Democratic Republic of Congo

Lee Worden; Rae Wannier; Nicole A. Hoff; Kamy Musene; Bernice Selo; Mathias Mossoko; Emile Okitolonda-Wemakoy; Jean Jacques Muyembe-Tamfum; George W. Rutherford; Thomas M. Lietman; Anne W. Rimoin; Travis C. Porco; J. Daniel Kelly

The stochastic and negative binomial auto-regression models were scored based on The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. . https://doi.org/10.1101/461285 doi: bioRxiv preprint and auto-regression models to project one-week, two-week, four-week, and eight-week 274 forecasts of outbreak size. As time lapsed, we compared predicted and known outbreak 275 sizes and found a higher probability of accurate forecasts at one week than at eight 276 weeks (Figs 2, 3) . Log-likelihood scores typically declined as projection times extended 277 further into the future. These lower-scoring longer-term projections tended to include 278 wider prediction intervals, reflecting less certain outcomes in which less probability was 279 assigned to any one value. The epidemic curve accelerated in early October, and 280 stochastic model forecasts occurring just before that change were especially low scoring, 281 as they failed to anticipate the coming rise in case counts. Subsequent projections took 282 into account the reported acceleration and their performance recovered. After our model validation process was completed, we used the stochastic and 284 auto-regression models to project one-week, two-week, four-week, and eight-week 285 outcomes (Figs 4, 5) . We used the Gott's rule and Theil-Sen regression models together 286 with the stochastic model to project final outbreak sizes (Figs 6, 7) . The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. . https://doi.org/10.1101/461285 doi: bioRxiv preprint The copyright holder for this preprint (which was not peer-reviewed) is the author/funder. . https://doi.org/10.1101/461285 doi: bioRxiv preprint quite likely by all the models, with the Gott's rule forecast being particularly pessimistic, 298 as it was in all cases. The probability of a very large outbreak (10,000 or more cases) 299 was calculated to be below 8% for Gott's rule and negligibly small for the other models. 300 A final outbreak size of a catastrophic outbreak larger than the West Africa outbreak 301 (28,616 or more cases) was projected to have probability less than 3% by Gott's rule 302 and negligibly small for the other models (Table S2 in S1 Supporting Information). 303 We generated short-and long-term projections from earlier snapshots of the current 304 outbreak case counts with each model, for the purpose of both scoring and forecasting. 305