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

During an Ebola outbreak, real-time forecasting has the potential to support 28 decision-making and allocation of resources, but highly accurate forecasts have proven 29 difficult for Ebola [8, 9] as well as other diseases [10] [11] [12] [13] . Highly accurate forecasts of 30 small, noisy outbreaks may be a fundamentally elusive ideal [14] . Previous work has 31 found that probabilistic forecasts can have relatively high accuracy within a few weeks, 32 though they become less useful as time horizons grow longer [15] . Thus, short-term against subsequently known counts. Although the epidemic was officially reported on 63 August 1 as a cluster of cases occurring in June and July, seven sporadic early cases 64 from April and May were subsequently linked to the current outbreak and were added 65 to later case totals [3] . This additional knowledge was added retrospectively to the time 66 series of cumulative case counts only for predictions made for days on or after 67 September 15, when these cases were officially linked to the current outbreak. 68 Stochastic model 69 We modeled Ebola virus (EBOV) transmission using a stochastic branching process 70 model, parameterized by transmission rates estimated from the dynamics of prior EVD 71 outbreaks and conditioned on agreement with reported case counts from the 2018 EVD 72 outbreak to date. We incorporated high and low estimates of vaccination coverage into 73 this model. We used this model to generate a set of probabilistic projections of the size 74 of simulated outbreaks in the current setting. This model is similar to one described in 75 previous work [16] , with the addition of a smoothing step allowing for a continuum of 76 transmission rates interpolated between those estimated from prior outbreaks. 77 On the assumption that past outbreaks provide a basis for projection of the current 78 outbreak, we used estimates of transmission rates from past EVD outbreaks to 79 parameterize simulations of the current outbreak. To estimate the reproduction number 80 R in past outbreaks as a function of the number of days from the beginning of the 81 outbreak, we included reported cases by date from fourteen prior outbreaks (Table S1 in 82 S1 Supporting Information), [17] [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] [28] [29] [30] . To reflect the Ebola response system in DRC 83 during what is now its tenth outbreak, the first historical outbreak reported in each 84 country was excluded (e.g. the 1976 outbreak in Yambuko, DRC), as there is a 85 difference in the Ebola response system as well as community sensitization to EVD 86 following a country's first outbreak. We used the Wallinga-Teunis technique to estimate 87 R for each case and therefore for each reporting date in these outbreaks [31] . The serial 88 interval distribution used for this estimation was a gamma distribution with a mean of 89 14.5 days and a standard deviation of 5 days, with intervals rounded to the nearest 90 whole number of days, consistent with the understanding that the serial interval of EVD 91 cases ranges from 3 to 36 days with mean 14 to 15 days. Transmission rates estimated by day in these outbreaks tend to decline from initially 93 high to eventually low values, though they may display substantial fluctuations. This 94 "quenching" of transmission may be driven by formal interventions such as quarantine, or 95 by informal changes in individuals' behavior in response to the disease or by depletion 96 of uninfected