BACKGROUND: Between 2014 and 2015, the largest outbreak of Ebola Virus Disease (EVD) occurred in West Africa, forcing the world to reconsider not only its perceived susceptibility to a virus, but also its grasp on a relatively unknown pathogen. Forty years after its discovery, there is reason to question the reliability of the universally accepted incubation period of 21 days. Based on a previous sample size of roughly 300 cases, the estimate of 21 days does not take into account potential measurement error or information bias. This study aimed to examine these influences on the estimation of the incubation period of EVD.
METHODS: Data for suspected, probable, and confirmed cases of EVD in Guinea, Liberia, Nigeria, and Sierra Leone between August 29th and September 14th, 2014 were digitized from weekly World Health Organization Ebola Situation Reports. Cases were defined as single versus multiple exposure, as it is hypothesized that exposure type is correlated with varying risk of and time to developing disease. Conservative random bias and error proxy terms between -7 and 7 days were assigned to the original reported date of exposure of each case. A non-parametric bootstrap was used to generate a representative sample. Outcome misclassification (incubation ≤ 21 days versus > 21 days) sensitivity analysis, multivariate logistic, and Poisson regression models were conducted using SAS v9.3.
RESULTS: Of the 4,507 EVD cases as of September 2014, 723 were digitized and bootstrapped to replicate the true case counts by May 2015. Total number of cases used in the analysis was 20,957. Cases with incubation period greater than 21 days were 60% less likely to have had a single exposure as compared to cases that fell within the 21 day period. In the unadjusted model, 6.4% of cases had incubation period exceeding 21 days; this number decreased to 5.5% after adjustment in the outcome misclassification sensitivity analysis. After accounting for misclassification bias and measurement error, incubation period exceeded 21 days in 9.5% of cases, with upper limits as great as 41 days.
CONCLUSIONS: This study supported the hypothesis that modeling information bias and measurement error in the estimation of incubation time yielded a substantially larger proportion of cases exceeding the previously cited period of 21 days. The accurate estimation of the incubation period of EVD—and countless pathogens alike—is crucial to timely and efficient outbreak response and prevention of future transmission.