Unreported cases in the 2014-2016 Ebola epidemic: Spatiotemporal variation, and implications for estimating transmission

Abstract

In the recent 2014-2016 Ebola epidemic in West Africa, non-hospitalized cases were an important component of the chain of transmission. However, non-hospitalized cases are at increased risk of going unreported because of barriers to access to healthcare. Furthermore, underreporting rates may fluctuate over space and time, biasing estimates of disease transmission rates, which are important for understanding spread and planning control measures. We performed a retrospective analysis on community deaths during the recent Ebola epidemic in Sierra Leone to estimate the number of unreported non-hospitalized cases, and to quantify how Ebola reporting rates varied across locations and over time. We then tested if variation in reporting rates affected the estimates of disease transmission rates that were used in surveillance and response. We found significant variation in reporting rates among districts, and district-specific rates of increase in reporting over time. Correcting time series of numbers of cases for variable reporting rates led, in some instances, to different estimates of the time-varying reproduction number of the epidemic, particularly outside the capital. Future analyses that compare Ebola transmission rates over time and across locations may be improved by considering the impacts of differential reporting rates.

Fig 1. Hierarchical binomial tree for classifying reported and unreported non-hospitalized Ebola deaths.

Variables highlighted in red are taken from data. Blue squares indicate the components of the total unreported community Ebola deaths per week.

Fig 2. Correlations between reported prevalence in the WHO patient database and reported prevalence in Red Cross burials of non-hospitalized deaths, in four districts of Sierra Leone: Western Area Urban (the capital district; red circles), Western Area Rural (orange circles), Bombali (blue triangles), and Bo (green squares).

Polygons enclose +/- 2 standard errors for Poisson regression of non-hospitalized burial prevalence as a function of WHO reported prevalence (confirmed and probable cases) in each district.

Fig 3. Contrasting estimates of Ebola incidence and reporting rates over space and time.

Upper row: The heights of the black polygons show weekly numbers of confirmed plus probable cases from the WHO data. Colored polygons enclose credible intervals for total number of cases per week (both reported and unreported). Lower row: Estimated reporting rates over time and across districts; polygons enclose the credible intervals. Horizontal lines show static reporting rates estimated by [17] (lower, solid), [19] (middle, dashed) and [18] (top, dotted). The central white region of each plot shows the temporal coverage of the SDB data. To illustrate the temporal coverage of our data relative to the epidemic as a whole, total cases and reporting rates in the grey regions are extrapolated using the mean of the nearest month of burial data. While extrapolated values match published estimates of static reporting rates, these extrapolations are not used in any of the analyses.

Fig 4. The estimated effective reproduction number of the Ebola epidemic over time in four districts of Sierra Leone, based the WHO data, either corrected for variable underreporting (colors) or uncorrected (grey).

Polygons enclose the interquartile range of the credible interval on the estimate over time, encompassing the central 50% of the posterior distribution at each time point[28].