Strategies for containing Ebola in West Africa

Abstract

The ongoing Ebola outbreak poses an alarming risk to the countries of West Africa and beyond. To assess the effectiveness of containment strategies, we developed a stochastic model of Ebola transmission between and within the general community, hospitals, and funerals, calibrated to incidence data from Liberia. We find that a combined approach of case isolation, contact-tracing with quarantine, and sanitary funeral practices must be implemented with utmost urgency in order to reverse the growth of the outbreak. As of 19 September, under status quo, our model predicts that the epidemic will continue to spread, generating a predicted 224 (134 to 358) daily cases by 1 December, 280 (184 to 441) by 15 December, and 348 (249 to 545) by 30 December.

Fig. 1. Model fit to data

(A) Cumulative non–health care worker cases. (B) Cumulative deaths. (C) Cumulative health care worker cases. (D) Cumulative hospital admissions. Cumulative Ebola cases and fatalities were obtained from the Liberian Ministry of Health and Social Welfare Situation Reports nos. 10 to 89 (red circles), to which the model was fit. A 95% prediction interval was generated by 10,000 runs of the model, with parameters randomly sampled from within their confidence intervals (gray fill) (supplementary materials). Validation of the model predictions was provided by comparison with data of the cumulative Ebola cases and fatalities from Situation Reports nos. 89 to 127 (blue circles), representing 12 August 2014 to 19 September 2014, which were not used for model fitting.

Fig. 2. Nonpharmaceutical intervention effectiveness

(A to F) Model predictions of the cumulative number of new cases after 1, 3, and 6 months after 20 September 2014, as well as the probability of fewer than one case daily after these months for (A) 80% reduction in transmission to health care workers combined with different reductions in community transmission, (B) increasing proportions of sanitary burial of hospital deaths, (C) increasing proportions of hospital case isolation, (D) 90% reduction in transmission to health care workers and increasing sanitary burial of hospital deaths, (E) 80% sanitary burial of hospitalized deaths with increasing sanitary burial of community deaths, and (F) 80% case isolation of hospitalized patients with concurrent contact-tracing and quarantine of infected contacts. One thousand simulations of the stochastic model were used to generate the cumulative case count error bars (95% prediction interval) and to estimate the probability of less than one new case per day.

Fig. 3. Effectiveness comparison of individual and combined intervention strategies

Model predictions of the daily number of new and cumulative cases after 6 months for sanitary burial and hospital deaths, sanitary burial of community deaths, case isolation of hospitalized patients, contact-tracing in the community, and quarantine of infected contacts.

Fig. 4. Sensitivities and elasticities with respect to epidemiological parameters

(A to C) Sensitivities and elasticities of cumulative cases under (A) 85% successful sanitary burial of hospital deaths with 95% reduction in transmissibility to health care workers for 6 months (Fig. 2D, blue line); (B) successful sanitary funeral of 90% hospital deaths and 50% community deaths, respectively; and (C) 90% successful hospital case isolation with concurrent contact-tracing and quarantine of 80% infected contacts. We varied the number of pre-outbreak health care workers [S_W(0)], the incubation period (1/α), the duration from symptom onset to death if not hospitalized (1/γ_DG), the duration from symptom onset to recovery if not hospitalized (1/γ_RG), the duration between symptom onset and hospitalization (1/γ_H), the hospitalization rate per person for reasons other than Ebola (h), the duration of hospitalization for reasons other than Ebola (1/f_HG), the number of funeral attendees with close contact to the body (M_F), the number of hospital visitors per patient (M_H), the transmission rate at funerals relative to general community (ω), the fraction of asymptomatic infections (1 – ε), and the hospital admission rate (θ). For each parameter varied, we recalibrated and reran the model.