Controlling Ebola: key role of Ebola treatment centres.

What initially was perceived to be a self-limited outbreak of Ebola virus disease in a forested area of Guinea has become an unprecedented epidemic of international concern that continues to spread unabated in parts of west Africa.1, 2 A lack of public health infrastructure together with delays in virus detection and implementation of control interventions have contributed to the widespread transmission of Ebola virus disease in a region inexperienced with this disease.

As in previous international health crises, such as the severe acute respiratory syndrome epidemic and the 2009 influenza pandemic, mathematical models are proving instrumental to guide the public health response against Ebola virus disease and monitor the effectiveness of control interventions.4, 5 Whereas earlier modelling efforts have relied on compartmental homogeneous-mixing models,6, 7, 8 the study by Stefano Merler and colleagues, reported in The Lancet Infectious Diseases, uses a microsimulation model to capture spatial heterogeneity in the transmission dynamics of Ebola virus disease in Liberia. Indeed, local epidemics of Ebola virus disease seem to be asynchronous and show slower than expected growth, a pattern probably driven by the geometry of the contact network or social behaviour changes.

The model of Merler and colleagues incorporates fine details of Liberia's population structure and geography, including location of households, hospitals, and Ebola treatment units. Infectiousness is assumed to intensify in the later and more severe stages of Ebola virus disease, when infectious individuals are confined at home or in the health-care setting, and exposed to a restricted number of caregivers. The resulting contact network is highly clustered, giving rise to a slow and local mode of dissemination, a pattern that homogeneous-mixing models have been unable to capture. Long-distance transmission events (eg, during unsafe funerals) are predicted to be uncommon; instead, the slow geographic spread of Ebola virus disease is best explained by distance travelled to reach a hospital. This point is interesting, because several long-distance transmission events were key in the dissemination of the infection to neighbouring countries (Mali, Nigeria, Senegal) and other continents (Europe [Spain], North America [USA]).

One important aspect of the model by Merler and colleagues is to assess the effectiveness of intervention measures put in place in Liberia since mid-August, 2014. In the pre-intervention period from June to mid-August, 2014, most of the Ebola infections were estimated to occur in hospitals (38%), followed by households (31%), the community (22%), and at funerals (9%). The rapid establishment of new Ebola treatment units was a key step to curb the epidemic in Liberia, although the model assumes near perfect isolation of infectious individuals in Ebola treatment units, which could be overly optimistic. Distribution of household protection kits and implementation of safe burial procedures were also associated with significant reduction in Ebola virus disease transmission. Of note, other interventions not explicitly modelled could have played a part, including use of rapid diagnostic kits in Ebola treatment units, which reduces the delay from presentation to isolation, and changes in population behaviour in response to mass education campaigns and accumulation of Ebola virus disease cases. Some of these effects could have been absorbed in the model by Merler and colleagues through the estimated effect of household protection kits.

The model by Merler and colleagues provides a substantial improvement compared with earlier homogeneous mixing models6, 7, 8 and in turn their more optimistic predictions align better with the observed trajectory of the epidemic in Liberia. However, the model tends to underestimate Ebola treatment unit admissions and predicts an earlier peak than reported, suggesting that further improvements could be useful. Ideally, future models should integrate more realistic population mobility patterns derived from cellphone usage data (eg, flowminder), proxies of social behaviour, and differences in reporting and hospitalisation rates between rural and urban areas. However, these important data are lacking for the region.

The outlook for Liberia has substantially improved over the past few weeks with news of an epidemic slowdown (figure ). By contrast, incidence has remained relatively stable in Guinea, whereas the epidemic continues to ascend quickly in Sierra Leone (figure). Difficulties in building and staffing Ebola treatment units could explain the worrisome turn of the outbreak in Sierra Leone. A key test of the robustness of the model by Merler and colleagues will be whether it can reproduce the highly distinct dynamics of Ebola virus disease in different countries. In the longer run, microsimulations and other modelling approaches could prove instrumental to optimise interventions to curb and ultimately eliminate Ebola virus disease in the region, especially as an Ebola vaccine might materialise in the near future.

Figure

Cumulative number of Ebola virus disease cases (log scale) in Guinea, Sierra Leone, and Liberia

As of Dec 10, 2014, 27 Ebola treatment centres were open in Liberia (1079 beds), 12 in Sierra Leone (565 beds), and four in Guinea (200 beds).