The modeling approach presented here adequately captures the complex spatiotemporal pattern of the Guinea EVD epidemic through microsimulations. This computational modeling approach was previously applied to the EVD epidemic in Liberia, allowing us to compare the results obtained for the two countries to determine how differences in healthcare infrastructure and the implementation of interventions can explain the distinct epidemic patterns. For Guinea, the model yields an estimated reproduction number (R?~?1.2) that is much lower than those reported for Liberia (estimates in the range 1.7–2.0) [5, 14, 22, 23] and Sierra Leone (estimates in the range of 1.4–2.0) [14, 23, 24]. The model suggests that this particular feature of the epidemic in Guinea stems from the early availability of ETUs (which were already open in the phase where we computed the reproduction number) and their more developed healthcare system that helped to prevent sustained exponential growth of the epidemic. This seems to be consistent with evidence from the field that a lower proportion of cases in Guinea were associated with transmission in healthcare settings compared with Liberia; in particular, the analysis of the initial transmission chain in Conakry attributed 9 % of EVD infections to hospitals over the period April to August, 2014 . Consistently, we estimate that 8.8 % (95 % CI, 3.6–15.8) of transmissions were linked to hospital/ETU contacts through August 15, 2014. A similar proportion of transmissions in hospital settings (namely 7.7 %) was reported in an analysis of an infection tree in Pujehun district of Sierra Leone . On the other hand, these estimates differ substantially from analogous estimates in Liberia where, according to modeling results , a much larger proportion of transmission (30.8 %) was attributed to hospitals, especially in the initial phase of the epidemic. This difference is likely due to the lack of sufficient beds in ETUs in Liberia in the initial phase of the epidemic. In Guinea, even at the time of the peak incidence (November–December 2014), the daily number of admissions to ETUs ranged from about 10 to 30, lower than the estimated bed capacity of the four ETUs already open at that time. The estimated proportion of transmission linked to unsafe burials is higher than that previously observed  over the period February 10 to August 25, 2014. However, that analysis preceded our current one and was not able to take the subsequent rise in the number of unsafe burials into account (Fig. 4d). Our analysis shows the key role of CT for limiting Ebola spread. An analysis of the Ebola epidemic in Liberia  showed that the distribution of household protection kits had the effect of significantly accelerating the elimination phase. In sum, the results of both studies highlight the importance of reducing transmission between household and extended family members for controlling Ebola spread.
The model also indicates that the long tail of the disease elimination phase in Guinea could be explained by fluctuations in the level of CT that, although adequate to prevent resurgence of a large epidemic , were not enough for timely elimination of the disease (Fig. 6). Specifically, we found numerical evidence that 7–10 contacts traced per Ebola case were consistently needed to achieve disease elimination with relatively high probability (90 %) before the end of 2015. To effectively perform CT, human and logistical resources are required. Human behavior, such as fear and hostility, can also drive acceptance of CT, as well as the fraction of population seeking care in ETUs and safe burials for the deceased . Thus, the estimated impact of the interventions that we report incorporates population compliance to public health directions; different human behavioral responses of the population would lead to increased or decreased effectiveness of control measures. Indeed, this represents a possible explanation for the observed larger effect of CT in 2015 than in 2014 (Fig. 4c). Contact tracers could have become more skilled over the course of the epidemic after acquiring experience on the ground. Alternatively, the population could have changed its behavior, allowing health personnel to perform CT more efficiently. However, our modeling analysis does not have the capability to disentangle these effects.
Another important element to consider in discussing disease elimination scenarios is the WHO-sponsored ring vaccination trial of the rVSV vaccine in Guinea. The interim data analysis shows an impressive 100 % vaccine efficacy . Numerical simulations show that it is reasonable to assume that the vaccination trial itself may have had a considerable impact on the disease control effort. These simulations show that the vaccination program provides the largest benefit in regions with low, deteriorating or fluctuating CT, thereby considerably accelerating disease elimination. Consistent with other modeling studies [27–29], we found that the major benefit of ring vaccination is observed at low levels of CT (5–10 individuals per case). Finally, vaccination gives direct protection to vaccinees, which is important for controlling epidemics on a local level as well as protecting those providing front line care.
Most of the assumptions and limitations of the model have already been discussed by Merler et al. . It is worth mentioning however that, similar to the model for Liberia, we assume that Ebola cases in Guinea are unlikely to travel long distances when they have active symptoms, unless they are seeking hospital care or help from relatives and friends. This assumption has been validated for Liberia and appears to adequately reproduce the geographical incidence in Guinea. However, we cannot exclude the possibility that local population mobility could drive EVD dynamics, especially during the elimination stage where the epidemic can be dominated by rare fluctuations. Another feature that we have not considered in the model is the effect of asymptomatic infection and acquired immunity . While the early stage of the epidemic is probably not overly affected by asymptomatic infections, acquired immunity may play a role in speeding up the elimination of the disease. Notwithstanding the mentioned limitations, the model is able to adequately describe the dynamics of the epidemic in Guinea, provide estimates of transmission by setting in agreement with reported data , and capture the highly heterogeneous pattern of transmission reported in previous studies [6, 7]. These findings support our choice of using an individual-based modeling approach, capturing the observed high heterogeneity in the number of secondary infections and the clustering effect due to the low number of effective contacts in the different settings.