A critical issue during the decision-making process for disease management is to identify the optimal intervention to efficiently bring an outbreak under control. To determine the best candidate intervention requires an understanding of uncertainties from both epidemiological and operational perspectives. While models are increasingly developed to address epidemiological uncertainties, information about the feasibility and cost to achieve a particular impact for a given intervention must also be considered. Crucially, these two types of uncertainty are rarely addressed concurrently in epidemic studies, impeding estimation of the role of different sources of uncertainty. Here, we present an approach that allows us to address both sources of uncertainty on the same platform. We use Ebola outbreak management as a study case. We identified and recoded a large ensemble of Ebola models to evaluate the impact of three classes of widely applied interventions (reducing community transmission, improving hospitalization and reducing funeral transmission) on three different management objectives (minimizing caseload, deaths and the probability of a major outbreak). To illustrate the potential impact that different cost-scaling relationships may have on the optimal intervention, we explored three representative possible cost functions for public health interventions.
Results/Conclusions
The models projected substantially different case numbers, indicating high levels of epidemiological uncertainty. Our study shows that the impact necessary to achieve a consensus threshold on optimal management differs between interventions. On average, for any management objective, the required impact was lowest for reductions in community transmission, intermediate for reductions in funeral transmission and highest for improving hospitalization. Comparison of model-intervention combinations shows that the amount of variation between models differs under different interventions, suggesting that epidemiological uncertainties need to be considered separately under particular intervention settings. Interestingly, results were very similar for all three management objectives. Most importantly, our study demonstrates that the rank of interventions is jointly determined by the epidemiological impact and the potential expected cost of the corresponding management actions. Thus, an epidemiologically impactful intervention might not be optimal if it is extremely expensive or logistically prohibitive. Overall, our study shows that it is critical to concurrently address epidemiological and operational uncertainties within the same modeling structure to efficiently inform the epidemic decision-making process. The approach developed in this study can easily be applied to decision-making for management of other diseases for which multiple models and multiple interventions are available.