In 2014–2015, a large Ebola outbreak afflicted Liberia, Guinea, and Sierra Leone. We performed a systematic review of 26 manuscripts, published between 2014 and April 2015, that forecasted the West African Ebola outbreak while it was occurring, and we derived implications for how results could be interpreted by policymakers. Forecasted case counts varied widely. An important determinant of forecast accuracy for case counts was how far into the future predictions were made. Generally, forecasts for less than 2 months into the future tended to be more accurate than those made for more than 10 weeks into the future. The exceptions were parsimonious statistical models in which the decay of the rate of spread of the pathogen among susceptible individuals was dealt with explicitly. The most important lessons for policymakers regarding future outbreaks, when using similar modeling results, are: 1) uncertainty of forecasts will be greater in the beginning of the outbreak; 2) when data are limited, forecasts produced by models designed to inform specific decisions should be used complementarily for robust decision-making (e.g., 2 statistical models produced the most reliable case-counts forecasts for the studied Ebola outbreak but did not enable understanding of interventions’ impact, whereas several compartmental models could estimate interventions’ impact but required unavailable data); and 3) timely collection of essential data is necessary for optimal model use.

中文翻译：

---

预测2014年西非埃博拉疫情。

2014-2015年，埃博拉大爆发，利比里亚，几内亚和塞拉利昂受灾。我们对2014年至2015年4月之间发表的26篇手稿进行了系统的回顾，该手稿预测了西非埃博拉病毒爆发的发生时间，并对政策制定者如何解释结果产生了启示。预测的病例数差异很大。病例数预测准确性的重要决定因素是对未来预测的预测程度。通常，对未来不到2个月的预测往往比对未来10周以上的预测更为准确。例外情况是简约统计模型，其中明确处理了易感者之间病原体传播速率的下降。对于政策制定者而言，关于未来爆发的最重要的教训，当使用类似的建模结果时，是：1）爆发开始时预测的不确定性会更大；2）在数据有限的情况下，应将旨在为特定决策提供信息的模型所产生的预测补充用于稳健的决策（例如，两个统计模型对所研究的埃博拉疫情产生了最可靠的病例数预测，但无法理解干预措施的影响，而几个分区模型可以估计干预措施的影响，但需要不可用的数据）；3）及时收集必要的数据对于优化模型的使用是必要的。旨在为特定决策提供依据的模型所产生的预测应互补地用于稳健的决策（例如，两个统计模型对所研究的埃博拉疫情产生了最可靠的病例数预测，但无法理解干预措施的影响，而多个阶段模型可以估计干预措施的影响，但需要不可用的数据）；3）及时收集必要的数据对于优化模型的使用是必要的。旨在为特定决策提供依据的模型所产生的预测应互补地用于稳健的决策（例如，两个统计模型对所研究的埃博拉疫情产生了最可靠的病例数预测，但无法理解干预措施的影响，而多个阶段模型可以估计干预措施的影响，但需要不可用的数据）；3）及时收集必要的数据对于优化模型的使用是必要的。