[论文解读] Real-time projections of epidemic transmission and estimation of vaccination impact during an Ebola virus disease outbreak in the Eastern region of the Democratic Republic of Congo

Lee Worden, Rae Wannier+11 more|arXiv (Cornell University)|Nov 3, 2018

Viral Infections and Outbreaks Research参考文献 12被引用 3

一句话总结

本研究开发并验证了一套多方法实时预测框架，用于2018年刚果民主共和国东部的埃博拉疫情，整合了随机分支过程、自回归模型、Theil-Sen回归和Gott定律，以预测病例数和疫苗影响。研究发现，在62%疫苗接种覆盖率下，传播率高于预期，短期预测优于长期预测，且疫情尚未达到峰值，由于冲突导致的干扰，疫苗影响有限。

ABSTRACT

As of October 12, 2018, 211 cases of Ebola virus disease (EVD) were reported in North Kivu Province, Democratic Republic of Congo. Since the beginning of October the outbreak has largely shifted into regions in which active armed conflict is occurring, and in which EVD cases and their contacts are difficult for health workers to reach. We modeled EVD transmission using a branching process with gradually quenching transmission estimated from past EVD outbreaks, with outbreak trajectories conditioned on agreement with the course of the current outbreak, and with multiple levels of vaccination coverage. We used an autoregression for short-term projections, a regression model for final sizes, and a simple Gott's law rule as an ensemble of forecasts. Short-term model projections were validated against actual case counts. During validation of short-term projections, models consistently scored higher on shorter-term forecasts. Based on case counts as of October 13, the stochastic model projected a median case count of 226 by October 27 (95% prediction interval: 205-268) and 245 by November 10 (95% PI: 208-315), while the auto-regression model projected median case counts of 240 (95% PI: 215-307) and 259 (95% PI: 216-395) for those dates, respectively. Projected median final counts range from 274 to 421. Except for Gott's law, the projected probability of an outbreak surpassing 2013-2016 is exceedingly small. The stochastic model estimates that vaccine coverage in this outbreak is lower than reported in its trial. Based on our projections we believe that the epidemic had not yet peaked at the time of these estimates, though an outbreak like 2013-2016 is not likely. We estimate that transmission rates are higher than under target levels of vaccine coverage, and this model estimate may offer a surrogate indicator for the outbreak response challenges.

研究动机与目标

• 在2018年刚果民主共和国埃博拉疫情期间，生成基于实时数据的传播和最终疫情规模预测。
• 评估在医疗资源有限的冲突地区，疫苗接种对传播动力学的影响。
• 通过将模型预测与2018年8月20日至10月13日期间实际病例数对比，验证短期预测模型的准确性。
• 比较多种预测方法——随机分支过程、自回归模型、Theil-Sen回归和Gott定律——在鲁棒性和准确性方面的表现。
• 评估当前疫情轨迹是否与历史埃博拉疫情一致，或是否表明一种独特且更严重的传播路径。

提出的方法

• 使用随机分支过程模拟传播，基于先前埃博拉疫情数据估计逐渐减弱的传播率。
• 应用负二项分布自回归模型进行短期病例数预测，基于实时病例数据进行条件预测。
• 采用Theil-Sen回归基于历史病例数趋势估计最终疫情规模。
• 应用Gott定律作为最小信息基线预测，形成集合预测。
• 通过将模型预测与2018年8月20日至10月13日期间收集的实际病例数对比，验证1至4周的短期预测。
• 在随机模型中纳入不同疫苗覆盖率情景（高、低、无），使用西非地区获得的有效性估计值，以评估疫苗影响。

实验结果

研究问题

• RQ1 在冲突地区疫情中，实时预测模型在预测短期和长期病例数方面的准确性如何？
• RQ2 当前疫情中疫苗覆盖率在多大程度上与以往疫情中有效接触追踪和疫苗接种所对应的62%目标水平一致？
• RQ3 冲突地区病例检测数的增加是否反映传播率上升，还是监测能力下降？
• RQ4 不同预测方法（随机模型、自回归模型、回归模型、Gott定律）在性能和可靠性方面如何比较？
• RQ5 当前疫情轨迹是否与历史埃博拉疫情一致，还是表明一种独特且更严重的流行病？

主要发现

• 随机模型预测2018年10月27日病例中位数为226例（95%预测区间：205–268），11月10日为245例（95%预测区间：208–315）。
• 自回归模型预测2018年10月27日病例数为240例（95%预测区间：215–307），11月10日为259例（95%预测区间：216–395）。
• 最终疫情规模的中位数预测在274至421例之间，具体取决于模型和假设条件。
• 除Gott定律外，所有模型均认为疫情规模与2013–2016年西非埃博拉疫情相当的可能性极低。
• 随机模型估计实际疫苗覆盖率低于62%的目标水平，表明由于冲突相关干扰，疫苗影响有限。
• 短期预测（1–4周后）始终比长期预测更准确，验证了实时模型更新的实用性。