Database for the modelling framework.

Background
The dynamics of the spread of cholera epidemics in the Democratic Republic of the Congo (DRC), from east to west and within western DRC, have been extensively studied. However, the drivers of these spread processes remain unclear. We therefore sought to better understand the factors associated with these spread dynamics and their potential underlying mechanisms.
Methods
In this eco-epidemiological study, we focused on the spread processes of cholera epidemics originating from the shores of Lake Kivu, involving the areas bordering Lake Kivu, the areas surrounding the lake areas, and the areas out of endemic eastern DRC (eastern and western non-endemic provinces). Over the period 2000–2018, we collected data on suspected cholera cases, and a set of several variables including types of conflicts, the number of internally displaced persons (IDPs), population density, transportation network density, and accessibility indicators. Using multivariate ordinal logistic regression models, we identified factors associated with the spread of cholera outside the endemic eastern DRC. We performed multivariate Vector Auto Regressive models to analyze potential underlying mechanisms involving the factors associated with these spread dynamics. Finally, we classified the affected health zones using hierarchical ascendant classification based on principal component analysis (PCA).
Findings
The increase in the number of suspected cholera cases, the exacerbation of conflict events, and the number of IDPs in eastern endemic areas were associated with an increased risk of cholera spreading outside the endemic eastern provinces. We found that the increase in suspected cholera cases was influenced by the increase in battles at lag of 4 weeks, which were influenced by the violence against civilians with a 1-week lag. The violent conflict events influenced the increase in the number of IDPs 4 to 6 weeks later. Other influences and uni- or bidirectional causal links were observed between violent and non-violent conflicts, and between conflicts and IDPs. Hierarchical clustering on PCA identified three categories of affected health zones: densely populated urban areas with few but large and longer epidemics; moderately and accessible areas with more but small epidemics; less populated and less accessible areas with more and larger epidemics.
Conclusion
Our findings argue for monitoring conflict dynamics to predict the risk of geographic expansion of cholera in the DRC. They also suggest areas where interventions should be appropriately focused to build their resilience to the disease.

### 研究背景
霍乱疫情在刚果民主共和国（Democratic Republic of the Congo, DRC）境内自东向西、以及该国西部境内的传播动态已得到广泛研究，但此类传播过程的驱动因素仍不明确。为此，本研究旨在深入探究与该传播动态相关的影响因素及其潜在作用机制。

### 研究方法
本项生态流行病学（eco-epidemiological）研究聚焦于源自基伍湖（Lake Kivu）沿岸的霍乱疫情传播过程，涉及基伍湖沿岸区域、环湖区域以及东部地方性流行区域之外的刚果民主共和国区域（东部和西部非地方性流行省份）。在2000年至2018年期间，我们收集了疑似霍乱病例数据，以及一系列变量数据，包括冲突类型、国内流离失所者（internally displaced persons, IDPs）数量、人口密度、交通网络密度以及可达性指标。本研究采用多变量有序logistic回归模型，识别出与东部地方性流行区域之外的霍乱传播相关的影响因素；通过多变量向量自回归（Vector Auto Regressive, VAR）模型，分析与上述传播动态相关的潜在作用机制；最后基于主成分分析（principal component analysis, PCA）的分层聚类法，对受影响的卫生区进行分类。

### 研究结果
东部地方性流行区域的疑似霍乱病例数增加、冲突事件加剧以及国内流离失所者数量上升，均与霍乱向东部非地方性流行省份扩散的风险升高显著相关。本研究发现，疑似霍乱病例数的增加受到4周滞后阶数的战斗次数增加的影响，而战斗次数的增加又受到1周滞后阶数的针对平民的暴力事件的影响。暴力冲突事件会在4至6周后导致国内流离失所者数量上升。此外，本研究还观察到暴力冲突与非暴力冲突之间、冲突与国内流离失所者之间存在单向或双向因果关联及其他影响路径。基于主成分分析的分层聚类将受影响卫生区划分为三类：人口稠密的城市区域，此类区域疫情规模较大、持续时间较长但暴发次数较少；中等人口密度且可达性良好的区域，此类区域疫情暴发次数较多但规模较小；人口稀少且可达性较差的区域，此类区域疫情暴发次数较多且规模较大。

### 研究结论
本研究结果表明，应通过监测冲突动态来预测刚果民主共和国境内霍乱的地理扩散风险；同时也提示，应针对性地在重点区域开展干预措施，以提升这些区域对霍乱的抵御能力。