Abstract: Background During the process of major epidemic prevention and control, rapid assessment of the risk level of infectious disease transmission is an important basis for subsequent emergency response measures. Objective This study aims to develop a risk assessment model based on machine learning (ML) algorithms to quickly identify high-risk places for respiratory infectious disease transmission, thereby facilitating rapid and precise epidemic prevention and control. Methods Data in 201 public places involving 43 COVID-19 cases in a District of Beijing during the implementation of the "dynamic zero-COVID" policy in April 2022 were collected, including place type, area, ventilation conditions, and personnel flow. An ML algorithm-based model with the secondary attack risk value of the case activity place as the dependent variable was created to assess the risk level of the infected places. Results The results of the Random Forest (RForest), Gradient Boosting Decision Tree (GBDT) and Extreme Gradient Boosting (XGBoost) models for the occurrence of close-contact transitions in the investigated places showed that GBDT performed best with an area under the curve (AUC) of 0.78 and a sensitivity of 0.647. The initial Ct value of the first case (the lowest value), health treasure scan code record query, mask wearing status of the first case, and residence time of the first case were the most important features in the model. Stratified by the place type, areas with high traffic, such as Chinese restaurants, fast food lunch boxes and bars weighed importantly. Conclusion The GBDT model established in this study can provide effective assistance for risk assessment of respiratory infectious disease epidemic sites, and help policy makers prioritize and make decisions under limited resources. Given the possible similar transmission mechanisms among respiratory infectious diseases, this model is expected to provide a reference for risk assessment of other respiratory infectious diseases.

Key words: Respiratory infectious diseases, Risk assessment, Machine learning, Public health

摘要： 背景 在重大疫情防控过程中，快速评估传染病传播风险等级是后续采取应急处置措施的重要基础。目的 基于机器学习算法，开发能快速识别呼吸道传染病传播高风险场所的风险评估模型，从而有利于快速精准防控疫情。方法 收集2022年4月中国实施“动态清零”政策期间，北京市某区43例新型冠状病毒感染病例涉及的 201个场所的相关数据，包括场所类型、面积、通风条件、人员流动情况等，以病例活动场所的续发风险值为因变量构建了机器学习算法模型，以评估涉疫场所的风险等级。结果 场所发生密接转阳的随机森林（RForest）、梯度提升决策树（GBDT）和极限梯度提升（XGBoost）模型效果显示，GBDT 模型以受试者工作特征曲线下面积（AUC）=0.78和灵敏度=0.647表现最佳。首例病例首次Ct值（最低值）、健康宝扫码记录查询、首例病例口罩佩戴情况、首例病例停留时间是模型中显示的重要特征。在细分场所类型，中餐馆、快餐便当以及酒吧等人流量比较大的区域相对重要性更高。结论 本研究建立的GBDT评估模型，可为呼吸道传染病疫情场所的风险评估提供有效帮助，有助于政策决策者在资源有限的情况下进行优先级排序和决策。鉴于呼吸道传染病间可能存在的相似传播机制，该模型有望为其他呼吸道传染病的风险评估提供参考。

关键词: 呼吸道传染病, 风险评估, 机器学习, 公共卫生