Ma Zhengwen, Zhu Min, Zhi Chen, Zhang Huaguo, Li Minye, Zhang Nan, Ma Hui
School of Nursing, Southern Medical University, Guangzhou, China.
Department of Infection Control, Sixth Medical Center, PLA General Hospital, Beijing, China.
Front Public Health. 2025 Apr 29;13:1578178. doi: 10.3389/fpubh.2025.1578178. eCollection 2025.
The research aims to develop a human behavior-based model to predict respiratory infectious diseases.
This research employs semi-supervised machine learning techniques in conjunction with an RGB-depth camera to collect micro-level data. We employed computational fluid dynamics to simulate the dispersion of virus concentration in outpatient environments. Furthermore, we evaluated the infection risk of respiratory infectious diseases (RIDs) by utilizing a dose-response model.
A total of 201,600 behavioral data points were collected. The average interpersonal distance observed during medical procedures was 0.62 meters. The most common facial orientation between patients and healthcare workers (HCWs) was face-to-face, accounting for 30.48% of interactions. The predicted average viral RNA load exposures per second during various medical procedures were as follows: Otoscopy: 0.014314 viral RNA loads/s; Rhinoscopy: 0.014411 viral RNA loads/s; Laryngoscopy: 0.014379 viral RNA loads/s; External auditory canal irrigation: 0.018803 viral RNA loads/s. Simulations of preventive measures indicated that N95 masks reduced the probability of infection to 2.44%, surgical masks to 14.81%, and cotton masks to 36.05%.
This research presents an innovative micro-level exposure risk model for respiratory infectious diseases (RIDs), which provides significant insights into the risk of infection. However, it is important to acknowledge certain limitations, including the distinctiveness of the data sources utilized and the insufficient examination of transmission pathways. Subsequent studies should aim to enhance the dataset, fine-tune model parameters, and integrate further transmission pathways to augment both the accuracy and applicability of the model.
本研究旨在开发一种基于人类行为的模型来预测呼吸道传染病。
本研究采用半监督机器学习技术，结合RGB深度相机收集微观层面的数据。我们运用计算流体动力学来模拟门诊环境中病毒浓度的扩散。此外，我们利用剂量反应模型评估呼吸道传染病（RIDs）的感染风险。
共收集了201,600个行为数据点。医疗程序中观察到的平均人际距离为0.62米。患者与医护人员（HCWs）之间最常见的面部朝向是面对面，占互动的30.48%。各种医疗程序期间预测的每秒平均病毒RNA负荷暴露量如下：耳镜检查：0.014314病毒RNA负荷/秒；鼻镜检查：0.014411病毒RNA负荷/秒；喉镜检查：0.014379病毒RNA负荷/秒；外耳道冲洗：0.018803病毒RNA负荷/秒。预防措施模拟表明，N95口罩将感染概率降低到2.44%，外科口罩降低到14.81%，棉质口罩降低到36.05%。
本研究提出了一种创新的呼吸道传染病（RIDs）微观层面暴露风险模型，该模型为感染风险提供了重要见解。然而，必须认识到某些局限性，包括所使用数据源的独特性以及对传播途径的审查不足。后续研究应旨在增强数据集、微调模型参数并整合更多传播途径以提高模型的准确性和适用性。
