Abstract：Objective To study the ventilation alarms of ventilators in clinical use by applying machine learning methods, obtain the
important parameters affecting the alarms and the alarm prediction model, identify invalid alarms and give clinical hints, so that
the clinic can respond to the ventilator alarms efficiently to avoid the negative effects of alarm fatigue and other negative impacts.
Methods A respiratory data management platform was established that conformed to standard data processes. According to the
alarm information of single center ventilator, the characteristic values were analyzed and the important parameters were sorted.
Hyperparameter optimization modeling was used to predict the true or false alarm. The confusion matrix and receiver operating
characteristic (ROC) were used to validate the machine learning model. Results The test set of 5936 ventilation alarms was
evaluated, with 88% invalid alarms rate (recall rate was 0.88). The model accuracy was 0.94, and the precision was 0.78, the area
under ROC curve was 0.92. The F1 score was 0.82. Conclusion The use of machine learning facilitates clinical single-center data
modeling can timely analyze and obtain the important parameters and alarm predictions of the real alarm of the ventilator, and
through the ventilator data management platform, it can effectively prompt the clinical invalid alarms, thus reducing the pressure of
the alarms on the healthcare personnel and improving the quality of medical care.
刘强a，郭瑞b，王勤c，孙凯a. 基于机器学习对呼吸机报警分析[J]. 中国医疗设备, 2024, 39(3): 53-57.
LIU Qianga, GUO Ruib, WANG Qinc, SUN Kaia. Analysis of Ventilator Alarms Based on Machine Learning. China Medical Devices, 2024, 39(3): 53-57.
