基于医院信息系统多源时序数据的急性脑卒中预测研究

Research on Acute Stroke Prediction Based on Multi-source Temporal Data from Information Systems in Hospitals

目的 开发能够实时预测急性脑卒中的模型,辅助医生提高脑卒中的早期识别。方法 利用无感式智能床实时监测数据,结合医院多源信息系统时序数据,通过逻辑回归、随机森林和门控循环单元(GRU)等人工智能算法,构建脑卒中实时预测模型。采用夏普利可加性解释值对重要预测因子进行排序。结果 短暂性脑缺血发作和急性缺血性脑卒中预测的最佳表现均来自GRU模型,受试者工作特征曲线下面积(AUROC)分别为0.79和0.94,急性大血管闭塞性缺血性脑卒中预测的最佳表现来自随机森林模型,AUROC为0.89。年龄、性别、疾病史以及血压、呼吸、脉搏对脑卒中的发生是重要的。结论 本研究通过结合医院信息系统多模态时序数据有效地识别了急性脑卒中,并揭示了脑卒中常见危险因素外的重要预测因子。

Objective To develop a model capable of providing real-time predictions for acute stroke and assist doctors to improve the early identification of stroke.Methods We constructed real-time stroke prediction models by integrating continuous monitoring data from non-invasive smart beds with temporal data from multi-source hospital information systems.Artificial intelligence algorithms such as logistic regression,random forest,and gated recurrent units(GRU)were employed to build these models.The Shapley additive explanations values were used to rank the importance of predictive factors.Results The best performance for predicting transient ischemic attacks and acute ischemic stroke were achieved by the GRU model, with area under the receiver operating characteristic curve (AUROC) values of 0.79 and 0.94, respectively. For acute large vessel occlusive ischemic stroke, the random forest model demonstrated the best performance, with an AUROC of 0.89. Age, gender, medical history, along with blood pressure, respiration, and pulse rate were identified as significant factors influencing the occurrence of stroke. Conclusion This study effectively identified acute stroke by combining multimodal temporal data from hospital information systems, and revealed important predictive factors beyond common risk factors for stroke.