基于连续无创生理参数驱动的死亡事件动态预警

打开文本图片集

中图分类号：TP391；R195.1 文献标志码：A 文章编号：1671-024X（2025）04-0013-06

Dynamic warning of mortality events based on continuous noninvasive physiological parameters

ZHAO Zhe1，2，3， ZHAO Xinhao1，2.3， GUO Yu1，2 ， XU Jiameng1，2，3， GENG Xueqiao4 （1.SchoolofLifeScience，Tiangong University，Tianjin3OO387，China；2.ResearchCenterforFuture TextileTechnolo gy，Tiagong University，Tianjin30o387，China;3.Tianjin KeyLaboratoryof QualityControlandEvaluation Technology for Medical Devices，Tiangong University，Tianjin3Oo387，China; 4. University Hospital，Tiangong UniversityTianjin 300387，China）

Abstract：Toachieve dynamicearlywarningof mortality events incriticallyillpatientsbasedonnon-invasivephysiological parameters，3O parameters from 4 738 adult patients were extracted from the eICU Collaborative Research Database.After preprocessing including one-hot encoding，samplingrate consistency processing，and data imputation， the data were input into and used to train a LightGBM model. A lookahead window was added between the learning windowand the prediction window to provide valuabledecision-making time fordoctors，anda1O-fold cross-validationmethod was used tooptimize the model parameters，constructing theoptimaldynamic early warning model for mortality events.The results showed that the prediction accuracyof the constructed model was O.852，and the area under the ROCcurve was O.875.The predictive performanceof the model was positively corelated with the learning window and negatively correlated with the lookahead window.Thedynamic early warning model based oncontinuousnon-invasiveparameterscantimelydetect mortalityevents incriticallillpatients，avoiding thedependenceon laboratory parameters intraditional methods，gaining valuable foresight time fortheformulationof patient reatment strategies，and greatly improving the work efficiency of medical staff and patient treatment outcomes.

Key words： non-invasive parameter; machine learning; death events prediction; dynamic warning

随着电子病历数据的不断积累，传统的统计方法在处理高维度和庞杂数量的重症监护数据时显现出效率低和准确性低等不足之处[1-3]。（剩余10238字）