深度学习功能在重症监护病房患者死亡关联因素预测中的应用

Application of deep learning function in the prediction of death-related factors for intensive care patients

目的利用FP-Growth和Apriori算法的深度学习(deep learning,DL)功能预测重症监护病房(ICU)患者院内死亡的关联因素。方法筛选美国重症监护医学信息数据库-Ⅲ(medical information mart for intensive care-Ⅲ,MIMIC-Ⅲ)中患者10000例,包含死亡患者1320例,收集其基线资料进行回顾性研究。使用SPSS Modeler 18.0软件编制FP-Growth和Apriori算法程序,通过DL功能计算1320例死亡患者的基线资料间有效强关联规则。对全部患者行Logistic回归分析导致死亡的独立风险因素。参考Logistic回归分析对患者死亡风险的预测结果来验证DL功能的预测结果。结果通过DL功能计算获得死亡患者的基线资料间有效强关联规则9项,其前项包括:年龄、急性生理学与慢性健康状况评估系统Ⅱ(APACHEⅡ)评分、序贯器官衰竭评分(SOFA)、院内感染、机械通气、动静脉插管、动静脉插管时间、导尿管插管。除“肝脏疾病”和“昏迷”外,DL功能同Logistic回归分析预测结果高度一致。两种方法预测结果的比较在一定程度上证实DL功能的科学性和可靠性。结论基于FP-Growth和Apriori算法的DL功能可用于预测ICU患者死亡的关联因素,具有一定应用和推广价值。

Objective To predict the related factors of death for the patients in intensive care unit(ICU)by using the deep learning(DL)function based on FP-Growth and Apriori algorithm.Methods 10000 patients were selected from the medical information mart for intensive careⅢ(MIMICⅢ),including 1320 dead patients.The clinical data of the patients were collected for prospective study.SPSS Modeler 18.0 software was used to compile the FP-Growth and Apriori algorithm program.The effective strong association rules between the clinical data of 1320 dead patients were calculated through DL function.The independent risk factors leading to death were analyzed through Logistic regression.The prediction results of DL function were verified by referring to the prediction results of Logistic regression analysis for the risk of death in the patients.Results 9 effective strong association rules between clinical data of dead patients were obtained through DL function calculation.The first items included age,acute physiology and chronic health evaluationⅡ(APACHEⅡ),sequential organ failure score(SOFA),hospital infection,mechanical ventilation,arteriovenous intubation,arteriovensous intubation time,catheter intubation.Except for"liver disease"and"coma",the DL function result was highly consistent with the prediction results of Logistic regression analysis.The comparison of the predicted results between the two methods confirmed the science and reliability of DL function to some extent.Conclusions DL function based on FP-Growth and Apriori algorithm can be used to predict the death-related factors of patients.This method has a certain clinical application and promotion value.