融合知识驱动和数据驱动的混合决策模型构建:以室性心动过速病因诊断为例  

Constructing A Knowledge⁃driven and Data⁃driven Hybrid Decision Model for Etiological Diagnosis of Ventricular Tachycardia

作　　者：王敏[1] 胡兆 徐晓巍[1] 郑思[1] 李姣[1] 姚焰[2] WANG Min;HU Zhao;XU Xiaowei;ZHENG Si;LI Jiao;YAO Yan(Institute of Medical Information,Chinese Academy of Medical Sciences&Peking Union Medical College,Beijing 100020,China;Arrhythmia Center,Fuwai Hospital,Chinese Academy of Medical Sciences&Peking Union Medical College,National Center for Cardiovascular Diseases,Beijing 100037,China)

机构地区：[1]中国医学科学院北京协和医学院医学信息研究所,北京100020 [2]国家心血管病中心中国医学科学院阜外医院心律失常中心,北京100037

出　　处：《协和医学杂志》2025年第2期454-461,共8页Medical Journal of Peking Union Medical College Hospital

基　　金：中国医学科学院医学与健康重大协同创新项目(2021-I2M-1-056);中央高水平医院临床科研业务费项目(2022-GSP-GG-25)。

摘　　要：目的构建一个融合知识驱动和数据驱动的混合决策模型,并将其应用于室性心动过速的病因诊断。方法检索2018—2023年心律失常疾病领域的临床实践指南、专家共识和医学文献作为知识源,并回顾性收集2013—2023年中国医学科学院阜外医院室性心动过速(ventricular tachycardia,VT)患者的电子病历信息作为数据集。采用基于知识规则的方法构建临床路径作为知识驱动模型;基于真实世界数据构建VT病因诊断三分类机器学习模型,并选取其中的最佳模型作为数据驱动模型代表;以临床路径为基本框架,将机器学习模型以自定义运算符的形式嵌入临床路径的决策节点中,作为混合模型。评价上述3种模型的精确率、召回率和F1分数。结果共纳入3部临床实践指南作为知识驱动模型的知识源;收集了1305条患者数据作为数据集,构建了5种机器学习模型,其中XGBoost模型最佳。混合模型采用知识驱动的决策思维,分别将XGBoost模型嵌入2层分类的决策节点中。3种模型的精确率、召回率和F1分数如下:知识驱动模型为80.4%、79.1%和79.7%;数据驱动模型分别为88.4%、88.5%和88.4%;混合模型分别为90.4%、90.2%和90.3%。结论融合知识与数据驱动的混合模型展现出更高的准确性,且混合模型的所有决策结果均基于循证证据,这更接近临床医生的实际诊断思维。未来需更严格地验证混合模型广泛应用于医学领域的可行性。Objective To construct a hybrid decision⁃making model that integrates knowledge⁃driven and data⁃driven approaches,and to apply it to the etiological diagnosis of ventricular tachycardia(VT).Methods Clinical practice guidelines,expert consensus documents,and medical literature in the field of ar⁃rhythmia diseases from 2018 to 2023 were retrieved as knowledge sources.Retrospective electronic medical re⁃cord data of VT patients from Fuwai Hospital,Chinese Academy of Medical Sciences&Peking Union Medical College,from 2013 to 2023 were collected as the dataset.A knowledge⁃driven model was constructed using a knowledge⁃rule⁃based approach to establish clinical pathways.A three⁃class machine learning model for VT eti⁃ology diagnosis was developed based on real⁃world data,and the best⁃performing model was selected as the rep⁃resentative of the data⁃driven approach.The machine learning model was embedded into the decision nodes of the clinical pathway in the form of custom operators,forming the hybrid model.The precision,recall,and F1 score of the three models were evaluated.Results Three clinical practice guidelines were included as knowl⁃edge sources for the knowledge⁃driven model.A total of 1305 patient records were collected as the dataset,and five machine learning models were constructed,with the XGBoost model performing the best.The hybrid model adopted a knowledge⁃driven decision⁃making framework,embedding the XGBoost model into the decision nodes of a two⁃level classification.The precision,recall,and F1 scores of the three models were as follows:the knowledge⁃driven model achieved 80.4%,79.1%,and 79.7%;the data⁃driven model achieved 88.4%,88.5%,and 88.4%;and the hybrid model achieved 90.4%,90.2%,and 90.3%.Conclusions The hybrid model integrating knowledge⁃driven and data⁃driven approaches demonstrated higher accuracy,and all its deci⁃sion outcomes were based on evidence⁃based practices,aligning more closely with the actual diagnostic reason⁃ing of clinicians.Furt

关 键 词：室性心动过速 知识驱动 数据驱动 混合模型 决策支持 

分 类 号：R541.7[医药卫生—心血管疾病] TP35[医药卫生—内科学]