中国全科医学

• •

基于特征性中医四诊信息构建冠心病预测模型的研究

李柯霖1,欧洋2,肖蕾3*   

  1. 1.110032 辽宁省沈阳市,辽宁中医药大学第一临床学院 2.110032 辽宁省沈阳市,辽宁中医药大学附属医院治未病科 3.110032 辽宁省沈阳市,辽宁中医药大学附属医院心内五科
  • 收稿日期:2025-07-31 接受日期:2025-09-01
  • 通讯作者: 肖蕾,主任中医师
  • 基金资助:
    沈阳市科技计划项目(22-321-34-14)

Development and External Validation of a Coronary Heart Disease Prediction Model Incorporating Characteristic TCM Four-Diagnostic Features

LI Kelin1, OU Yang2, XIAO Lei3*   

  1. 1.First Clinical College of Liaoning University of Traditional Chinese Medicine, Shenyang 110032, China 2.Department of Preventive Treatment, Affiliated Hospital of Liaoning University of Traditional Chinese Medicine, Shenyang 110032, China 3.Fifth Department of Cardiology, Affiliated Hospital of Liaoning University of Traditional Chinese Medicine, Shenyang 110032, China
  • Received:2025-07-31 Accepted:2025-09-01
  • Contact: XIAO Lei, Chief Physician of Traditional Chinese Medicine

摘要: 背景 冠心病的筛查可以借助诊断预测模型,但目前较为公认的预测模型多是基于西医研究,如果能发挥中医“司外揣内”的优势,则可以提高模型的准确性。目的 基于特征性中医四诊信息,构建冠心病无创诊断预测模型。方法 采用临床流行病学调查法,选择2022年5月—2023年4月于辽宁中医药大学附属医院心内科门诊或病房就诊的冠心病患者138例、非冠心病患者125例,以此临床信息作为数据来源。采用Logistic回归分析结合临床专家经验,构建中医冠心病预测模型,并通过外部验证评价模型。结果 中医冠心病诊断预测模型的公式:P=1/{1+exp[-(2.396×年龄男性≥45,女性≥55)+1.435×(合并疾病≥3个)+1.981×典型心绞痛+2.354×目内眦血络增粗、色红、延长+1.561×(瘀血表现≥2个)+1.266×(肾虚表现≥2个)-4.302]},当截断值为-0.502 5时,AUC最大为0.953,约登指数最佳为0.760,外部验证结果提示其AUC为0.881,且与目前较为公认的更新的Diamond Forrester模型(UDFM)比较,净重新分类改善指数(NRI)=0.275、综合区分改善度(IDI)=56.95。结论 中医冠心病预测模型能够准确预测冠心病,较UDFM的诊断效能更具优势,可以作为临床筛查冠心病的无创参考工具。

关键词: 冠心病;预测学习模型;诊断;中国医学, 传统

Abstract: Background Diagnostic prediction models facilitate screening for coronary heart disease (CHD), but most existing models derive from Westernmedicine variables. Incorporating characteristic traditional Chinese medicine (TCM) fourdiagnostic information, leveraging TCM's integrated external assessment of internal pathology, may enhance predictive performance. Objective To develop and externally validate a TCM CHD diagnostic prediction model integrating characteristic TCM fourdiagnostic features. Methods In a clinical epidemiologic study, we enrolled consecutive patients evaluated in the Cardiology Outpatient Clinic and Wards at the Affiliated Hospital of Liaoning University of Traditional Chinese Medicine between May 2022 and April 2023: 138 patients with confirmed CHD and 125 nonCHD patients. Clinical and TCM fourdiagnostic features were collected. Candidate predictors were selected using clinical expertise and statistical methods, and a multivariable logistic regression model was derived. External validation was performed on an independent dataset. Results The final predictive equation: P=1/{1+exp[-(2.396×age indicator+1.435×(comorbidities ≥ 3)+1.981×typical angina+2.354×conjunctival canthal vascular changes+1.561×(bloodstasis_signs ≥ 2)+1.266×(kidney deficiency signs ≥ 2)-4.302}, where age indicator=1 for men≥45 years or women ≥ 55 years, and the other variables are binary indicators as described. At an optimal cutoff (Logit=-0.5025), the model achieved an area under the receiver operating characteristic curve (AUC) of 0.953 and an optimal Youden index of 0.760 in the development sample. External validation yielded an AUC of 0.881. Compared with the updated diamond-forrester method (UDFM), the TCM model produced net reclassification improvement (NRI)=0.275 and integrated discrimination improvement (IDI)=56.95, indicating net improvement in classification and discrimination. Conclusion A prediction model that integrates characteristic TCM fourdiagnostic information accurately identifies patients with CHD and outperforms the UDFM in our datasets.This TCMbased noninvasive tool may aid clinical CHD screening.

Key words: Coronary disease, Predictive learning models, Diagnosis, Traditional Chinese medicine

中图分类号: