ObjectiveFrom the perspective of cardiovascular-kidney-metabolic （CKM） syndrome， this study aims to construct and validate a diagnostic machine learning model integrating traditional Chinese and Western medicine for severe coronary artery stenosis in patients with CKM， thereby providing clinical decision-making support for patients with borderline lesions. MethodsBased on a retrospective study design， a total of 535 hospitalized patients from two independent campuses of Jiangsu Province Hospital of Traditional Chinese Medicine： the main campus （from January 2024 to August 2024） and Zidong branch （from September 2024 to December 2024） were screened. Data from the main campus were randomly divided into the training dataset （376 cases） and the internal validation dataset （95 cases） at a 4∶1 ratio， while data from Zidong branch served as the external validation dataset （64 cases）. Risk factors were analyzed and screened through literature review， expert interviews， and the least absolute shrinkage and selection operator （LASSO） algorithm. Nine machine learning algorithms were utilized to construct diagnostic models. Comparative analyses of common evaluation metrics， calibration curves， and decision curves were conducted to select the model through internal and external validation. The Shapley additive explanations （SHAP） method and two cases were utilized to help understand the operational logic of the best model. Finally， the best model was applied to patients with borderline lesions to calculate diagnostic efficacy. ResultsNine risk factors were screened by LASSO regression， such as phlegm， hematoma， stagnation， deficiency， hypertension duration， gender， arterial stiffness index （ASI）， uric acid to high-density lipoprotein cholesterol ratio （UHR）， and glycosylated hemoglobin （HbA1c）. After comparison from multiple dimensions， the light gradient boosting machine （LightGBM） was the best model， achieving area under the curve （AUC） of 0.918 （95% confidence interval （CI）： 0.890-0.945） in the training dataset， 0.885 （95%CI： 0.820-0.951） in the internal validation dataset， and 0.897 （95%CI： 0.818-0.975） in the external validation dataset. Calibration curves indicated good consistency in the predicted probabilities， while decision curve analysis showed clinical benefit when threshold probabilities were less than 90%. SHAP importance rankings were stagnation， deficiency， hematoma， HbA1c， gender， phlegm， hypertension duration， ASI， and UHR. When applied to the patients with borderline lesions， the diagnostic model achieved an AUC of 0.783 （95%CI： 0.637-0.930）， with 73% of patients with actual severe stenosis getting benefit. ConclusionGuided by clinical value， the diagnostic model integrating traditional Chinese and Western medicine established in this study demonstrates favorable performance， providing a basis for clinical diagnosis， treatment， and decision-making in patients with CKM.

ObjectiveFrom the perspective of cardiovascular-kidney-metabolic （CKM） syndrome， this study aims to construct and validate a diagnostic machine learning model integrating traditional Chinese and Western medicine for severe coronary artery stenosis in patients with CKM， thereby providing clinical decision-making support for patients with borderline lesions. MethodsBased on a retrospective study design， a total of 535 hospitalized patients from two independent campuses of Jiangsu Province Hospital of Traditional Chinese Medicine： the main campus （from January 2024 to August 2024） and Zidong branch （from September 2024 to December 2024） were screened. Data from the main campus were randomly divided into the training dataset （376 cases） and the internal validation dataset （95 cases） at a 4∶1 ratio， while data from Zidong branch served as the external validation dataset （64 cases）. Risk factors were analyzed and screened through literature review， expert interviews， and the least absolute shrinkage and selection operator （LASSO） algorithm. Nine machine learning algorithms were utilized to construct diagnostic models. Comparative analyses of common evaluation metrics， calibration curves， and decision curves were conducted to select the model through internal and external validation. The Shapley additive explanations （SHAP） method and two cases were utilized to help understand the operational logic of the best model. Finally， the best model was applied to patients with borderline lesions to calculate diagnostic efficacy. ResultsNine risk factors were screened by LASSO regression， such as phlegm， hematoma， stagnation， deficiency， hypertension duration， gender， arterial stiffness index （ASI）， uric acid to high-density lipoprotein cholesterol ratio （UHR）， and glycosylated hemoglobin （HbA1c）. After comparison from multiple dimensions， the light gradient boosting machine （LightGBM） was the best model， achieving area under the curve （AUC） of 0.918 （95% confidence interval （CI）： 0.890-0.945） in the training dataset， 0.885 （95%CI： 0.820-0.951） in the internal validation dataset， and 0.897 （95%CI： 0.818-0.975） in the external validation dataset. Calibration curves indicated good consistency in the predicted probabilities， while decision curve analysis showed clinical benefit when threshold probabilities were less than 90%. SHAP importance rankings were stagnation， deficiency， hematoma， HbA1c， gender， phlegm， hypertension duration， ASI， and UHR. When applied to the patients with borderline lesions， the diagnostic model achieved an AUC of 0.783 （95%CI： 0.637-0.930）， with 73% of patients with actual severe stenosis getting benefit. ConclusionGuided by clinical value， the diagnostic model integrating traditional Chinese and Western medicine established in this study demonstrates favorable performance， providing a basis for clinical diagnosis， treatment， and decision-making in patients with CKM.