| Objective:To construct traditional Chinese medicine(TCM)intelligent diagnostic models for unstable angina pectoris(UAP)in coronary heart disease based on machine learning algorithms,evaluate the classification prediction performance of the machine learning model,and provide artificial intelligence-assisted methods for the diagnosis and treatment of cardiovascular diseases.Methods:Clinical medical records of 400 inpatients from the Department of Cardiovascular Medicine of Dong fang hospital of Beijing University from January to December 2020 were collected.After data cleaning and preprocessing,a clinical dataset was established,and 382 cases that met the inclusion criteria were finally included.SPSS,Medcalc,and GraphPad Prism software were used to perform descriptive statistics,parametric tests(independent samples Ttest,one-way ANOVA),and non-parametric tests(Kruskal-Wallis H test)to analyze the general demographic characteristics of the study subjects,the distribution characteristics of TCM syndromes in coronary heart disease unstable angina pectoris,and their relationship with the general demographic characteristics of the study subjects.Based on the least absolute shrinkage and selection operator(LASSO)method in Python 3.11 software,feature variables closely related to the differentiation of UAP syndromes were selected as predictive factors,and the rationality of the feature variables was discussed through literature research and professional consultation.In addition,the contribution of feature variables to syndrome differentiation was quantified through the SHAP machine learning model interpretability analysis.Finally,the clinical data were divided into a training set and a test set at a ratio of 8:2,and the above feature variables were included in the construction of the TCM intelligent diagnostic model for coronary heart disease unstable angina pectoris.Gaussian Naive Bayes,Random Forest,Extreme Gradient Boosting(XGBoost),Light Gradient Boosting Machine(LightGBM),and Gradient Boosting Decision Tree(GDBT)parameters were optimized through grid search and cross-validation,and TCM intelligent diagnostic models were established separately.The predictive performance of the models was evaluated by the area under the receiver operating characteristic curve(AUC),accuracy,precision,recall,and F1 score.Results:1.Among the 382 study subjects,there were 203 males(53.14%)and 179 females(46.61%),with an average age of 63.86±10.22 years and an age range of 38-77 years.There were more male than female patients in the study,and the overall age was relatively high.The distribution of TCM syndromes and their proportions from high to low were as follows:phlegm-turbidity obstruction syndrome 121(31.68%),qi deficiency and blood stasis syndrome 108(28.27%),phlegm and blood stasis syndrome 100(26.18%),qi stagnation and blood stasis syndrome 53(13.9%),and other syndromes(not included in the case analysis due to the small number).In TCM syndromes,male patients had a higher proportion of phlegm and blood stasis syndrome and phlegm-turbidity obstruction syndrome,while female patients had a higher proportion of qi stagnation and blood stasis syndrome,and qi deficiency and blood stasis syndrome.Patients with qi deficiency and blood stasis syndrome had a significantly higher average age than those with other syndromes.Hypertensive patients were more common in phlegm and blood stasis syndrome and qi stagnation and blood stasis syndrome,hyperlipidemia patients were more common in phlegm-turbidity obstruction syndrome,and diabetic patients were more common in qi deficiency and blood stasis syndrome.2.Predictive factors closely related to the TCM differentiation of UAP were screened through the LASSO feature variable selection algorithm,and their overall contribution was ranked as follows:prothrombin time,fibrinogen content,total bilirubin,small and low-density lipoprotein cholesterol,D-dimer,jaundice index,homocysteine,indirect bilirubin,white blood cell count,thrombin time,red blood cell distribution width coefficient of variation,and direct bilirubin.The contribution(correlation)of each feature variable to the TCM syndrome of coronary heart disease unstable angina pectoris was different.Among them,prothrombin time,homocysteine,D-dimer,and small and low-density lipoprotein cholesterol had higher predictive contributions to the differentiation of phlegm-turbidity obstruction syndrome;Ddimer,total bilirubin,icterus index,homocysteine,indirect bilirubin,fibrinogen content,prothrombin time,and small and low-density lipoprotein cholesterol had higher predictive contributions to the differentiation of qi deficiency and blood stasis syndrome;prothrombin time,total bilirubin,icterus index,small and low-density lipoprotein cholesterol,fibrinogen content,and red blood cell distribution width had higher predictive contributions to the differentiation of phlegm and blood stasis syndrome;fibrinogen content,jaundice index,total bilirubin,small and low-density lipoprotein cholesterol,thrombin time,and red blood cell distribution width had higher predictive contributions to the differentiation of qi stagnation and blood stasis syndrome.3.The prediction performance of the five TCM intelligent diagnostic models constructed in this study was excellent(AUC>0.900),with performance ranking from high to low as follows:XGBoost,LightGBM,GDBT,Random Forest,and Naive Bayes.The machine learning model based on the XGBoost algorithm was the best performing TCM intelligent diagnostic model in this study.Conclusion:1.This study reveals a certain pattern in the distribution of traditional Chinese medicine(TCM)syndrome types in patients with unstable angina pectoris(UAP).The frequency of each syndrome type is as follows:phlegm turbidity obstructing syndrome,qi deficiency and blood stasis syndrome,phlegm and blood stasis intertwining syndrome,and qi stagnation and blood stasis syndrome.Among them,phlegm turbidity obstructing syndrome is more common in males and patients with hyperlipidemia,while qi deficiency and blood stasis syndrome are more prevalent in females,elderly individuals,and patients with hypertension.Phlegm and blood stasis intertwining syndrome is more common in males with hypertension and hyperlipidemia,while qi stagnation and blood stasis syndrome is more prevalent in females with hypertension.2.From a modern medical perspective,there is a close relationship between TCM syndrome types in UAP and clinical laboratory indicators such as inflammation,blood lipids,coagulation,and hemorheology.According to TCM syndrome differentiation,elevated blood lipids and inflammation indicators are mainly associated with phlegm turbidity obstructing syndrome,increased coagulation and inflammation indicators are mainly associated with qi deficiency and blood stasis syndrome,elevated blood lipids and coagulation indicators are primarily related to phlegm and blood stasis intertwining syndrome,and increased coagulation,hemorheology,and blood lipid indicators are predominantly associated with qi stagnation and blood stasis syndrome.According to the TCM pathogenesis,inflammation and blood lipid indicators are correlated with "phlegm" syndrome,while the coagulation-fibrinolysis system and hemorheology indicators are associated with "stasis" and "qi stagnation" syndromes.3.Considering the research objectives,data types,and algorithm characteristics,this study employed the XGBoost algorithm to construct a machine learning model for intelligent TCM syndrome differentiation in UAP.The aim is to guide clinical decision-making and promote the objective and standardized development of TCM syndrome differentiation. |
PDF Full Text Request