Establishment And Validation Of Prediction Model For Elderly Patients With Coronary Heart Disease Complicated With Heart Failure Based On Real World

Background:Cardiovascular disease(CVD)is the leading cause of death and disability,and its incidence is on the rise globally.Coronary Heart Disease(CHD)is one of the most common CVD,and its high hospitalization rate and high mortality pose serious challenges to public health in China.Since traditional disease prediction models cannot handle high-dimensional medical big data,Artificial Intelligence(AI)needs to be introduced to integrate and develop an efficient,convenient and widely applicable CHD diagnosis model to meet the health needs of China’s aging population.It is of great significance to assist doctors in clinical decision-making,reduce medical cost and optimize medical resource allocation.Objective:To develop and validate predictive models for HF in patients with CHD based on large-scale multi-center medical data,improving the clinical application value of the model,providing a basis for clinical auxiliary diagnosis,realizing the construction of HF early warning,and promoting clinical monitoring and intervention.Materials and Methods:1.Bibliometric analysis method and software including COOC,Citespace5.6.R2,VOSviewer,Pajek and Scimago Graphica were used for visualization of AI in CVD during 2003～2022 publications and selection of technique or method of AI to construct CVD prediction model.2.Collected patient data from the big data platform of the Institute of Medical Data of Chongqing Medical University,and used SPSS 25.0software to analyze the demographic characteristics,length of stay,hospitalization cost and disease spectrum of inpatients in the department of cardiovascular medicine;SQL Server is used to preprocess the original data encoded by diagnosis,to realize data integration and conversion,and Apriori algorithm is used to calculate the major diagnosis and other frequent item sets of diagnosis of patients,and to identify the CVD diseases and populations that need the most attention.3.Literature was obtained and screened from Pub Med,Web of Science,Cochrane Library,Embase,CNKI,Chinese Biomedical Literature Database,Wanfang Database.Meta-analysis was performed by Stata17.0software to discover the risk factors for HF in CHD patients.4.The data of elderly CHD patients from the big data platform of the Institute of Medical Data of Chongqing Medical University and a hospital in Sichuan Province were obtained,and the demographic characteristics,disease diagnosis code and laboratory test results were extracted,and the data were preprocessed.Next,the index filtered by the Least Absolute Shrinkage and Selection Operator(LASSO)is used to model CHD concurrent HF.Logistic Regression(LR),Random Forest(RF),Decision Tree(DT),Support Vector Machine(SVM),SVM),Extreme Gradient Boosting(XGBoost),Artificial Neural Network(ANN),K-Nearest Neighbor,KNN),Lightweight Gradient Lifting Machine(light GBM),and Naive Bayes(Naive Bayes,NB)Nine algorithms were used to construct Acute Coronary Syndrome(ACS)and Chronic Coronary Syndrome(CCS)in elderly patients in modeling group,internal validation group,temporal external validation group and spatial external validation group,respectively.concurrent HF prediction model;Decision curve analysis(DCA)and SHAP were used to evaluate the clinical practicality of the model and rank the importance of the indicators,respectively.Results:1.The research on AI integration into CVD field in China is at an explosive stage.Research hotspots mainly focus on CVD prediction model construction based on Machine Learning(ML),Electrocardiogram(ECG)classification based on feature extraction,and Deep Learning(ECG).DL)image segmentation,robotic system based catheter ablation or minimally invasive surgery,and Natural Language Processing(NLP)based Electronic Health Record(EHR)feature extraction.2.Coronary heart disease(CHD)was the leading cause of hospitalization,accounting for 36.8% of the total,followed by hypertension and chronic obstructive pulmonary disease(COPD),accounting for 12%and 7.3% of the total,respectively.At the same time,CHD was also the first disease in the 45～64 group and the 65-plus group,and hypertension is the first disease in the 18～64.According to the association rules,"I50.903(NYHA class II)→I20-I25(CHD)" and "I50.904(NYHA class III)→I20-I25(CHD)" were the most important.3.Body Mass Index(BMI),diabetes,age,hypertension,History of Myocardial Infarction(History of MI),Left Ventricular ejection fraction(LVEF)Ejection Fraction,LVEF),History of Atrial Fibrillation,History of AF),dyslipidemia,White Blood Cell Count(WBC),C-reaction protein(CRP),and COPD,stroke,Creatinine(Cr)>93.6,Low-Density Lipoprotein Cholesterol(LDL-C)>2.46,Left atrial volume index(LAVI),E/A,and Global Longitudinal Strain(GLS)were CHD Risk factors for HF(P<0.05).4.(1)In the modeling group,the AUC value of the ACS concurrent HF diagnosis and prediction model constructed by XGBoost reached 0.957(95%CI: 0.950-0.965),and the AUC value of the RF model reached 0.787 in the internal verification(95%CI: 0.758-0.816),the AUC value of XGBoost model in temporal external validation was 0.750(95%CI:0.698-0.802),and the AUC value of LR model in spatial external validation was 0.70(95%CI: 0.675-0.724).The prediction model of ACS complicated with HF built by XGBoost has the best overall clinical practicability.The importance of risk factors for ACS complicated with HF in the elderly was as follows: hypertension,RDW,WBC,diabetes,LMR,total cholesterol,hyperlipidemia and chronic gastritis.(2)In the modeling group,the AUC value of the elderly CCS concurrent HF prediction model constructed by XGBoost was 0.877(95%CI: 0.868-0.885),the AUC value of NB model in time external verification was 0.746(0.721,0.770),and the AUC value of ANN model under the ROC curve in space external verification was 0.736(95%CI: 0.720-0.753).The prediction model of CCS with HF constructed by XGBoost has the best overall clinical practicability.The importance of risk factors for CCS with HF in the elderly was as follows: hypertension,UA,RDW,albumin,total cholesterol,age,chronic gastritis,PLR and whole blood calcium.The prediction results of the prediction model of ACS and CCS complicated with HF in the elderly were consistent with the actual clinical outcomes of patients,and the results of risk factor analysis were also consistent with clinical experience.Conclusion:1.The development of ML prediction model based on medical big data is one of the research hotspots at present.ML especially has advantages in building data-driven inference and prediction models as a method to build CVD prediction models.2.CHD is the most common disease in the department of cardiovascular medicine.Heart Failure(HF)is the main complication of CHD,and elderly people over 65 are used as research objects.3.BMI,diabetes,age,hypertension,History of MI,History of AF,LVEF,dyslipidemia,WBC,CPR,COPD,stroke,Cr,LDL-C,LAVI,E/A,and GLS are the risk factors for HF in patients with CHD.It can be used to establish the prediction model of CHD concurrent HF.4.The prediction model of ACS and CCS with HF established in this study has achieved good prediction efficiency in the elderly population,and has been verified by the dual external test of temporal validation and geography validation,which has good adaptability,generalization and robustness,and has certain reference value for clinical promotion and application.