Establishment of Disease Characteristics and a Chinese Medicine Prognosis Risk Model Based on a Large-Scale Database After Coronary Revascularization

This is a multicenter, prospective cohort study conducted at five sites in China. The study aims to develop and validate a prognostic risk model for major adverse cardiovascular events (MACE) in patients after coronary revascularization (percutaneous coronary intervention or coronary artery bypass grafting).

Study population: A total of 600 eligible patients aged ≥18 years who have undergone coronary revascularization will be enrolled consecutively. Key exclusion criteria include severe heart failure, malignant arrhythmias, severe pulmonary or liver/kidney dysfunction, pregnancy, psychiatric disorders, and poor compliance.

Data collection: At baseline (enrollment), the following data are collected: demographics, medical history, surgical characteristics (e.g., access route, number of stents, target vessels), vital signs, laboratory tests (complete blood count, cardiac enzymes, liver/kidney function, lipids, glucose), echocardiography, 24-hour ambulatory electrocardiography, and a standardized Traditional Chinese Medicine (TCM) case report form covering symptom scores, tongue/pulse findings, and pattern elements. In addition, biological samples (blood and tongue coating) are obtained for proteomics, metabolomics, and tongue-coating microbiomics.

Follow-up: Participants are followed at 12, 24, 36, and 48 weeks post-enrollment. At each follow-up, the TCM case report form is reassessed, MACE (including all-cause death, subacute stent thrombosis, perioperative myocardial infarction, recurrent myocardial infarction, recurrent unstable angina, repeat revascularization, and rehospitalization for angina or heart failure) are recorded, and NYHA functional class and current medications are updated.

Statistical analysis: Missing data will be handled by mean imputation or K-nearest neighbors imputation. Continuous variables will be standardized using Z-scores, and categorical variables will be one-hot encoded. Feature selection will be performed using LASSO regression. Three nested prediction models will be built: