# A study on the correlation between TCM syndrome types, TCM symptoms and myocardial injury markers in patients with coronary heart disease

## TL;DR

This study explores how traditional Chinese medicine (TCM) syndrome types in coronary heart disease patients correlate with symptoms and heart injury markers, using machine learning to predict syndrome types.

## Contribution

The study introduces a machine learning approach to predict TCM syndrome types based on symptoms and myocardial injury markers.

## Key findings

- Phlegm blocking the heart meridian syndrome was the most common (20.5%) among CHD patients.
- BNP and NT-proBNP levels varied significantly across different TCM syndrome types.
- The LGBM machine learning model achieved 72.51% accuracy in predicting TCM syndrome types.

## Abstract

Coronary heart disease (CHD) is the leading cause of death from cardiovascular diseases. Previous studies related to traditional Chinese medicine (TCM) mostly lacked objective basis for TCM syndrome types, which may affect the accuracy of syndrome differentiation.

To explore the distribution pattern of TCM syndrome types in patients with CHD, analyze its correlation with TCM symptoms and myocardial injury markers.

This study adopted a cross-sectional research design. A total of 1,503 patients with CHD from January 2023 to January 2025 were included. The clinical and TCM four diagnostic data of the selected patients were systematically collected, aiming to analyze the association between TCM syndrome types and TCM symptoms as well as myocardial injury markers, and to construct and verify the predictive model of TCM syndrome types.

A total of 1435 patients were included and grouped according to the syndrome types in TCM. The main types were 295 cases (20.5%) of phlegm blocking the heart meridian syndrome, 289 cases (20.1%) of qi and Yin deficiency syndrome, 257 cases (17.9%) of heart and kidney Yin deficiency syndrome, 199 cases (13.8%) of qi deficiency and blood stasis syndrome, 177 cases (12.3%) of qi stagnation and blood stasis syndrome, and 186 cases (12.9%) of heart blood stasis obstruction syndrome. BNP and NT-proBNP were statistically significant differences among different TCM syndrome types (P < 0.05). Further, unordered multi-class regression analysis was used to compare the differences in TCM symptom indicators among different TCM syndrome types. Meaningful statistical results and TCM symptoms were combined for discriminant analysis of TCM syndrome types (coincidence rate of discrimination = 62.8%), and machine learning models (LGBM, XGBoost, etc.) were used to construct TCM syndrome type prediction models. Ultimately, the best-performing model LGBM (validation set accuracy = 72.51%) was selected and SHAP was used to explain the contribution of the model.

The combination of TCM symptoms and myocardial injury markers can be used to distinguish and predict the syndrome types of patients.