What type of study is this?

This is a Cox regression study (also classified as: Observational).

September 10, 2025Open Access

The Relationship Between C‐Reactive Protein–Triacylglycerol–Glucose Index and All‐Cause Mortality in Patients With Cardiovascular Disease: A Retrospective Cohort Study and Development of a Machine Learning Prediction Model

Key Points

The highest quartile of CTI significantly increases the risk of all-cause mortality in CVD patients, highlighting its potential as a vital marker.
Cox regression analysis revealed a hazard ratio of 1.38 for Q4 CTI versus Q1, indicating a strong connection to mortality risk.
Machine learning models, particularly LightGBM, achieved high accuracy in predicting mortality risk based on CTI, underscoring its utility in clinical settings.
SHAP analysis identified age, BUN, and CTI as critical factors, providing insights for future research on mortality risk in cardiovascular disease.

Abstract

Objective: The CTI is increasingly recognized as a new marker for assessing inflammation and insulin resistance. However, the relationship between CTI and all‐cause mortality risk in patients with CVD remains unclear. Methods: We analyzed data from the NHANES from 1999 to 2010. The correlation between CTI and all‐cause mortality risk in CVD patients was examined using Cox regression analysis. Nonlinear relationships between CTI and all‐cause mortality risk were explored through restricted cubic splines and Cox proportional hazards regression. We employed six ML models, including RF, LightGBM, DT, XGBoost, LR, and KNN, to predict all‐cause mortality risk in CVD patients based on CTI and SHAP for interpretability. Results: A total of 1429 CVD patients were included, with 849 all‐cause deaths recorded during the follow‐up period. After adjusting for potential confounding factors, the highest quartile of CTI (Q4) significantly increased the risk of all‐cause mortality compared to the lowest quartile (Q1) (HR = 1.38, 95% CI: 1.04–1.84, p = 0.03). Restricted cubic splines demonstrated a nonlinear relationship between CTI and all‐cause mortality risk in CVD patients. Among the machine learning models, the LightGBM model exhibited the best predictive performance, with an ROC of 0.967, accuracy of 0.909, sensitivity of 0.906, specificity of 0.914, F 1 score of 0.922, recall of 0.906, and PR of 0.979. SHAP analysis identified age, BU, and CTI as the primary predictive factors, with CTI positively correlated with all‐cause mortality risk in CVD patients. Conclusion: There is a nonlinear relationship between CTI and all‐cause mortality risk in CVD patients, with elevated CTI levels significantly associated with increased mortality risk. Additionally, for the first time, this study constructed a machine learning model to predict all‐cause mortality risk in cardiovascular disease using CTI, with LightGBM demonstrating the best predictive performance. SHAP analysis identified age, BUN, and CTI as critical factors in the prediction, providing valuable references for future related research.

Read Full Paperexternally

AIに質問

Bookmark

View Full Paper