Coronary health index based on immunoglobulin light chains to assess coronary heart disease risk with machine learning: a diagnostic trial

Figures

• 

  Fig. 1. Details of the sample inclusion criteria and grouping procedure. After initial screening, the 486 included samples were divided into control and CHD groups. The CHD group was classified into mild, moderate, and severe lesion groups via the SYNTAX scoring system

• 

  Fig. 2. CHD patients show distinct features and lower CHI than controls. Comparison of the classical diagnostic indicators as well as the novel CHI between the control group and the CHD group

• 

  Fig. 3. Logic regression is used for feature selection. Results of the univariate regression analysis used to screen for baseline characteristics significantly associated with CHD

• 

  Fig. 4. Screening of machine learning classification models. (A) The forest plot displays the results of ROC curve analysis for the CHD prediction models, with the data indicating the means (dots) and SDs (error bars) of the model AUCs. Through 5-fold cross-validation, XGBoost emerged as the top performer in the training set in terms of AUC. (B) Plots of the model ROC curves in the training set. (C) Plots of the model ROC curves in the validation set. The AUC value of the XGBoost classification model was the highest. (D) The DeLong test results indicate that the area under the ROC curve of the XGBoost model is significantly different from the AUCs of the other models

• 

  Fig. 5. Performance evaluation of the XGBoost classification model. (A) 10-fold cross-validation ROC curves in the training set. (B) 10-fold cross-validation ROC curves in the validation set. (C) ROC curve in the test set. (D) The learning curve shows that model performance in both the training set and the validation set improves steadily and that the difference between them remains stable, showing that the model neither overfits nor underfits the data and is able to effectively learn from the training data and generalize to the validation data. (E) The decision curve highlights the model’s excellent performance across multiple thresholds, indicating that it could help guide clinical decision-making. (F) The calibration curve shows close alignment between the predicted and actual probabilities, demonstrating the accuracy of the model’s predictive performance

• 

  Fig. 6. CHI decreases with severity of CHD patients. (A) As the severity of CHD increases, the CHI decreases, as depicted in a set of box plots. (B) The forest plot displays the ROC curve results for the CHD prediction models, with dots and error bars indicating the means and SDs of the AUC. Through 2-fold cross-validation, XGBoost emerged as the top performer in the training set according to the AUC

• 

  Fig. 7. XGBoost model identifies patients with severe CHD. (A) 10-fold cross-validation ROC curves in the training set. (B) 10-fold cross-validation ROC curves in the validation set. (C) ROC curve in the test set. (D) The learning curves exhibit similar trends between the training and validation sets with small differences, indicating strong generalization capabilities and near-optimal performance. (E) Decision curve analysis reveals that the model outperforms alternative treatment strategies and is clinically effective. (F) The calibration curve reveals the model’s overconfidence in predicting severe CHD

• 

  Fig. 8. Development of online prediction tool. The visualization of the prediction model through Roche i-Research platform