Machine Learning Model Shows Moderate Accuracy for Complex Coronary Lesions in Diabetes

A retrospective study of 846 patients diagnosed concurrently with type 2 diabetes mellitus and coronary heart disease found that an interpretable gradient boosting machine incorporating routine clinical and laboratory markers achieved moderate performance in identifying complex coronary lesions, but failed to demonstrate statistical superiority over simpler classification models. Noninvasive preoperative risk stratification for these lesions has remained difficult in people living with both conditions, prompting researchers to evaluate machine learning frameworks integrating left ventricular ejection fraction and the systemic immune-inflammation index alongside standard clinical indicators.

Across the complete cohort of 846 individuals examined in the single-center analysis, 475 patients, or 56.1 percent, were classified as having complex coronary lesions. To develop and evaluate the models, researchers separated the clinical records into a training set of 593 individuals and an internal validation set of 253 individuals using a seven-to-three data split. Predictor selection using least absolute shrinkage and selection operator regression followed by multivariable logistic screening identified seven key variables: left ventricular ejection fraction, log-transformed systemic immune-inflammation index, triglyceride-glucose index, high-sensitivity C-reactive protein, log-transformed N-terminal pro-B-type natriuretic peptide, aspartate aminotransferase, and hypertension.

Researchers constructed seven distinct classifiers and carried out a separately seeded internal sensitivity analysis conditional on the completed data matrix. This sensitivity analysis evaluated paired discrimination, model calibration, probability-scale Shapley additive explanations, nested feature selection, and fixed left ventricular ejection fraction reference breakpoints positioned at 40 percent and 50 percent. In initial internal validation testing, the gradient boosting machine achieved an area under the receiver operating characteristic curve of 0.743, with a 95 percent confidence interval spanning from 0.683 to 0.804.

In a supplementary split assessment, the gradient boosting machine recorded an area under the curve of 0.757 and a Brier score of 0.199. However, the model showed no statistically significant differences in discrimination compared with standard multivariable logistic regression, which yielded a P value of 0.644, or with AdaBoost, which yielded a P value of 0.280. Probability-scale Shapley additive explanations indicated that left ventricular ejection fraction, high-sensitivity C-reactive protein, and the log-transformed systemic immune-inflammation index provided the largest mean absolute contributions to the model’s predictions.

Further multivariable analysis identified contrasting associations between individual markers and lesion complexity. After full multivariable adjustment, left ventricular ejection fraction remained significantly and inversely associated with complex coronary lesions, showing an odds ratio of 0.92 per percentage point with a 95 percent confidence interval of 0.89 to 0.95. Conversely, the association between the log-transformed systemic immune-inflammation index and lesion complexity attenuated after multivariable adjustment, producing an odds ratio of 1.35 with a 95 percent confidence interval of 0.91 to 2.00.

Despite achieving moderate discrimination from routine clinical and laboratory variables, the gradient boosting model did not establish superiority over simpler statistical approaches, and the analysis could not define a clinically actionable left ventricular ejection fraction threshold. In addition, the authors reported that the investigation’s single-center retrospective structure and the absence of external validation currently limit the clinical application of the predictive algorithm.

Subject of Research: Machine learning prediction of complex coronary lesions in patients with type 2 diabetes and coronary heart disease

Article Title: Development and internal validation of an interpretable machine learning model incorporating LVEF and SII for Predicting complex coronary lesions in patients with type 2 diabetes mellitus and coronary heart disease

Article References: Ma, X., Wang, M., Zhang, J., Yang, S., & Gou, L. (2026). Development and internal validation of an interpretable machine learning model incorporating LVEF and SII for Predicting complex coronary lesions in patients with type 2 diabetes mellitus and coronary heart disease. BMC Cardiovascular Disorders. https://doi.org/10.1186/s12872-026-06907-1

Image Credits: AI Generated

DOI: 10.1186/s12872-026-06907-1

Keywords: cardiology, coronary heart disease, type 2 diabetes, machine learning, gradient boosting machine, left ventricular ejection fraction, systemic immune-inflammation index, complex coronary lesions, Development, internal, validation, interpretable