Machine learning model flags unstable carotid plaques from routine clinical data

Researchers have built machine learning models that predict whether a carotid plaque is likely to become unstable, using only routinely collected clinical data from patients at a single hospital in China. The work, published as a peer-reviewed research article in BMC Cardiovascular Disorders, reports that a model based on the light gradient boosting machine (LightGBM) algorithm delivered the strongest performance among four machine learning approaches tested.

The study was a retrospective analysis of data from 449 patients with carotid plaques treated at The Affiliated Hospital of Qingdao University between July 2022 and November 2024. The patients were randomly divided into a training set and a testing set at a 7:3 ratio, meaning roughly seven in ten records were used to teach the models and the remaining three in ten were held back to test how well they generalized to unseen cases.

Four machine learning algorithms were used to establish prediction models. The researchers then evaluated each model’s predictive performance using the area under the receiver operating characteristic curve (AUC), calibration curves, and decision curve analysis (DCA), a combination that captures both discrimination and clinical usefulness.

LightGBM came out on top. In the training set it achieved an AUC of 0.921 (95% confidence interval: 0.887–0.956) and an accuracy of 0.837. In the testing set it reached an AUC of 0.864 (95% CI: 0.784–0.944) with an accuracy of 0.811. The somewhat lower testing-set figures reflect performance on data the model had never seen during training, and the confidence intervals indicate the uncertainty around those point estimates.

To understand what drove the predictions, the team applied SHAP (SHapley Additive exPlanations), a technique that quantifies each feature’s contribution to a model’s output. This analysis identified C-reactive protein (CRP), leukocytes, high-density lipoprotein (HDL), platelets, neutrophils, the lymphocyte-to-monocyte ratio (LMR), fibrinogen, low-density lipoprotein (LDL), and monocytes as the key contributors to the model’s predictive performance. Many of these are markers of inflammation and lipid metabolism, measured in blood tests that are already part of standard clinical workups.

Beyond discrimination, the calibration curves and decision curve analysis both supported the favorable predictive performance of the models, according to the authors.

The authors conclude that machine learning–based predictive models offer a valuable approach for evaluating the risk of carotid plaque instability by utilizing routinely collected clinical data, and that such models hold potential as decision-support tools to guide personalized treatment planning. Because the inputs are ordinary laboratory values rather than specialized imaging, the approach could in principle be integrated into existing hospital workflows.

Several caveats are worth noting. The study was retrospective and drew on patients from a single center over a roughly two-and-a-half-year window, so performance at other hospitals and in other populations remains to be established. The reported AUCs and accuracies describe how well the models ranked and classified patients in this dataset; they do not by themselves demonstrate that using the models changes clinical decisions or patient outcomes. The article was shared early as a citable, peer-reviewed accepted version carrying a permanent DOI, subject to further edits before the final Version of Record.

By pairing strong discrimination with SHAP-based explanations of which blood markers matter most, the Qingdao team’s LightGBM model offers a transparent starting point for clinicians weighing which patients with carotid plaques may need closer monitoring or more aggressive intervention.

Subject of Research: Machine learning prediction of carotid plaque instability from routine clinical data

Article Title: A study on analyzing risk factors for carotid plaque instability using machine learning models

Article References: Zhang, T., Ma, H., Sun, X., Li, G., Li, Y., & Guo, M. (2026). A study on analyzing risk factors for carotid plaque instability using machine learning models. BMC Cardiovascular Disorders. https://doi.org/10.1186/s12872-026-06931-1

Image Credits: AI Generated

DOI: 10.1186/s12872-026-06931-1

Keywords: carotid plaque, machine learning, LightGBM, SHAP, AUC, retrospective study, C-reactive protein, cardiovascular risk, BMC Cardiovascular Disorders, predictive modeling, decision curve analysis, Qingdao University