Development and validation of explainable machine-learning models for carotid atherosclerosis early screening

Area of Science:

• Medical Informatics
• Public Health
• Cardiovascular Disease Research

Background:

• Carotid atherosclerosis (CAS) is a significant risk factor for stroke and a major public health concern.
• Early detection of CAS is crucial for stroke prevention strategies.
• Routine health check-up indicators offer a potential resource for widespread CAS screening.

Purpose of the Study:

• To develop and validate machine learning (ML) models for early screening of CAS.
• To utilize routine health check-up data for CAS identification.
• To assess the performance and interpretability of ML models in CAS screening.

Main Methods:

• Utilized 69,601 health check-up records from northeast China (2018-2019).
• Trained and tested ten ML algorithms, including Gradient Boosting Decision Tree (GBDT), on the dataset.
• Employed area under the receiver operating characteristic curve (auROC) and SHapley Additive exPlanations (SHAP) for performance evaluation and interpretability.

Main Results:

• The GBDT model demonstrated high performance, with an auROC of 0.860 (internal) and 0.851 (external validation).
• Key predictive factors identified by SHAP analysis include age, sex, and non-high-density lipoprotein cholesterol.
• Lower negative predictive values were observed in individuals with diabetes or those over 65 years.

Conclusions:

• Developed ML models show promise for effective CAS screening using routine health indicators.
• The models can potentially be applied in diverse populations for CAS prevention.
• Interpretability analysis highlights critical demographic and clinical factors for CAS risk.