Sex-specific machine learning models for carotid plaque prediction in individuals with fatty liver disease: a cross-sectional study.
Early detection of carotid plaque prevents stroke and myocardial infarction. Individuals with fatty liver might be at an increased risk of developing carotid plaque, yet limited access to carotid artery ultrasound underscores the need for predictive models.
We aimed to construct six predictive models for males and females separately to predict carotid plaque among individuals with fatty liver disease.
A cross-sectional study.
We included 8361 participants aged ≥40 years (4871 males; 3490 females) with fatty liver who underwent at least one health check-up between 1 January 2020 and 31 December 2023.
The sex-stratified dataset was randomly divided into 70% training and 30% internal testing datasets. With 24 potential predictors, we applied four machine learning (ML) algorithms and two conventional logistic regression (LR) models: stepwise LR and LR based on ML-selected features (LR-ML) to develop sex-specific carotid plaque prediction models. The performances were evaluated by area under the receiver operating characteristic curve (AUC), sensitivity, specificity, precision, F1-score, accuracy, calibration curve and decision curve analysis.
Carotid plaque was determined when the local carotid intima-media thickness was ≥1.5 mm in any of the arterial segments.
Four predictors (age, hypertension, total bilirubin, total cholesterol and white blood cell count) in males and three (age, systolic blood pressure and fasting blood glucose) in females were identified by consensus across the four ML algorithms and subsequently used to construct LR models. Among all 4 ML and two LR models, the gradient boosting machine model demonstrated the best overall performance in males (AUC=0.773, 95% CI 0.749 to 0.797), while the LR-ML model was optimal in females (AUC=0.817, 95% CI 0.791 to 0.843). Calibration and decision curve analyses further demonstrated satisfactory agreement and higher net benefit across sexes. Risk stratification identified distinct low-, intermediate- and high-risk groups with progressively higher observed prevalence of carotid plaque (20.25%, 48.58% and 69.41% in males; 15.28%, 50.89% and 66.56% in females).
Our findings highlight significant sex differences in practical carotid plaque prediction, providing crucial insights for public health implications in the early identification and risk assessment of carotid plaque among individuals with fatty liver.
We aimed to construct six predictive models for males and females separately to predict carotid plaque among individuals with fatty liver disease.
A cross-sectional study.
We included 8361 participants aged ≥40 years (4871 males; 3490 females) with fatty liver who underwent at least one health check-up between 1 January 2020 and 31 December 2023.
The sex-stratified dataset was randomly divided into 70% training and 30% internal testing datasets. With 24 potential predictors, we applied four machine learning (ML) algorithms and two conventional logistic regression (LR) models: stepwise LR and LR based on ML-selected features (LR-ML) to develop sex-specific carotid plaque prediction models. The performances were evaluated by area under the receiver operating characteristic curve (AUC), sensitivity, specificity, precision, F1-score, accuracy, calibration curve and decision curve analysis.
Carotid plaque was determined when the local carotid intima-media thickness was ≥1.5 mm in any of the arterial segments.
Four predictors (age, hypertension, total bilirubin, total cholesterol and white blood cell count) in males and three (age, systolic blood pressure and fasting blood glucose) in females were identified by consensus across the four ML algorithms and subsequently used to construct LR models. Among all 4 ML and two LR models, the gradient boosting machine model demonstrated the best overall performance in males (AUC=0.773, 95% CI 0.749 to 0.797), while the LR-ML model was optimal in females (AUC=0.817, 95% CI 0.791 to 0.843). Calibration and decision curve analyses further demonstrated satisfactory agreement and higher net benefit across sexes. Risk stratification identified distinct low-, intermediate- and high-risk groups with progressively higher observed prevalence of carotid plaque (20.25%, 48.58% and 69.41% in males; 15.28%, 50.89% and 66.56% in females).
Our findings highlight significant sex differences in practical carotid plaque prediction, providing crucial insights for public health implications in the early identification and risk assessment of carotid plaque among individuals with fatty liver.