Development and Validation of a Predictive Model Using Logistic Regression and Machine Learning for Carotid Artery...

Area of Science:

• Cardiovascular Disease Research
• Genetics and Genomics
• Epidemiology

Background:

• Postmenopausal women face increased carotid atherosclerosis risk due to estrogen decline.
• Current risk models lack genetic markers, necessitating improved prediction tools.
• Estrogen receptor 1 (ESR1) rs9340799 genotype is investigated as a potential genetic marker.

Purpose of the Study:

• To develop and validate a risk prediction model for carotid atherosclerosis (CAS) in postmenopausal Han women.
• To integrate clinical variables with the ESR1 rs9340799 genotype for enhanced prediction.
• To compare logistic regression with machine learning approaches for model development.

Main Methods:

• Recruited 276 postmenopausal Han women, categorizing them into CAS cases (n=162) and controls (n=114).
• Genotyped the ESR1 rs9340799 polymorphism using high-resolution melting PCR.
• Developed a prediction model using logistic regression, incorporating age, systolic blood pressure (SBP), glucose (GLU), and rs9340799, and validated it externally.

Main Results:

• Age, SBP, GLU, and ESR1 rs9340799 genotype were significant predictors of CAS.
• Logistic regression yielded the highest Area Under the Curve (AUC) of 0.709, demonstrating moderate predictive performance.
• External validation showed an AUC of 0.65, with sensitivity of 0.82 and specificity of 0.55, indicating a need for larger sample sizes.

Conclusions:

• A preliminary logistic regression model integrating clinical factors and ESR1 rs9340799 genotype was developed.
• The model shows moderate discrimination for predicting CAS risk in the target population.
• Further validation in larger, independent, multicenter cohorts is essential for clinical application.