Predicting increased blood pressure using machine learning.

Area of Science:

• Cardiology
• Biostatistics
• Machine Learning

Background:

• Increased blood pressure is a significant health concern.
• Anthropometric measurements like Body Mass Index (BMI), waist circumference (WC), hip circumference (HC), and waist-hip ratio (WHR) are potential indicators of cardiovascular risk.
• Predictive modeling can aid in early identification and management of hypertension.

Purpose of the Study:

• To investigate the predictive power of BMI, WC, HC, and WHR for increased blood pressure using classification trees.
• To compare the performance of classification trees against traditional logistic regression for hypertension prediction.
• To identify optimal combinations of anthropometric predictors for hypertension risk assessment in college students.

Main Methods:

• A dataset of 400 college students (ages 16-63) was utilized.
• Classification tree models were developed for both sexes, with multiple trees generated using varying predictor combinations.
• Model performance was evaluated using metrics such as deviance, misclassification rate, pseudo R-squared, sensitivity, and specificity on training and test sets.

Main Results:

• For women, the combination of BMI, WC, and WHR yielded the best prediction model (deviance: 87.42, misclassification: 0.19, pseudo R-squared: 0.43).
• For men, BMI, WC, HC, and WHC provided the optimal prediction (deviance: 57.25, misclassification: 0.16, pseudo R-squared: 0.46).
• Classification trees demonstrated superior predictive power compared to logistic regression.

Conclusions:

• Anthropometric measures, particularly BMI, WC, and WHR, are effective predictors of increased blood pressure in college students.
• Classification tree analysis offers a robust and more accurate method for hypertension risk prediction than logistic regression.
• Sex-specific models utilizing distinct combinations of anthropometric variables enhance prediction accuracy for hypertension.