Korean J Fam Pract.  2022 Jun;12(3):173-178. 10.21215/kjfp.2022.12.3.173.

A Machine-Learning-Based Risk Factor Analysis for Hypertension: Korea National Health and Nutrition Examination Survey 2016–2019

• Oh T
• Kim D
• Won C
• Kim S
• Jeong E
• Yang J
• Yu J
• Kim B
• Lee J

• KMID: 2530659
• DOI: http://doi.org/10.21215/kjfp.2022.12.3.173

Abstract

Background
The purpose of this study was to use machine learning to identify risk factors (other than systolic and diastolic blood pressure) for hypertension.
Methods
The study population comprised 23,170 adults (selected from the KNHANES 2016–2019), of whom 7,500 (32.4%) had hypertension. We developed machine learning-based classification models for diagnosing hypertension using the computerized demographic and examination survey database of subjects from the KNHANES study. Random forest (RF)- and gradient boosting machine (GBM)-based classification algorithms were trained with 5-fold cross-validation, and factors related to hypertension were identified through post-hoc analysis using the permutation feature importance (PFI) technique. The classifiers used 59 variables whose data could be easily extracted on medical examination, excluding directly related variables like systolic and diastolic blood pressure.
Results
The classification performance of GBM (area under the curve [AUC], 0.852; 95% confidence interval [CI], 0.842–0.862) was slightly higher than that of RF (AUC, 0.847; 95% CI, 0.837–0.857). Post-hoc analysis of model classification using the PFI technique revealed age, cholesterol level, fraternal hypertension, education level, and height as risk factors for hypertension.
Conclusion
Although hypertension diagnosis is based on systolic and diastolic blood pressure measurements, hypertension could also be diagnosed by analyzing easily extractable variables such as age, cholesterol level, and family history of hypertension using machine learning.

Keyword

Hypertension; Machine Learning; Risk Factors; Data Mining; Artificial Intelligence