JMIR Publications

ABSTRACT

Background:

Globally, cardiovascular diseases (CVDs) are the primary cause of death, with hypertension as a key contributor. In 2019, CVD led to 17.9 million deaths, predicted to reach 23 million by 2030.

Objective:

This study presents a new method to predict hypertension using demographic data, employing six machine learning models for enhanced reliability and applicability. The goal is to harness AI for early and accurate hypertension diagnosis across diverse populations.

Methods:

Data from two national cohort studies, NHIS-NSC (South Korea, n=244,814), conducted between 2002 and 2013 were utilized to train and test machine learning models designed to anticipate incident hypertension within five years of a health checkup involving ≥20 years of age, and JMDC (Japan, n=1,296,649) were utilized for extra validation. An ensemble from six diverse machine learning models was employed to identify the five most salient features contributing to hypertension by presenting a feature importance analysis to confirm the contribution of each future.

Results:

The AdaBoost and logistic regression ensemble showed superior balanced accuracy (0.812; sensitivity, 0.806; specificity, 0.818; area under the receiver operating characteristic curve [AUROC], 0.901). The five key hypertension indicators were age, diastolic blood pressure, body mass index, systolic blood pressure, and fasting blood glucose. The JMDC dataset (extra-validation set) corroborated these findings (balanced accuracy, 0.741; AUROC, 0.824). The ensemble model was integrated into a public web portal (http://ai-wm.khu.ac.kr/Hypertension/) for predicting hypertension onset based on health checkup data.

Conclusions:

Comparative evaluation of our machine learning models against classical statistical models across two distinct studies emphasized the former's enhanced stability, generalizability, and reproducibility in predicting hypertension onset.