Table 3.

Examples of ML for the Care of Hypertension

First Author (Ref. #)	Disease Application	Sample Size	Variable Input	Output	Algorithms	Results
Huang et al (30)	Prediction of hypertension	3,054	Occupation, family history, educational level, alcohol intake, vegetable and fruit intake, salt, animal insides intake, physical exercise, body mass index, and blood pressure	Prevalent hypertension	LRM and ANN	ANN model was better than LRM in predicting the presence of hypertension
AlKaabi et al (31)	Prediction of hypertension	987	Age, sex, education, employment, tobacco use, physical activity, consumption of fruits and vegetables, mother history of hypertension, diabetes, cholesterol, and abdominal obesity	Prevalent hypertension	DT, RF and LRM	RF model had better prediction accuracy for screening the presence of hypertension
Kanegae et al (38)	Prediction of hypertension	18,258	Medical history, lifestyle factors, anthropometrics, and biochemical measurements	Incident hypertension	XGBoost, ensemble, and LRM	ML developed a highly precise model for predicting incident hypertension
Katz et al (35)	Classification of hypertension	1,273	Demographics, physical characteristics, laboratory, and echocardiographic indices	Hypertension phenotypes	Agglomerative hierarchical clustering	2 distinct types of hypertension with different cardiac substrate
Wu et al (20)	Prediction of outcome	508	Left atrial diameter, HDL-C, big endothelin-1, right arm diastolic BP, right/left leg systolic BP, right leg diastolic BP, left arm systolic BP, mean nocturnal arterial oxygen saturation, past maximum systolic BP, and urea	Clinical outcomes	Recursive feature elimination and XGBoost	ML model was comparable with Cox proportional model for outcome prediction and better than recalibrated Framingham risk score model

ANN = artificial neural network; BP = blood pressure; DT = decision tree; HDL-C = high-density lipoprotein–cholesterol; LRM = logistic regression model; ML = machine learning; RF = random forest.