Artificial Intelligence and Hypertension: Recent Advances and Future Outlook

Abstract

Prevention and treatment of hypertension (HTN) are a challenging public health problem. Recent evidence suggests that artificial intelligence (AI) has potential to be a promising tool for reducing the global burden of HTN, and furthering precision medicine related to cardiovascular (CV) diseases including HTN. Since AI can stimulate human thought processes and learning with complex algorithms and advanced computational power, AI can be applied to multimodal and big data, including genetics, epigenetics, proteomics, metabolomics, CV imaging, socioeconomic, behavioral, and environmental factors. AI demonstrates the ability to identify risk factors and phenotypes of HTN, predict the risk of incident HTN, diagnose HTN, estimate blood pressure (BP), develop novel cuffless methods for BP measurement, and comprehensively identify factors associated with treatment adherence and success. Moreover, AI has also been used to analyze data from major randomized controlled trials exploring different BP targets to uncover previously undescribed factors associated with CV outcomes. Therefore, AI-integrated HTN care has the potential to transform clinical practice by incorporating personalized prevention and treatment approaches, such as determining optimal and patient-specific BP goals, identifying the most effective antihypertensive medication regimen for an individual, and developing interventions targeting modifiable risk factors. Although the role of AI in HTN has been increasingly recognized over the past decade, it remains in its infancy, and future studies with big data analysis and N-of-1 study design are needed to further demonstrate the applicability of AI in HTN prevention and treatment.

Hypertension (HTN) is the leading cause of cardiovascular disease (CVD) morbidity and mortality worldwide. Over the past decades, the prevalence of patients with HTN has increased from 594 million in 1975 to 1.13 billion in 2015, especially in low- and middle-income countries, and is mostly driven by aging, changes in lifestyle, and population growth.¹ Recent HTN guidelines from American College of Cardiology (ACC)/American Heart Association (AHA)² lowered the threshold for diagnosing stage 1 HTN from an office-measured 140/90 to 130/80 mm Hg based on results from recent randomized controlled trials and meta-analyses, resulting in significantly increased prevalence and incidence of HTN,^3,4 again emphasizing the global burden of HTN.

Many challenges present themselves in the management of HTN. Despite straightforward strategies for blood pressure (BP) control suggested by clinical practice guidelines, the pathophysiology and pathogenesis of HTN are heterogeneous and complex, involving interactions among genetics, epigenetics, social determinants, environmental, and behavioral factors,² which support the hypothesis of different phenotypes and clusters of HTN patients, and explain different treatment response rates with only 14% of patients globally achieving a systolic blood pressure (SBP) of less than 140 mm Hg.⁵ Moreover, the heterogeneity of patients with HTN may be responsible for conflicting results from different BP targets demonstrated by different major randomized controlled trials,^6,7 which lead to different recommendations from major HTN guidelines.^2,8 Hence, accumulating evidence suggests that a personalized treatment approach, often referred to as precision medicine, is necessary to improve patient outcomes and quality of life.⁹

The potential role of artificial intelligence (AI) in cardiovascular (CV) medicine and HTN has been increasingly recognized.^10–12 AI is the science and engineering of developing machines that are capable of stimulating human thought processes and learning based on applying complex algorithms and advanced computational power to extremely large amounts of data, which can also be referred to as “big data.” ¹⁰ Emerging digital technologies, such as social media, mobile applications, and wearable devices, that can provide substantial amount of continuous and real-time health data, shed light on the possibility of using AI and big data analytics to identify hidden risk factors or phenotypes of HTN that could not be described or uncovered by conventional demographics, statistics, study designs, and study populations. Additionally AI may aid in developing precise risk prediction models for patients with HTN by integrating traditional CV risk factors with multiomics (e.g., genomics, proteomics, and metabolomics), socioeconomic, behavioral, and environmental factors, as well as developing personalized treatment approaches for patients with HTN. In this review, we discuss the potential of AI in the prevention, diagnosis, and management of HTN.

MACHINE LEARNING AND DEEP LEARNING: BASIC CONCEPTS

Machine learning (ML) and deep learning (DL) are 2 common subtypes of AI that are promising and capable of solving problems with different approaches and strategies. While ML research in precision CV medicine has grown in recent years, DL is newer, and more sophisticated with different advantages and limitations compared with ML.

Machine learning

ML is used to perform predictive analyses by investigating mechanisms and associations among given variables from training datasets, which may consist of a variety and large amounts of data inputs, such as standardized electronic health records (EHRs), multiomics, wearable devices, and social media.¹³ ML can be classified into supervised and unsupervised learning. Supervised learning is used for predicting known outputs from a labeled dataset, hypothesis, and appropriate algorithm, such as artificial neural network (ANN), support vector machine, and K-nearest neighbor, which must be selected based on characteristics of the dataset, number of variables, learning curve, training, and computation time.^10,14 Predicted outputs for supervised learning can be labeled categories (classification), values (regression), and patterns (anomaly detection).¹⁰ Although supervised learning is promising in providing accurate prediction from big data analytics, there are limitations that need to be considered, such as the requirement for manually labeled datasets, and biases that can arise from the dataset itself or the algorithm.

Alternatively, unsupervised learning analyzes unlabeled datasets to predict and compare unknown outputs, such as previously unidentified clusters or patterns within the dataset, which can be subsequently incorporated into supervised learning.¹¹ The algorithms used in unsupervised learning are broadly classified into clustering algorithm, and association rule-learning algorithms.¹⁰ Given a dataset of patients with the same diagnosis, clustering algorithm is able to stratify patients into clusters in which they have some common characteristics, thereby identifying hidden phenotypes of the disease from the dataset, whereas association rule-learning algorithms can determine the interactions among different variables within the dataset.^10,11 However, despite its capability of using unlabeled datasets, unsupervised learning still requires denoised datasets in order to identify relevant clusters or associations. Another major limitation is the generalizability of cluster patterns identified from a cohort of patients, which can be overfitting to the training dataset, and need to be validated in different large datasets.

Deep learning

DL is another type of AI that resembles the human brain by utilizing multiple layers of neural networks, which has been widely used for pattern recognition, especially voice and image analysis, such as input from CV imaging (cardiac computerized tomography, magnetic resonance imaging, and echocardiography), and electrocardiograms.¹⁵ DL can perform automated predictive analysis with both supervised and unsupervised algorithms, including deep, recurrent, and convolutional neural networks.¹⁰ Given an increase in availability of EHR databases, CV imaging, and data from emerging wearable technologies, the ability of DL to analyze large amounts of data, and predict adverse outcomes or hidden phenotypes is promising for the development of precision CV medicine. However, using DL in medicine does present significant challenges. First, DL is widely known for its “black box” methodology, which means that the results from DL may not be easily explained because of the complexity and lack of standardization in DL design.¹⁵ Second, using multiple layers of neuronal networks, nonlinear analysis, and a variety of variables within the dataset may lead to overfitting, resulting in reduced generalizability of the results, which can be improved by providing sufficient amount of training datasets, optimizing the number of hidden layers, and dropout techniques.^12,16 Lastly, DL requires sophisticated machines with extreme computational power, which are not widely available.

ML and DL can solve problems with different approaches and inputs. Implementing these AI subtypes, particularly given accelerating growth in digital technology, including social media, and wearable devices, is necessary for the further development of precision HTN prevention and treatment.

CURRENT AI RESEARCH IN HTN

Accumulating evidence demonstrates the ability of AI to measure BP in novel ways, as well as predict, diagnose, and refine treatment paradigms for patients with HTN (Table 1). AI is being used to gain more insight into the complexity of the pathogenesis and pathophysiology of HTN. Additionally, AI is also an important tool for the development of personalized treatment approaches for patients with HTN.

Table 1.

Potential artificial intelligence applications for the prevention and management of hypertension

	AI applications
Predicting development of hypertension	- Predict the risk of developing HTN by using medical data,17–20 treadmill stress test,21 behavioral, environmental, socioeconomic factors, and genetics.22–25
	- Identify new genes associated with HTN.26
Diagnosing hypertension	- Accurately diagnosing HTN by using demographic data, vital signs, traditional CV risk factors, and routine laboratories in large patient cohorts.27,28
Predicting blood pressure	- Predict BP from demographic data, lifestyle (alcohol, smoking, and exercise),29 and retinal fundus images.30
Measuring Blood Pressure	- Estimate BP by analyzing PPG signal from pulse oximeter with ML algorithms31–36 and DL algorithms.37,38
	- Estimate BP from PPG signal recorded by a smartphone39 and a smartwatch.40
Predicting cardioavascular risk in hypertension	- Predict CV outcomes in HTN patients,41–44 and stratify patients based on their risk.45–47
Predicting and identifying barriers to blood pressure control	- Predict the risk of developing uncontrolled BP.48,49
	- Identify factors contributing to treatment adherence50–52 and success.53
Refining blood pressure targets	- Uncover factors associated with CV outcomes54,55 and adverse event56 in major RCTs suggesting different BP targets.

Abbreviations: AI, artificial intelligence; BP, blood pressure; CV, cardiovascular; DL, deep learning; HTN, hypertension; ML, machine learning; PPG, photoplethysmograph; RCTs, randomized, controlled trials.

Measurement of blood pressure

Since emerging wearable health devices with continuous and real-time data became widely available and more affordable,⁵⁷ AI may play an important role of the development of cuffless BP measurement, especially via photoplethysmograph (PPG), which has been used in conjunction with AI algorithms to estimate BP. PPG is commonly used in pulse oximetry for measuring oxygen saturation and pulse waveforms by analyzing the amount of light absorbed or reflected by blood vessels. This measurement is directly affected by changes in blood volume, and provides a variety of parameters that can be used for BP estimation, such as pulse transit time.⁵⁸ In 2011, Monte-Moreno³¹ demonstrated that age, weight, body mass index (BMI), and PPG signal could be used by ML to predict BP in 410 subjects. Similarly, Solà et al.³² studied the ability of ANN to estimate BP from PPG signal in patients undergoing surgery with general anesthesia, and revealed that ANN could estimate BP with less than 8 mm Hg different from invasive radial artery BP measurement. Miao et al.³³ demonstrated that ML with support vector machine algorithms was able to analyze data from electrocardiogram and PPG signal obtained from 73 subjects for estimating SBP and diastolic blood pressure (DBP) with mean error −0.001 ± 3.102 and −0.004 ± 2.199 mm Hg, respectively. In contrast, Wang et al.³⁴ revealed that ANN could accurately estimate BP from PPG signal without using electrocardiogram with mean error 4.02 ± 2.79 and 2.27 ± 1.82 mm Hg for SBP and DBP, respectively. Moreover, preliminary data from a study on 1,750 individuals also suggested that ANN could use PPG signal for predicting SBP and DBP with less than 3 mm Hg error.⁵⁹ These findings were also corroborated by subsequent studies using ML^35,36 and DL.^37,38 Interestingly, PPG signal can also be obtained from a smartphone or smartwatch. Banerjee et al.³⁹ demonstrated that ANN was able to estimate BP from PPG signal obtained from a smartphone with approximately 10% error from the measured BP values. Similarly, preliminary data from Tison et al.⁴⁰ demonstrated that step count and longitudinal heart rate variability measured by a smartwatch with PPG could be used as inputs by DL to diagnose HTN with 84.8% sensitivity and 63.6% specificity.

Although the ability to estimate BP using AI and continuous, real-time data from PPG technology is promising, conventional BP measurement using oscillometric devices are still recommended by recent HTN guidelines as a gold-standard method for BP measurement.² Future studies on AI, big data analytics, and wearable devices need to confirm the above findings, and provide convincing clinical data to support using a combination of AI and wearable device obtained data to accurately perform BP measurement, which may ultimately provide an alternative to current oscillometric technology.

Measurement of BP with AI-enhanced wearable devices theoretically would substantially improve access to 24-hour ambulatory BP monitoring, which is the gold standard for out-of-office BP measurement, and is used to detect masked and white coat HTN.² Moreover, AI also has potential to analyze inputs from wearable devices and predict central BP, which is believed to be the actual pressure affecting important target organs, including the brain, heart, and kidneys.⁶⁰ Readily available central BP data may offer insight into the pathogenesis and pathophysiology of HTN, and potentially change the diagnosis and treatment strategies of HTN.

Predicting development of HTN

Because HTN is a chronic disease that increases one’s risk of CVD, and in most cases be controlled by lifestyle modification and BP lowering medications, the ultimate goal of public health is to prevent healthy individuals from developing HTN (primordial prevention).⁶¹ By identifying traditional and novel risk factors for each individual, as well as implementing recommendations and interventions targeting modifiable risk factors, preventing HTN may be possible. Over the past decade, research consistently demonstrates the potential of AI to identify relevant factors and predict one’s risk of HTN. A study on 3,054 subjects from Huang et al.²² revealed that ANN was more precise than a model of conventional statistics, logistic regression analysis, in predicting HTN with area under the curve (AUC) of 0.9. The inputs included behavioral, environmental, and socioeconomic factors, such as educational level, dietary habits, and physical activity. Interestingly, genetic data can also be used by AI to predict HTN. Held et al.²³ demonstrated that ML tended to outperform a logistic regression model in predicting HTN from gene expression data. Similarly, Li et al.²⁶ revealed that ML with support vector machine algorithm predicted 177 new genes that could be related to HTN. Subsequent studies also demonstrated that ML with support vector machine algorithm was able to accurately predict the risk of developing HTN by integrating environmental and genetic factors.^24,25 Moreover, ML was able to predict the risk of having elevated SBP, defined as SBP of 120 mm Hg or higher,² with an accuracy of 80.86% in 225 women by using BMI, waist and hip circumference, and waist-to-hip circumference ratio.¹⁷

Recently, the prognostic implications of AI have been validated in studies using large cohorts. Maxwell et al.¹⁸ studied a cohort of 110,300 individuals, and demonstrated that DL could predict HTN from a combination of 4 physical examination findings, 26 blood routine tests, 12 urine routine tests, and 20 items from liver function testing. Ye et al.¹⁹ revealed that ML with K-nearest neighbor algorithms was able to analyze input from EHR, such as age, gender, race, and underlying diseases, and predict the risk of incident HTN with predictive accuracy of 91.7% in retrospective cohort with 823,627 subjects, and 87% prospective (validation) cohort with 680,810 subjects. Similarly, another study on 18,258 subjects demonstrated that ML could predict the risk of developing HTN with AUC of more than 0.85 by using demographic data, waist circumference, BP, 4 blood routine tests, underlying medical diagnoses, smoking status, and alcohol use.²⁰ Moreover, Sakr et al.²¹ demonstrated that ML could also use cardiorespiratory fitness data from treadmill stress testing, such as metabolic equivalents of task, as well as resting and peak BP to precisely predict the risk of incident HTN with an AUC of 0.93. Overall, these findings suggest that AI has potential to predict the risk of HTN. Broad application of tools such as the above in primary care or school/work based health screening programs, could help prevent or delay the development of HTN through early interventions. Future research needs to validate this prognostic application of AI.

Diagnosing HTN

AI is able to accurately diagnose HTN when using inputs from BP, demographic data, comorbidities, and some routine laboratories. For example, Samant et al.²⁷ demonstrated that ANN could diagnose HTN with an accuracy of 92.85% from the dataset of 981 subjects by using age, BP, pulse, and routine blood chemistry. Similarly, LaFreniere et al.²⁸ revealed that ML could use age, BP, BMI, pulse, lipid profile, history of smoking, and exercise as input to diagnose HTN in 379,027 individuals (185,371 patients and 193,656 controls) with an accuracy of 82%. In addition, AI can also predict BP from different variables with limited success. Wu et al.²⁹ attempted to use an ANN to estimate SBP from age, BMI, alcohol consumption, smoking level, and exercise in 498 healthy individuals. Of note, 50% of predicted BP were more than 10 mm Hg different from measured BP²⁹. Interestingly, Poplin et al.³⁰ demonstrated that DL could predict SBP in 284,335 subjects from retinal fundus images with mean absolute error within 11.23 mm Hg. Clearly differences of 10 mm Hg in BP are not nearly accurate enough to state that BP can be predicted via AI; future investigations need to expand and validate the ability of AI to diagnose HTN and predict BP. If studies demonstrate more accuracy in the future, AI may theoretically assist physicians in clinical practice, and potentially facilitate early detection of existing HTN.

Managing HTN

Once HTN is diagnosed, individuals should be counseled regarding other CVD risk factors, optimal BP treatment targets, possible consequences from poorly controlled HTN, and interventions that can help reduce BP, and the ultimately the risk of CVD. These interventions include lifestyle modification, out-of-office BP monitoring, and pharmacologic interventions.² During the past decade, AI has been shown to be a promising tool to identify CVD risk factors, stratify patients with HTN based on their future risk, and potentially transform the future care of HTN towards highly accessible and personalized CV medicine.^10,12,13

An important initial step in HTN management is CV risk stratification for each patient. This may be performed via different region-specific risk calculators. The 10-year risk of atherosclerotic CVD, as determined by the pooled-cohort equation, is endorsed in the most recent American BP guidelines.² AI can provide more insight into prognosis and risk stratification of patients with HTN in addition to traditional CV risk prediction models.^41–44 Additionally, AI provides more insight into the risk of having uncontrolled BP and can help delineate factors contributing to poorly controlled HTN. Early studies demonstrated that ML with fuzzy expert system algorithm was able to classify patients with HTN as low, moderate, and high HTN risk based on age, BMI, vital signs,^45,46 including history of smoking and exercise.⁴⁷ In 2014, Sun et al.⁴⁸ developed a risk prediction model from ML algorithms that could accurately predict the risk and timing of worsening BP control in 1,692 patients with an accuracy of 77.3%. The input was extracted from EHR, such as demographic data, BMI, BP, underlying comorbidities, and routine laboratory values, including serum creatinine, potassium, troponin, and brain natriuretic peptide. Subsequent study from Mohammadi et al.⁴⁹ revealed that ML could predict the risk of developing uncontrolled HTN within a period of 3 months in patients with previously controlled HTN with an AUC of 0.719 by using demographic data, vital signs, routine blood, and urine laboratories. ML has also been used to identify factors associated with medication adherence and treatment success of HTN. Li et al.⁵⁰ developed a prediction model to predict and identify factors contributing to medication nonadherence using ML, and revealed that age, race, neighborhood poverty, history of diabetes mellitus, household income, and health insurance coverage could be used to cluster HTN patients, and predict medication adherence.

Since health care and clinical practice have been transformed by emerging technologies, AI is important for improving the efficiency of novel digital interventions by identifying factors associated with treatment adherence and predicting treatment success.⁵³ Guthrie et al.⁵¹ developed a smartphone application providing behavioral therapy for HTN patients, such as goal setting, skill learning, self-monitoring of biometrics, and personalized feedback, which could significantly reduce BP in a cohort of 172 patients. They demonstrated that ML was able to predict intervention completion at day 7 with an AUC of 0.78, and predictive factors included self-monitoring of biometrics, completing application interactions, an increase in the amount of exercise, and brand of smartphone. Recently, da Silva et al.⁵² developed a comprehensive digital intervention using ML to determine adherence to antihypertensive medications by analyzing data recorded from a smartphone application in order to promote patients’ awareness, self-monitoring, and treatment compliance, which is an example of personalized and holistic treatment approach for patients with HTN. There are multiple ongoing clinical trials assessing the effectiveness of AI-integrated HTN care on treatment compliance, optimization, and success (i.e., NCT04223934, NCT03969056).

With additional study and implementation, AI will become a tool for the development of comprehensive and precise risk prediction models integrating traditional CV risk, behavioral, and environmental factors. AI also has potential to predict treatment failure, and identify factors contributing to uncontrolled BP and treatment adherence in each patient. Furthermore, AI can be used to develop intelligent health care systems that provides personalized recommendations and treatment strategies.

Defining optimal blood pressure targets

A major challenge in the management of HTN is to define optimal BP targets as evidenced by recent American and European HTN guidelines.⁶² 2017 ACC/AHA HTN guidelines of HTN recommend BP target of 130/80 mm Hg for most of HTN patients,² whereas current HTN guidelines from European Society of Cardiology (ESC)/European Society of Hypertension (ESH)⁸ still recommend BP target of 140/90 mm Hg, and only 130/80 mm Hg for patients with high CV risk. Most of the push toward tighter BP control was driven by the results of the Systolic Blood Pressure Intervention Trial (SPRINT)⁶ that demonstrated clear benefit in aiming for a SBP goal of less than 120 vs. 135–140 mm Hg in a patient population with, or at elevated risk for, CV disease, but without diabetes mellitus. This stands in contrast to results from the Action to Control Cardiovascular Risk in Diabetes Blood Pressure Trial (ACCORD-BP)⁷ which studied similar BP targets, but in patients with diabetes mellitus, and showed no benefit from tighter BP control. Although attempts have been made to assess differences between study populations and study design in these 2 trials,^63,64 AI and big data analytics may provide more insight into additional factors contributing to heterogeneous treatment effects from intensive BP control. Data from SPRINT were reanalyzed by Lacson et al.⁵⁴ using ML with random forest algorithm to identify factors associated with CV outcomes. They demonstrated that ML could predict CV outcomes with an AUC of 0.71 from significant features, including urine albumin/creatinine ratio, estimated glomerular filtration rate, age, serum creatinine, the presence of subclinical CVD, cholesterol, SBP signals from wavelet transformation, high-density lipoprotein, 90th percentile of SBP, and triglyceride levels. Similarly, Duan et al.⁵⁵ demonstrated the ability of ML to predict absolute risk reduction in CV events using data from both SPRINT and ACCORD-BP trial. ML outperformed logistic regression with higher C-statistic by using 17 covariates. Important variables included BMI, age, serum cholesterol, triglyceride, SBP, DBP, serum creatinine, and high-density lipoprotein. Furthermore, Israel et al.⁵⁶ used ML to analyze factors associated with hyponatremia in SPRINT trial, an adverse event from intensive BP control, and found that elevated high-density lipoprotein was associated with increased risk of hyponatremia, which was also validated in another cohort of 16,501 subjects.

Undoubtedly, AI has potential to uncover additional factors associated with heterogeneous treatment outcomes from different strategies of BP control. This is particularly important because trials such as SPRINT enrolled a relatively patient robust (not frail) population. These trials may not be representative of the population at large and may not account for factors such as the J-curve (impact of low DBPs) in the management of HTN.⁶⁵ Performing randomized controlled trials in this population may be impractical but big data and AI can serve to better define optimal targets apart from reanalyzing clinical trial data.⁶⁶

FUTURE DIRECTION OF AI RESEARCH AND APPLICATIONS

Although current AI research is still in early stages, accumulating evidence suggests that AI-guided HTN research allows investigators to explore novel risk factors contributing to the pathogenesis and pathophysiology of HTN by applying ML or DL to big data with multimodal inputs, such as demographic data, vital signs, traditional CV risk, CV imaging, multiomics, behavioral, environmental, and socioeconomic factors (Figure 1). The ability of AI to comprehensively identify factors associated with the development of HTN can be used for targeting healthy individuals who are at higher risk of incident HTN, and may benefit from lifestyle modification for primordial prevention of CV disease. Moreover, AI has potential to transform the diagnosis of HTN into more precise and feasible strategies with the use of AI algorithms and emerging wearable technologies, such as devices with real-time electrocardiogram and PPG signals. The standardization of convenient, continuous, noninvasive, and cuffless BP measurement needs AI algorithms to improve precision, accuracy, and reproducibility.

Figure 1.

Artificial intelligence research in hypertension. Given sufficient amount of training and validation datasets, artificial intelligence can analyze multimodal, big data from electronic health records, wearable devices, smartphone, and social media, with machine learning or deep learning algorithms to identify hidden risk factors, predict incident hypertension, estimate blood pressure, identify phenotypes and clusters of hypertension, factors determining outcomes from different blood pressure targets, and factors associated with treatment success and adherence. Abbreviations: BP, blood pressure; CV, cardiovascular; ECG, electrocardiogram; EHRs, electronic health records; HR, heart rate; HTN, hypertension.

In order to develop efficient, personalized treatment approach for HTN, big data analytics, and AI need to be implemented in the N-of-1 study designs,^67,68 which integrate all factors associated with prognosis and treatment outcomes to uncover different phenotypes of HTN, and determine the optimal BP target and interventions for each HTN patient. Data from the previous randomized controlled trials can be further reanalyzed by using AI algorithms in order to explain different treatment response across phenotypes and clusters of HTN patients. Since HTN is a heterogeneous disease, clustering patients is an important initial step of precision medicine for HTN, followed by the development of comprehensive, personalized treatment approach that may increase effectiveness of BP control and prevent complications from HTN. Thus, databases of standardized EHR with multimodal data are necessary to further this component of AI research.

Additionally, future AI-enhanced HTN care will promote patient awareness, self-monitoring, healthy behaviors, and treatment adherence, as well as the development of digital technologies. For instance, AI can be integrated into health coaching applications, which automatically analyze inputs from patients’ activity, social media, and wearable devices, and provide personalized feedback and suggestions of BP medications and lifestyle modification. Findings from the ongoing and future clinical trials of AI-integrated health care will provide more information regarding the benefits and feasibility of using AI in the clinical practice.

Future research needs to focus on precision HTN medicine utilizing AI-based technologies with multiomics data, socioeconomic, behavioral, and environmental factors will be able to prevent or delay the development of HTN, identify novel risk factors and phenotypes of patients, and improve the treatment outcomes, which can subsequently reduce the global burden of HTN.

FUNDING

Dr Tang is partially supported by grants from the National Institutes of Health and the Office of Dietary Supplements (R01DK106000 and R01HL126827).

DISCLOSURE

Dr Tang is a consultant for Sequana Medical A.G. and Owkin Inc, and has received honorarium from Springer Nature for authorship/editorship, all unrelated to the contents of this paper. All other authors have no relationships to disclose. Dr. Laffin has received honorarium from Vascular Dynamics Inc for serving as member of hypertension eligibility committee for the CALM-2 Trial.