Is AI the future of personalized hypertension risk assessment, diagnosis, and treatment?

By Hugo Francisco de SouzaJan 16 2024Reviewed by Lily Ramsey, LLM

In a recent study published in Hypertension Research, researchers reviewed recent advances in artificial intelligence (AI) applications for treating hypertension (HTN) and its comorbidities.

Study: Recent developments in machine learning modeling methods for hypertension treatment. Image Credit: chayanuphol/Shutterstock.com

Background

They summarize more than 35 studies and reveal that recent advances in home blood pressure (BP) monitoring and the increased integration of diverse data sources may result in the development of personalized machine-learning solutions for HTN risk assessments, diagnosis, and treatment.

They further explore current challenges faced in the field and suggest increased implementations of interdisciplinary collaborations between AI experts, healthcare providers, and clinicians as the way forward for AI's increased application in HTN care.

Why do we need AI for HTN care?

Hypertension (HTN) is a common condition characterized by high blood pressure (>140/90 mmHg). Patients' hearts must work significantly harder than normal during the condition to ensure adequate blood flow to different somatic features.

While rarely life-threatening on its own, HTN, especially if left untreated, is associated with a host of cardiovascular comorbidities, including heart disease and stroke.

Unfortunately, HTN is usually a silent disease with almost no externally visible symptoms. Despite nearly 50% of adults living with the condition, 30% or more remain unaware of the disease, posing severe challenges in its diagnosis and management.

In recent years, advances in computational power and the influx of artificial intelligence (AI) into medical research have spurned significant growth in the field, with proponents of AI suggesting that machine learning (ML) models can reduce the burden on the limited availability of cardiovascular experts by taking the lead in HTN risk assessment and diagnoses.

Several crucial advancements have paved the way for the influx of ML models in HTN research, most notably the extensive recent integration of hitherto distinct data sources – electronic health records, environmental data, and personalized health records, traditionally analyzed separately, have provided ML models with the raw materials needed to reveal population- and individual-specific patterns in HTN pathology.

As noteworthy is the increased proliferation of smartwatches and other wearable data loggers, which, while not as accurate as clinical tools, provide high-density and constant data on lifelong and other vital parameters, including heart rate, physical activity levels, and blood pressure (BP).

Finally, deep-learning models, a significant step forward for traditional AI models, have enabled researchers to achieve unprecedented accuracy in HTN and BP diagnoses by capturing intricate patterns in interdisciplinary data.

This has resulted in the increased personalization of these approaches, potentially benefitting patients by tailoring clinical interventions to their specific needs instead of the 'one-size-fits-all' approach.

About the study

In the present study, researchers summarised more than 35 recent publications on the applications of AI in HTN risk assessment, diagnosis, and treatment. The study focuses on modeling methods utilizing high-density home BP data and research prioritizing explainability for personalized treatment.

"Home BP is strongly associated with the development of cardiovascular diseases and is the most important index for HTN management"

Big data – advances in BP measurements and their role in AI development

Conventionally, HTN patients recorded two or more BP readings (usually in the morning and evening), which were noted in a 'BP diary' for later evaluation by a clinician.

The size reduction of BP measurement devices and the increasing popularity of wearables (e.g., smart watches) with integrated BP measurement have resulted in the advent of data digitalization and the advent of big data.

Many recent studies have highlighted the advantages of wearables and smartphone applications in BP treatment assistance.

These approaches, however, face one significant challenge – accuracy validation. Despite research by Fleischhauer et al. and Tan et al. developing novel measurement methodologies, including photoplethysmography (PPG) and deep learning, to improve wearable accuracy (average error rates <7 mmHg), these methodologies remain novel and lack clinical validation.

The European Society of Hypertension (ESH) has gone so far as to validate cuffless BP monitors separately from their conventional cuff counterparts.

"Currently, cuffless BP monitors are not intended to replace traditional BP monitors using cuff. Further validation of the device's performance as a measurement device and its clinical efficacy and effectiveness is needed to clarify its clinical position."

This challenge notwithstanding, cuffless BP monitors have provided an extensive digital dataset for training and improving AI models. Combinations of continuously logged BP data with readily available hospital test values, health records, and demographic data have resulted in the development of 'time-series' analyses capable of predicting HTN risk months or even years before disease onset.

"If AIs can predict future BP values based on past BP trends, doctors will know when BP levels exceed the HTN criteria at an early stage. Li et al. developed an AI that sequentially predicts mean BP values every month for three months using BP, BMI, sex, age, latitude, and longitude"

Personalized HTN treatment – the AI advantage

One of the strongest arguments against using AI in medical research was AI's lack of simultaneous accuracy and explainability.

The publications of LIME (2016) and SHAP (2017) overcame these arguments by introducing the concept of "local explanation," thereby removing generalization as a requirement for explainability.

This concept, encouragingly, unintentionally ushered in the age of personalized, AI-guided patient care wherein personal data is used to inform clinical interventions in place of the 'one-size-fits-all' medical approaches of the past.

"In particular, for personalized medicine aimed at preemptive disease prevention, it is essential to encourage lifestyle modifications in response to early signs of disease. Accordingly, it is expected that other XAI methods and comprehensive systems utilizing machine learning models will be implemented in addition to technologies such as SHAP."

Counterfactual explanations and individual conditional expectations (ICE) are two novel concepts driving personalized predictive healthcare. When a patient's current body metrics suggest the future risk of HTN or a similar disease, these concepts allow AI to simulate the factors and behavioral modifications required to reduce or alleviate said risk.

A current challenge in these approaches is that of modification feasibility. Just because a set of modifications can reduce HTN risk doesn't mean the patient can easily achieve those 'optimal' lifestyle changes (it's unlikely that a patient who uses a wheelchair could do a 10 km daily run to improve their HTN risk assessment).

The future integration of SHAP technologies with ICEs and other AI techniques may allow for the development of guided AI interventions that provide recommendations informed by the patient's health records and daily routine, allowing for small cumulative modifications with synergistic effects against HTN.

Conclusions

The present study summarizes the progress, challenges, and future directions of AI's application in HTN risk assessment, personalized treatment, and diagnosis.

AI has the potential to aid or even replace human medical practitioners in some of these fields but faces a notable limitation in its novelty and lack of clinical validation.

Nonetheless, deep learning models are increasingly being developed to improve patient well-being, fueled by the ever-increasing digital outputs of smart devices and wearables that log real-time patient data in a way never before seen.

With targeted studies aimed at overcoming current limitations in measurement accuracy, a future where our smartwatches and smartphones suggest lifestyle changes aimed at improving our overall health is closer than ever before.