Page 13 - hypertension_newsletter

REFLECTIONS
Hypertension
Hypertension Global Newsletter #10 2026

an individual’s stress levels, smooth muscle contractions,
and changes in body position, all of which alter pulse times Hypertension
independently of actual BP changes. As a result, studies
demonstrate that current cuffless devices perform poorly
when tracking BP changes induced by exercise, sleep, and
daily activities.

The AHA notes a critical lack of appropriate clinical validation
testing for these products. While protocols for cuffed
devices have existed for decades, international standards CLICK HERE
for intermittent cuffless devices (like the upcoming ISO WATCH AUTHORS DR. JORDANA B.
81060-7) are only just being developed. The statement COHEN AND DR. TAMMY M. BRADY
warns that Food and Drug Administration (FDA) clearance DISCUSS THEIR PUBLICATION ON THE
is not synonymous with measurement accuracy, as formal CLINICAL LIMITATIONS OF CUFFLESS
clinical validation is not required for a BP device to receive DEVICES FOR BLOOD PRESSURE
FDA clearance. Therefore, the AHA suggests that currently MEASUREMENT (3:44)
available cuffless BP devices should not be used for the
diagnosis or clinical management of hypertension until
they demonstrate improved precision, reliability, and clear CLICK HERE
correlations with patient outcomes. FOR THE LINK TO FULL ARTICLE

ARTIFICIAL INTELLIGENCE

Performance of large language models in analyzing common hypertension scenarios.
Zand J, et al. Hypertension. 2026 Jan;83(1):225-234.

Despite the widespread availability of hypertension Graphical abstract
treatments, BP control remains highly suboptimal, often due
to a combination of patient factors and clinician constraints,
such as limited time and therapeutic inertia. Recently,
generative artificial intelligence (AI) and large language
models (LLMs) have rapidly evolved as potential tools to
augment clinical decision-making and improve workflow
efficiency. However, their reliability in strictly adhering to
medical guidelines for managing complex, real-world clinical
scenarios remains unverified.

This study evaluated the accuracy and safety of three
publicly available – LLMs, GPT-4, Gemini 1.5 Pro, and
MedLM-medium – against human expertise. Researchers
developed 51 clinical vignettes depicting common primary
care hypertension scenarios based on AHA concepts.
These vignettes included patient data such as age, body
mass index (BMI), BP readings, and current medications.
For instance, sample vignettes evaluated the management
of patients taking specific antihypertensives. Each
vignette was processed by the three LLMs and a human
hypertension expert to answer a single prompt: “What
should be the next step to manage the hypertension?”

TABLE OF CONTENTS