AI may flag blood-pressure crisis risk three days early, preliminary study finds

An AI system analyzing blood-pressure readings and wrist-wearable signals identified elevated risk of a dangerous blood-pressure crisis up to three days ahead, researchers report. The American Heart Association released the preliminary findings November 2, offering a possible window for doctors to intervene before an emergency—not a prediction that every alert will become one.

The research followed 4,190 adults ages 40–85 with stage 2 hypertension, or blood pressure of at least 140/90 mm Hg. All received care at one hospital in Sindh, Pakistan. Researchers tracked them from January 2025 to January 2026; 312 participants experienced 418 hypertensive crises during the study.

The findings are scheduled for presentation at the American Heart Association’s Scientific Sessions, November 6–9 in Chicago. They are a conference abstract, not a peer-reviewed paper. The association considers such findings preliminary until a full manuscript appears in a peer-reviewed journal.

Participants used a portable cuff-based monitor to collect repeated blood-pressure measurements during daily activities and sleep. A separate wrist-worn device used light-based sensors to capture cardiovascular signals. The AI combined blood-pressure patterns with heart rate and heart-rate variability, the variation in timing between heartbeats.

Researchers examined 47 characteristics in those data, looking for combinations associated with a crisis within the next 72 hours. Lead author Waleed Razzaq Chaudhry, an internal medicine resident at the University of Maryland Capital Region Medical Center, described the goal as finding warning patterns that a single blood-pressure reading cannot capture.

• No overnight decrease: participants whose blood pressure did not fall during sleep had nearly five times the odds of an upcoming crisis.
• A sharp morning rise: an increase of more than 35 mm Hg after waking, compared with nighttime pressure, was associated with nearly four times the odds.

About 32%Participants with a positive alert

About 32% of participants with a positive AI alert experienced a hypertensive crisis.

97.4%Participants without an alert

Among participants without an alert, 97.4% remained crisis-free.

A high-risk prediction triggered an alert prompting a physician to contact the patient, review symptoms and adjust medication if needed. The condition being anticipated was not simply an elevated reading: the study defined a hypertensive crisis as pressure above 180/120 mm Hg accompanied by acute organ damage.

Researchers reported about 61% fewer hypertensive crises and 45% fewer emergency-department visits among patients whose doctors received an alert and adjusted medication within 24 hours, compared with those without timely alert-triggered intervention. Those comparisons link prompt care with better outcomes; they do not establish that the AI itself caused the reductions.

The alert figures also need that treatment context. Most participants with an alert did not experience a crisis, but alerts could lead to medication changes. The proportion who went on to have a crisis therefore should not be read as a simple count of useful versus useless warnings.

The authors say the single-hospital findings need testing in more diverse populations and healthcare settings before wider use. They also call for research on how clinicians should respond: confirming blood pressure, checking symptoms and medication adherence, contacting patients, considering treatment changes, and deciding when urgent or emergency evaluation is necessary.

The wrist device’s role here should not be confused with replacing a blood-pressure cuff. For home blood-pressure measurement, the American Heart Association recommends an automatic upper-arm cuff monitor. It does not recommend wrist or finger monitors because their readings are less reliable.