Heart disease has long been the quiet leader among causes of death for women, and yet fewer of them now recognize it as such than did a decade earlier. Into that gap, University of California researchers led by nurse scientist Yoshimi Fukuoka have introduced HeartBot, a conversational app meant to meet women where they already are, on their phones, in plain language. Whether a patient-facing chatbot can make heart attack warning signs familiar enough to act on in a frightening moment is a question the field is still answering, and a six-month trial of 201 women at UCSF will show whether what i
Can an AI Chatbot Help Women Recognize the Warning Signs of Heart Disease?
Accurate information was not enough; it had to be understandable too.
Mark
Why build a chatbot when there have been decades of heart disease campaigns?
Mimi
Because the campaigns haven't held. Women who knew heart disease was the leading cause of death dropped from 65% in 2009 to 44% in 2019, and the drops were steepest among Black, Hispanic and younger women.
Mark
What does HeartBot actually do in a conversation?
Mimi
It teaches basics, symptoms, risk factors and prevention, and it remembers earlier chats. If you mentioned a relative's heart attack, it can bring that up.
Luke
Remembering is nice, but the real question is whether it's safe. What stops a language model from making things up?
Mimi
Its answers come from a core knowledge bank the team built to reduce false or misleading medical information. The team started from almost 600 anticipated questions with research-based answers.
Luke
That's a design claim, though. The source doesn't report test results on how often it errs, and that's why the public release waits on safety.
Mark
And the wording issue, why did that matter?
Mimi
Users found words like "risk" and "risk factors" unclear. The team rewrote them in everyday language and feedback was positive.
Luke
Positive feedback isn't the same as better knowledge. Engagement was the earlier question.
Mimi
Right, and that's what the 201-patient, six-month randomized trial tests: retention, with half the women getting access later as a control group.
Luke
And whether anyone calls 911 sooner is still unknown. The trial measures knowledge retention, not behavior in an emergency.
Der Puls
- Women's awareness that heart disease is their leading killer slid from 65% to 44% in a decade, even as campaigns continued.
- Symptoms in women often hide in plain sight: shortness of breath, nausea, and back or jaw pain can be mistaken for something else.
- Researchers are rebuilding the conversation itself, swapping clinical jargon like 'risk factors' for everyday words and tethering the AI to a curated knowledge bank.
- The app remains limited to trial participants, with a randomized study testing whether women still remember warning signs months later.
Heart disease has long been the quiet leader among causes of death for women, and yet fewer of them now recognize it as such than did a decade earlier. Into that gap, University of California researchers led by nurse scientist Yoshimi Fukuoka have introduced HeartBot, a conversational app meant to meet women where they already are, on their phones, in plain language. Whether a patient-facing chatbot can make heart attack warning signs familiar enough to act on in a frightening moment is a question the field is still answering, and a six-month trial of 201 women at UCSF will show whether what is learned stays learned.
A woman opens an app on her phone and asks a question about heart disease. The answer comes back in plain language, and if she has mentioned that a relative had a heart attack, the app remembers. That app is HeartBot, built by University of California researchers and led by Yoshimi Fukuoka, a nurse scientist at UC San Francisco.
The purpose is narrow and practical. Heart disease kills more women than any other cause and affects more than 60 million in the United States, yet awareness has slipped: the share of women who knew it was the leading killer fell from 65% in 2009 to 44% in 2019, with the steepest drops among Black, Hispanic and younger women. Fukuoka believed campaigns alone were not reaching them. HeartBot explains that women are more likely than men to experience shortness of breath, nausea, or back and jaw pain alongside chest pain, signs that are easy to dismiss.
The team, drawing on nursing scientists, linguists, computer scientists and others across three UC campuses, started with nearly 600 questions and studied how women actually talk about their hearts. The current version, HeartBot II, runs on a large language model but draws its answers from a core knowledge bank meant to limit misleading output. Feedback also showed that words like 'risk' were confusing, so the language was made everyday. As Fukuoka put it, accurate information was not enough; it had to be engaging enough to use.
The real test now is memory. In a six-month randomized trial with 201 UCSF patients, half have immediate access and half wait, so researchers can compare what each group retains, including how to recognize a heart attack and when to call 911. The app is available only to participants for now. If the results hold, Fukuoka hopes to expand to thousands of women and perhaps place QR codes in clinics and on transit, but for the moment the question is whether a conversation with an AI helps women know what to do when minutes matter.
A woman opens an app on her phone and asks a question about heart disease. The answer comes back in plain language, and if she has mentioned before that a relative had a heart attack, the app remembers. That app is HeartBot, built by University of California researchers and led by Yoshimi Fukuoka, a nurse scientist and professor at UC San Francisco's School of Nursing, who holds a Ph.D. and is a registered nurse and Fellow of the American Academy of Nursing.
The purpose is narrow and practical. Fukuoka's team wants to know whether carrying a heart-health chatbot would raise participants' awareness of heart disease risks and sharpen their recognition of heart attack symptoms, so that they seek medical help when they need it. In conversations, participants cover the basics of heart disease, symptoms, risk factors and prevention. HeartBot explains, for example, that chest pain is common in both women and men, but women are more likely to also have shortness of breath, nausea, vomiting, and pain in the back or jaw. Those signs are less obvious and can be mistaken for something else.
Fukuoka's motivation comes from the numbers. Heart disease is the leading cause of death among women and affects more than 60 million women in the United States, with the highest prevalence among non-Hispanic Black women. The American Heart Association says one in three women dies of it. Yet awareness has slipped despite decades of public campaigns. The share of women who knew heart disease was the leading cause of death fell from 65% in 2009 to 44% in 2019, with the steepest drops among Black, Hispanic and younger women. Fukuoka said she believed an approach beyond traditional media campaigns was needed, especially for underserved populations.
HeartBot launched in 2021, built on work by cardiovascular nursing scientists, linguists, communication scientists, computer scientists, biostatisticians, programmers and graduate students from UCSF, UCLA and UC Davis. They compiled almost 600 potential user questions, each with research-based answers to variations. They also studied the language, rhythm and cadence of how women talk about heart disease, hoping the exchange would feel closer to talking with another person. The central aim, Fukuoka said, was to raise awareness of heart disease as a precursor to heart attack and to encourage appropriate care-seeking, particularly calling 911 when potential heart attack symptoms appear.
The current version, HeartBot II, was enhanced in 2025. It adds learning modules on the basics, symptoms, risk factors and prevention, and it moved from SMS text messaging to a chatbot app built on a large language model, which supports natural, personalized conversation. To keep the AI on a short leash, its responses draw from a core knowledge bank the team developed, designed to reduce the chance of false or misleading medical information.
Wording mattered too. Feedback showed that phrases that seem ordinary in medicine, such as "risk" and "risk factors," were not always clear to users. Later versions switched to everyday language, and user feedback on the change was positive. "Getting accurate health information to women was not enough," Fukuoka said. "The information also had to be understandable and engaging enough for women to use it."
The new study asks a harder question than whether women will engage: whether they retain what they learn, including how to recognize a heart attack. Fukuoka is completing a six-month randomized controlled trial with 201 female UCSF Health patients. Half get immediate access to the HeartBot II educational sessions; the other half serve as a control group and get access later. Researchers will compare how well each group retains heart disease information over time.
The app is currently available only to trial participants, through the Apple App Store and Google Play Store. The team plans a public release once efficacy and safety are established. Longer term, Fukuoka hopes to test the next version with several thousand women for a longer period, and possibly place HeartBot QR codes in clinics and public transportation spaces. For now the experiment is smaller, and the trial's results will say whether a conversation with AI helps women know what to do when minutes matter.
Bemerkenswerte Zitate
Getting accurate health information to women was not enough. The information also had to be understandable and engaging enough for women to use it.— Yoshimi Fukuoka, UCSF School of Nursing
Möchten Sie die ganze Geschichte?
Das Original lesen bei Medical Xpress ↗
Health & Medicine