Vinithasree Subbhuraam, PhD
AI & Digital Health Executive
June 2026
Let me begin with a story you may recognize. A 52-year-old woman arrives at an emergency department feeling extremely tired, nauseous, with a burning sensation in her jaw, and struggling with anxiety. The doctor orders standard tests. Her troponin levels are slightly elevated but remain within a range largely based on studies of male patients. She is sent home. Two days later, she returns by ambulance. She has had a heart attack.
Similar situations happen thousands of times each year in emergency rooms around the world. If you are curious, you can visit the American Heart Association (AHA)’s Go Red for Women Class of Survivors to read real accounts from female survivors about their experiences in managing heart disease and stroke symptoms.
Cardiovascular disease (CVD) is still the main threat to women’s longevity worldwide, causing about 35% of all female deaths as of 2019.1 It is the leading cause of death for women globally, and is responsible for 1 in 3 deaths in the United States. Despite years of progress, several reasons, such as lack of awareness, misdiagnosis, delayed diagnosis, and misinterpretation of heart problems as psychological or non-heart-related causes, still negatively affect the clinical outcomes for women. Women under 55 with acute cardiac ischemia have been found to be far more likely than men to be mistakenly discharged from the emergency department, according to a multicenter study published in The New England Journal of Medicine.2 Women not only face more diagnostic errors in emergencies but also encounter a distinct range of health issues, such as spontaneous coronary artery dissection (SCAD) and ischemia with non-obstructive coronary arteries (INOCA), which traditional diagnostic models, generally based on men, fail to detect.3 They also remain underrepresented in clinical trials that create the evidence used by cardiologists to make treatment decisions.
Having spent nearly two decades building AI-driven solutions in healthcare, including time specifically focused on the intersection of biomarkers, wearable technology, and women’s physiology, I have watched this crisis from both the research and product/solution perspectives. What I want to argue in this article is not simply that a problem exists. That case has been made. What I want to argue is that we now have the technical infrastructure to do something transformative about it, and that the opportunity to do so cannot wait.
The invisibility of women in cardiovascular medicine occurred because, for decades, scientific institutions treated male physiology as the standard. Landmark cardiovascular trials that established the first evidence-based treatments for heart disease largely included male participants. The Physicians’ Health Study, which showed the heart-protective effects of aspirin, included about 22,000 men and no women. The Multiple Risk Factor Intervention Trial, which shaped our current understanding of coronary risk, also left women out. It wasn’t until 1993 that the National Institutes of Health (NIH) enacted the NIH Revitalization Act. This act required the inclusion of women and minorities in federally funded clinical research. Today, even after years of reform efforts, the enrollment gap still exists. Some supporting evidence is shown below.
38%
Women enrolled in cardiovascular trials on average4
27%
Women enrolled across 141 coronary artery disease trials, 2010-2017
20.8%
Women enrolled across 51 cardiac surgery randomized trials6
Besides trial enrollment, studies show that the gender bias from physicians may also contribute to gender disparities in cardiovascular testing. Research shows that a majority of male physicians implicitly associate strength and risk-taking with male patients.7 In terms of diagnosis protocols, across multiple studies, it was observed that women with chest pain received fewer complementary examinations and were 2.5 times less likely to be referred to a cardiologist than men with comparable presentations.8
The consequence is that, when heart disease is at stake, this hesitation around appropriate testing is not merely an equity issue. It is a survival issue as we continue to see higher mortality rates and rehospitalizations in women, compared with men.9
Perhaps the deepest structural problem in women’s cardiovascular medicine is the assumption that cardiovascular disease in women is simply a delayed or attenuated version of the disease in men. It is not.
Men predominantly develop obstructive coronary artery disease (plaques accumulating in the large epicardial vessels, building up until they rupture and trigger the classic heart attack). Women, particularly younger women, more frequently develop coronary microvascular dysfunction (CMD), which is impaired blood flow through the tiny arteries that cannot be visualized on a standard angiogram. They are also disproportionately affected by SCAD, a tearing of the coronary artery wall that generally strikes women under 50. They experience takotsubo (stress-induced) cardiomyopathy at far higher rates.
The hormonal aspect of women introduces another layer of biological complexity. Pregnancy-related conditions like preeclampsia, gestational diabetes, and peripartum cardiomyopathy leave lasting effects on a woman’s cardiovascular health that standard risk models do not address or capture. Estrogen offers benefits such as improved blood vessel function and reduced inflammation, providing some cardiovascular protection before menopause. However, this protection is not guaranteed and varies widely among individuals. The transitions of perimenopause and menopause are linked to increased arterial stiffness, deteriorating lipid levels, visceral fat redistribution, and autonomic dysfunction, all of which significantly change cardiovascular risk in a short period.10 These factors are not included in standard risk calculators and are not sufficiently represented in the training data of traditional diagnostic algorithms.
What this means in practice is that a 47-year-old woman with a history of preeclampsia, a perimenopause-associated lipid shift, and new-onset fatigue may have a cardiovascular risk profile that the current models cannot adequately determine. The Framingham Risk Score, the pooled cohort equations, and their successors were developed using data in which women were underrepresented and in which these female-specific risk factors were either not measured or not modeled. They systematically underestimate risk in this population.
Now the question is, can AI help address decades of systemic inequality in heart disease research, screening, diagnosis, and care? I believe the answer is yes, particularly if these technologies are developed and validated intentionally.
Understanding where AI can have the greatest impact requires mapping it across the full care continuum: from the moment a woman first becomes aware of her cardiac risk to the moment when management and treatment decisions are made. Below are the four primary domains where AI-enabled transformation will have the most impact in women’s CVD care.
According to the AHA’s recent survey, nearly half of all American women (44%) do not recognize CVD as their leading cause of death. Among women of color and young women, awareness has actually declined over the past decade11. This is a cultural and educational problem. But AI and digital tools can play a meaningful role at the community level to address this. Here are three types of AI-based digital health solutions that I propose to help improve CVD awareness in women:
Traditional risk models struggle to capture the interactions among dozens of biomarkers, especially how they differ in female versus male physiology. They cannot adequately capture the pattern interplay between the signal embedded in an ECG waveform’s subtle morphology, menopausal timing, and lipid dynamics, and connect it to the longitudinal digital phenotypes captured over months by a wearable device. The core advantage of modern machine learning, especially the latest foundational models, over traditional risk scoring is its capacity to identify such complex, nonlinear, high-dimensional patterns in heterogeneous data.
In March 2025, a landmark paper was published in The Lancet Digital Health by researchers at Imperial College London.12 They trained a convolutional neural network on over one million ECGs to predict the biological sex from the 12-lead ECG with remarkable accuracy. More importantly, they proposed a novel biomarker called the “sex discordance score” which is the difference between the model’s prediction of sex and the patient’s self-reported sex. They also showed that women with higher discordance scores showed dramatically elevated risks of heart
failure, myocardial infarction, and all-cause mortality. What does that mean? When a woman had a higher sex discordance score, her heart functioned in ways that are more typical of male heart patterns. These women were at higher risk because their biology did not match the female reference that standard thresholds expected. The AI found a risk continuum where conventional medicine saw only a binary sex classification. If validated prospectively in diverse populations, such models could become powerful tools for personalized cardiovascular risk assessment in women.
Despite progress, conventional risk models still omit critical women-specific factors, limiting their accuracy in cardiovascular risk prediction. More and more studies show that a woman’s lifetime cardiovascular risk should not solely be a function of traditional risk factors. Rather, it should also be a function of her reproductive biography: age at menarche, number of pregnancies, pregnancy complications, age at menopause, hormonal therapy history, etc. In a very recent review, Dakhil and colleagues propose that precision medicine supported by AI provides a framework for integrating these overlooked women-specific determinants and may help close existing gaps in cardiovascular risk prediction.3
However, while the authors suggest that we include sex-specific biomarkers and hormonal influences in prediction models, the paper serves as a narrative review of existing literature. It does not present new evidence showing that this integration actually improves predictive performance in clinical practice. To obtain this important empirical evidence, we need well-designed prospective studies with large, diverse groups of women. These studies will help us determine if female-specific markers, such as reproductive history, pregnancy complications, hormonal biomarkers, longitudinal wearables data, and imaging-derived variables, truly enhance discrimination and reclassification beyond traditional risk factors in different populations. We can then use explainable AI models and foundational models to examine patterns and estimate risk from such complex, varied, longitudinal data.
Women are generally offered fewer diagnostic tests than men, mainly because of the different manifestations of CVD in women compared to men, as explained before. Women show a delayed onset of CVDs than men, have less visible blockages in the coronary arteries, present with symptoms often described as atypical (shortness of breath, fatigue, nausea, epigastric discomfort, back pain, and anxiety), and get misdiagnosed as having gastrointestinal or mental health-related symptoms. Another critical aspect is that women presenting with signs and symptoms of myocardial ischemia are more likely than men to have no obstructive coronary artery disease. Some conditions, such as microvascular angina, SCAD, and myocardial infarction with non-obstructive coronary arteries (MINOCA, INOCA), disproportionately affect women and often evade traditional diagnostic frameworks.13
For AI to effectively close these diagnostic gaps, several changes are necessary, especially to gather enough evidence to boost the general awareness among providers about sex-specific presentation and risk factors. Here are a few approaches:
Furthermore, AI tools must be trained on diverse female populations. They must be validated to identify not only obstructive coronary disease but also microvascular dysfunction. This area has historically lacked diagnostic protocols and representation in training data.
The treatment gap in women’s cardiovascular care is arguably the most consequential and least discussed of the disparities. Women with established cardiovascular disease are less likely to be prescribed statins, less likely to receive antiplatelet therapy, less likely to be referred for cardiac rehabilitation, and less likely to receive implantable devices. One key reason for this gap is that many of these guideline recommendations were derived from trials with a majority of male enrolment, and clinicians express uncertainty about applying male-derived evidence to female patients.
Just as they help generate evidence supporting risk prediction and diagnosis, AI models can play a critical role in clinical trial design and patient matching to determine sex-specific treatment protocols. AI-driven enrolment algorithms can identify women in real-world patient populations who closely match the characteristics of underrepresented subgroups in existing trials.
Building equitable AI solutions for women’s cardiovascular health requires intentional and sustained effort to ensure that female-specific data are collected systematically and comprehensively. It requires sex-stratified training cohorts with robust representation of female-specific risk factors such as pregnancy complications, hormonal trajectories, and reproductive histories. It requires outcome labels that capture the full spectrum of female cardiovascular presentations, not just the events that the system was trained to look for in men. It requires validation in prospective, diverse, sex-balanced cohorts before any clinical deployment.
Beyond gender bias, another important aspect to consider is access bias. The most sophisticated AI risk score does not help a woman who cannot afford the follow-up visit, who lives in a cardiology-sparse rural county, or whose providers lack training to recognize and manage women’s CVD. Digital health equity requires attention to distribution, access, and trust, and not just AI-based solutions.
The future of AI-powered women-centric cardiovascular care is one:
I have spent my entire career building medical AI systems at the intersection of computational accuracy and clinical impact. My experience has taught me that the toughest problems in this area are not just technical. They involve deciding whose outcomes we want to improve, whose data we want to use, and whose experience of the healthcare system we treat as the standard. For too long, women’s cardiovascular health was not one of them.
The good news is that things are changing on that front, especially due to awareness campaigns and research efforts supported by organizations such as the AHA, WomenHeart, etc. We now have increasingly sophisticated AI algorithms, wearable sensors, and digital infrastructure that create new opportunities to improve women’s cardiovascular care. What we need now is the intentional integration of these components to build systems that see women’s hearts for what they actually are and respond accordingly.
I believe we are making progress. But urgency is crucial. Every day we delay is another day in which a woman with “atypical symptoms” is sent home too soon without the proper diagnosis and treatment she deserves.
Vinitha Subbhuraam, PhD, is an applied Health AI researcher and innovator with over two decades of experience turning research into real-world healthcare solutions, including an FDA-cleared SaaS for menopausal status detection, a wearable IoT platform for breast health monitoring, and a biomarker-based cardiovascular risk-reduction platform for women in midlife. She is a highly cited data scientist who has authored 110+ peer-reviewed publications, two books on Predictive Analytics in Healthcare (IOP Publishing), and serves as the co-Editor-in-Chief of Computer Methods and Programs in Biomedicine. She is the Founder & CEO of AI First Innovations, where she directs projects on women’s cardiovascular health and oculomics, and also the Founder & Program Director of IMPACT Labs Academy, where she mentors the next generation of applied AI innovators. She is also the mentor for Johns Hopkins AI in Healthcare executive program. She writes at the intersection of AI, women’s health, and the future of health and wellness.
