Recent research from Rice University and Houston Methodist shows how data-driven methods can sharpen doctors’ decisions for patients with aortic regurgitation, a common heart condition where the heart valve doesn’t close properly and blood leaks backward into the heart. By applying unsupervised machine learning to comprehensive cardiac MRI and clinical data from 972 patients at four U.S. centers, the team discovered four distinct patient groups or “phenoclusters” that track with very different outcomes, including a high-risk group of women whose disease often looks milder by traditional measures but is linked to higher mortality. The research was published this spring in JACC: Cardiovascular Imaging.
The work was led by Meng Li, associate professor of statistics at Rice and co-corresponding author; Dr. Dipan Shah, the Beverly B. and Daniel C. Arnold Distinguished Centennial Chair in the Department of Cardiology at Houston Methodist and co-corresponding author; Dr. Maan Malahfji, cardiologist at Houston Methodist and first author; and Xin Tan, a Rice doctoral student in statistics, second author and the study’s machine learning lead.
“Clinicians have long suspected that not all aortic regurgitation is the same,” Li said. “Unsupervised learning of this data revealed clear, reproducible subgroups that conventional one-size-fits-all approaches can miss. This is a first step toward precision cardiology for these patients.”
Using a robust clustering pipeline on 23 clinical and cardiac MRI variables, the team identified four phenoclusters and validated them across independent sites. Two clusters were largely male with bicuspid or tricuspid aortic valves and showed better to intermediate survival, while a third cluster — older men with more comorbidities, left-ventricular scarring and dysfunction — had the highest mortality. Most striking was a female-predominant cluster with fewer signs of dramatic heart remodeling yet similarly high mortality and comparatively lower referral rates for valve replacement.
“This is exactly the kind of signal we need to uncover with advanced analytics,” Shah said. “Women in our cohort had worse outcomes despite what looked like less severe remodeling. That finding urges us to reexamine referral thresholds and ensure women are not being undertreated.”
When the researchers added the new cluster assignments to the standard risk models for predicting outcomes (like age, symptoms, heart function through ejection fraction and time-dependent surgery), the predictions became more accurate, and the computer’s grouping method gave doctors a clearer picture of which patients were at higher risk.
“Even modest gains in prediction can change our practice, such as who we watch more closely, when we refer for surgery and how we counsel patients,” Malahfji said. “In valve disease, those decisions are critically important.”
“Our pipeline handles missing values, mixed data types and complex relationships that challenge traditional statistics,” Tan added. “We also built an early prototype calculator, so clinicians can explore which cluster a new patient most likely fits into.”
Aortic regurgitation can quietly enlarge and weaken the heart over years, and while current guidelines often treat patients similarly, this study reveals meaningful differences. Younger patients, often with bicuspid valves, may tolerate significant remodeling yet still have excellent outcomes. In contrast, older patients with more comorbidities, scarring and dysfunction face high risk even when regurgitation volumes aren’t extreme. Notably, women with this condition may face elevated risk despite less dramatic remodeling, pointing to the need for earlier intervention and closer follow-up.
“The takeaway for patients is hopeful,” Malahfji said. “With better phenotyping, we can personalize surveillance and treatment — intervening earlier for those who need it and sparing low-risk patients from unnecessary procedures.”
This research was enabled by the Digital Health Institute, a multiyear initiative uniting Houston Methodist’s academic medicine and research infrastructure with Rice’s leadership in engineering, digital health and artificial intelligence. The institute is designed to accelerate collaborative research, translate ideas into scalable clinical solutions and train the next generation of digital-health leaders.
“The Digital Health Institute is about shortening the distance from lab insight to bedside impact,” Shah said. “Houston Methodist brings deep clinical expertise and world-class imaging; Rice brings cutting-edge data science and engineering. Together, we can answer questions neither could tackle alone.”
“This is a model for how universities and health systems can co-create value,” Li said. “Our teams share data, build tools and iterate quickly with patients benefitting.”
This research was supported by the Houston Methodist Research Institute, the Rice-Methodist Cardiovascular Seed Grant Program, the Rice ENRICH Office and the National Science Foundation.