An automated method developed by UM-IHC/UMSOM researchers scours patient records, predicts absent data and flags missed diagnoses of pulmonary hypertension in clinical practice.
Diagnosing pulmonary hypertension just got a little bit easier. A new artificial intelligence (AI) clinical tool developed by researchers at the University of Maryland Institute for Health Computing (UM-IHC) and the University of Maryland School of Medicine (UMSOM) could help physicians identify patients with pulmonary hypertension who might otherwise go undiagnosed—potentially connecting them more quickly with lifesaving care.
Pulmonary hypertension is a serious condition where abnormally high blood pressure in the arteries of the lungs forces the heart to work harder, increasing the risk of heart failure. Early diagnosis is crucial but can be delayed, as the disease can be hard to recognize, especially for non-specialist clinicians, and important diagnostic data can be buried in dense medical records.
The new AI-powered clinical support tool, described in a paper published in the European Respiratory Journal on May 28, 2026, automatically analyzes electronic health records, finds key diagnostic information, and predicts and fills in missing data needed to determine whether a patient has pulmonary hypertension. The paper’s authors used their tool to analyze electronic health records from over 11,000 patients treated between 2016 and 2024 within the University of Maryland Medical System (UMMS).
“Having access to this large dataset gave us an unprecedented opportunity to test AI’s capabilities against a real clinical challenge,” said study co-author Katarina Zeder, a research associate at UM-IHC and UMSOM.
Finding missed diagnoses
Diagnosing pulmonary hypertension requires measuring mean pulmonary artery pressure (mPAP) during a procedure called right heart catheterization. An mPAP above 20 mmHg indicates a patient has the condition.
However, reports from right heart catheterization can be inconsistent, with important measurements hidden or absent—especially outside specialized pulmonary hypertension centers. For example, mPAP was reported to be missing in 7-10% of cases in prior electronic health record datasets and in 33% of cases in a large right-heart catheterization-based registry. Meanwhile, with thousands of patients undergoing these procedures each year, reviewing every report manually for hidden data is not practical.
In response to this challenge, the researchers developed an AI system based on a large language model—AI technology that can quickly recognize, organize and translate vast sums of information. Their system performs a series of steps: First, it identifies whether a report contains a pulmonary artery pressure measurement, then it extracts that information and checks the result against the original medical record to confirm accuracy.
The system was trained and tested using 17,292 right heart catheterization reports from 11,029 patients treated within UMMS between 2016 and 2024. The AI tool accurately pulled key pressure measurements more than 99% of the time when compared with reviews performed by experienced pulmonary hypertension clinicians.
Filling in the data gaps
With their tool, the researchers addressed a second major challenge in the electronic health record: missing data.
In some cases, the specific value needed to diagnose pulmonary hypertension is absent from right heart catheterization reports. By combining machine learning with the AI language model, the team developed a method that accurately predicts those missing mPAP measurements—using a simple formula based on other available pressure measurements.
When the researchers applied this method to 507 patient records where the key measurement was missing, the AI-assisted approach identified 382 patients whose estimated pressure met the criteria for pulmonary hypertension.
“Finding these patients lets us connect them with specialized care and treatment, and in some cases, clinical trials,” said study co-author Bradley Maron, UM-IHC co-executive director and the Melvin Sharoky, MD Professor of Medicine at UMSOM. “This means better management of the condition and potential lives saved.”
A partner to physicians
The researchers say the tool is not a replacement for physicians, but rather a way to support clinical decision-making by helping doctors find important data that may otherwise be overlooked.
“This tool showcases the power of where clinical medicine and academic research intersect in the treatment of complex diseases,” said Ian Brooks, senior director for research informatics at UMMS and the UM-IHC, who was not an author on the study. “It’s a great example of how an IHC partnership lets us find and help thousands of at-risk patients using the de-identified clinical data available from across UMMS.”
As electronic health records contain vast sums of data in disparate formats, AI tools like this one could also be applied to identifying missed diagnoses of other complex diseases.
“The next step is to test the pulmonary hypertension tool prospectively within the electronic health record system to see how it performs in real-time clinical care,” Maron said. “Ultimately, we hope the technology will help thousands of patients receive earlier diagnoses and specialist referrals, leading to speedier treatments and better long-term health.”
***
The paper, “A Clinical Support Tool Using Artificial Intelligence to Diagnose Pulmonary Hypertension,” by Seyed M. Shams, Mary E. Maldarelli, Yijun Yin, Steven Cassady, Gautam Ramani, Colleen M. Ennett, Bradley A. Maron, and Katarina Zeder, was published in Eur Respir J on May 28, 2026.
Research reported on in this publication was supported by United Therapeutics and Cardiovascular Medical Research and Education Fund (CMREF) and the National Institutes of Health under grant numbers 5R01HL139613-03, R01HL163960, R01HL153502 and R01HL155096-01. A total of 1% percent of the work was financed with federal funds. A total of 99% percent was financed with non-governmental funds.
The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.
