Frank Lee, M.D., watched an AI model sort through nearly 23,000 patient records spanning four decades, searching for a hidden diagnosis that affects millions but rarely gets caught. What the model found could transform how clinicians identify primary aldosteronism—a treatable cause of high blood pressure that often masquerades as ordinary hypertension, silently raising patients' risk of heart attack, stroke, and kidney failure.
Primary aldosteronism happens when the adrenal glands, small organs perched atop each kidney, overproduce the hormone aldosterone. This throws the body's sodium and potassium balance dangerously out of whack. The condition is dangerous precisely because it's invisible: up to 20% of people diagnosed with high blood pressure actually have primary aldosteronism instead, yet most never receive the correct diagnosis. For them, the wrong diagnosis means the wrong treatment—and years of preventable cardiovascular damage.
The breakthrough came from Lee and his team at Mayo Clinic in Rochester, Minnesota, who trained their AI model on 30 years of electronic health records and then tested it on 225,887 adults with hypertension. Using an XGBoost machine learning architecture, the model learned to spot the subtle patterns—age, gender, unusual potassium levels, blood pressure measurements, medication combinations—that quietly signal primary aldosteronism. Remarkably, it could predict who would be diagnosed up to 12 months before diagnosis actually occurred.
The results were striking. When researchers set the screening threshold to err on the side of caution, the model correctly identified more than 90% of primary aldosteronism cases while missing fewer than 10%. In practical terms, about two-thirds of hypertensive patients would be flagged as candidates for further testing. "During testing on patients with high blood pressure who had never been screened previously for primary aldosteronism, our model identified approximately two out of every three patients for further work-up," Lee explained when presenting the findings at ENDO 2026, the Endocrine Society's annual meeting in Chicago.
Why does this matter so urgently? Effective treatments already exist for primary aldosteronism—medications that address the root cause rather than just the symptom. Early diagnosis can prevent future cardiovascular complications and reduce overall health care costs. Yet clinicians have lacked a reliable tool to identify who needs screening. The Endocrine Society itself called for more widespread screening in its 2025 clinical practice guideline, recognizing that primary aldosteronism increases risk of stroke, coronary artery disease, atrial fibrillation, heart failure, and kidney disease far more than ordinary high blood pressure does.
Lee's AI model solves a real bottleneck in clinical practice. It works with routine information already collected in patient medical records—no expensive new tests required. The tool draws on 39 years of Mayo Clinic data, fed through a privacy-preserving infrastructure using de-identified patient information. This approach suggests that underdiagnosis isn't inevitable; it's a problem waiting for the right solution.
For millions of people taking blood pressure medication that doesn't address their actual condition, this AI screening tool offers genuine hope—the chance to move from hidden diagnosis to visible treatment, from preventable complications to protected health.