A groundbreaking study presented at the ENDO 2026 annual meeting in Chicago, Illinois, has unveiled a sophisticated artificial intelligence model designed to revolutionize the detection of primary aldosteronism, a frequently overlooked but dangerous cause of high blood pressure. By analyzing three decades of electronic health records (EHR) from the Mayo Clinic, researchers have developed a tool capable of identifying at-risk patients long before traditional clinical triggers might prompt a diagnostic work-up. This innovation arrives at a critical juncture, as medical guidelines shift toward more aggressive screening to combat the rising tide of cardiovascular complications associated with hormonal imbalances.
The Hidden Epidemic of Primary Aldosteronism
Primary aldosteronism (PA), historically known as Conn’s Syndrome when caused by a specific adrenal tumor, occurs when the adrenal glands produce an excess of the hormone aldosterone. This hormone plays a vital role in the body’s electrolyte balance, specifically by signaling the kidneys to retain sodium and excrete potassium. When aldosterone levels are pathologically high, the resulting sodium retention leads to increased blood volume and, consequently, severe hypertension.
While primary hypertension (essential hypertension) is often attributed to genetics, diet, and lifestyle, primary aldosteronism is a secondary form of hypertension with a specific, treatable cause. However, despite its prevalence, it remains one of the most underdiagnosed conditions in clinical medicine. Dr. Frank Lee, the lead researcher from the Mayo Clinic in Rochester, Minnesota, noted during the ENDO 2026 presentation that while the exact prevalence is debated, estimates suggest that up to 20 percent of all patients with hypertension may actually have primary aldosteronism.
The stakes for diagnosis are high. Patients with PA do not merely have high blood pressure; they suffer from the direct toxic effects of excess aldosterone on the cardiovascular system. Research indicates that these individuals face a significantly higher risk of stroke, coronary artery disease, atrial fibrillation, and heart failure compared to patients with essential hypertension, even when their blood pressure readings are identical. Furthermore, chronic exposure to high aldosterone levels can lead to permanent renal damage.
Technical Architecture and Data Foundations
The development of the AI model was made possible through the Mayo Clinic Platform, a sophisticated, privacy-preserving infrastructure that allows for the analysis of multimodal clinical data. The research team utilized de-identified data from a cohort of more than 22,000 patients, spanning a 39-year period from 1986 to 2025. This longitudinal dataset provided a rich tapestry of clinical history, allowing the AI to learn the subtle precursors of a PA diagnosis.
The model utilizes an XGBoost (Extreme Gradient Boosting) architecture. XGBoost is a highly efficient machine learning library that implements gradient-boosted decision trees designed for speed and performance. By feeding this architecture a variety of variables—including age, gender, systolic blood pressure measurements, and potassium blood levels—the researchers trained the system to recognize patterns that human clinicians might miss.
Crucially, the model also factored in ICD (International Classification of Diseases) codes related to hypertension and hypokalemia (low potassium), as well as the specific types of antihypertensive medications and potassium supplements prescribed to patients. This holistic view of the patient’s medical journey allowed the XGBoost model to predict a patient’s risk for primary aldosteronism up to 12 months prior to a formal clinical diagnosis.
Testing and Validating the AI Model
Following the training phase, the model was tested on a massive validation set consisting of 225,887 adults diagnosed with hypertension. The objective was to determine how effectively the AI could flag potential PA cases in a real-world population where the condition had not yet been suspected.
The results were statistically significant. When the researchers adjusted the model’s threshold to prioritize sensitivity (identifying as many cases as possible), the AI correctly flagged more than 90 percent of confirmed primary aldosteronism cases. In this "low-risk threshold" setting, the model missed fewer than 10 percent of cases. Under these parameters, the model identified approximately two-thirds of the hypertensive study participants as candidates for further screening.
Dr. Lee emphasized that the tool is intended to assist, not replace, clinical judgment. "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 stated. "Clinicians have been challenged to screen primary aldosteronism effectively. The tool developed by our team could offer a solution based on routine information available in a patient’s medical records."
A Timeline of Shifting Clinical Guidelines
The timing of this AI breakthrough aligns with a major shift in the endocrine community’s approach to hypertension. For decades, primary aldosteronism was considered a rare condition, and screening was typically reserved for patients with "resistant hypertension" (blood pressure that remains high despite three or more medications) or those with unexplained low potassium.
However, a chronological look at clinical guidelines shows a steady move toward broader intervention:
- 2008: The Endocrine Society releases its first major guideline, focusing screening on high-risk subgroups.
- 2016: Updated guidelines begin to acknowledge that PA is more common than previously thought, but screening rates remain below 3 percent in most primary care settings.
- 2025: The Endocrine Society releases "Primary Aldosteronism: An Endocrine Society Clinical Practice Guideline," calling for more widespread, almost universal screening of hypertensive patients. This guideline emphasizes that early detection can lead to surgical cures or targeted medical therapy (mineralocorticoid receptor antagonists), which are far more effective than standard blood pressure medications for these patients.
- 2026: The Mayo Clinic AI study is presented, providing a technological "engine" to power the widespread screening mandated by the previous year’s guidelines.
Economic and Public Health Implications
The failure to diagnose primary aldosteronism is not only a clinical failure but an economic one. Patients with undiagnosed PA often undergo years of "trial and error" with various antihypertensive drugs, many of which do not address the underlying hormonal issue. During this time, the risk of expensive and debilitating events—such as strokes or heart failure—increases.
By implementing an AI-based screening model within EHR systems, healthcare networks could potentially save billions of dollars in long-term care costs. Early diagnosis allows for targeted treatments:
- Adrenalectomy: If the excess aldosterone is caused by a benign tumor on one gland, surgical removal can often cure the hypertension entirely.
- Targeted Pharmacotherapy: If both glands are overactive, specific drugs like spironolactone or eplerenone can block the effects of aldosterone, providing much better protection against cardiovascular damage than general vasodilators.
The AI model’s ability to use "routine information" is its greatest strength. It does not require expensive new tests to flag a patient; it simply re-evaluates the data already being collected during standard check-ups. This makes it an ideal tool for resource-strapped primary care environments where specialized endocrine knowledge may be limited.
Analysis of Potential Challenges
Despite the model’s success, several hurdles remain before it can be implemented globally. The "two out of three" flagging rate, while highly sensitive, implies a high volume of follow-up testing. The gold standard for confirming PA is the Aldosterone-to-Renin Ratio (ARR) test, which can be sensitive to patient positioning, salt intake, and the very medications the patients are already taking.
If an AI flags 66 percent of all hypertensive patients for screening, the laboratory infrastructure must be prepared to handle the surge in ARR testing. Furthermore, there is the risk of "alert fatigue" among physicians. If an EHR system generates too many notifications, clinicians may begin to ignore them. Therefore, the integration of Dr. Lee’s model will require careful "user interface" design to ensure that flags are presented at the right time in the clinical workflow.
Expert Reactions and Future Directions
The presentation at ENDO 2026 sparked significant discussion among attendees. Independent endocrinologists noted that the use of XGBoost reflects a growing trend in "precision medicine," where machine learning bridges the gap between massive datasets and individual patient care.
"The challenge with primary aldosteronism has always been the ‘screening gap,’" said one attendee not involved in the study. "We know the patients are out there, but the criteria for screening have been too narrow for too long. Using AI to scan the EHR 12 months before a clinician would even think to order a renin test is a game-changer."
The Mayo Clinic team plans to continue refining the model by incorporating more diverse datasets from other healthcare systems to ensure the AI’s accuracy across different demographics. As the medical community moves toward the 2030s, the integration of such predictive models into standard EHR platforms like Epic or Cerner could become the new standard of care, ensuring that "secondary" causes of hypertension are no longer treated as secondary priorities.
By turning 30 years of medical history into a predictive engine, this study provides a roadmap for the future of endocrinology—one where data-driven insights prevent the complications of tomorrow by identifying the hidden risks of today.
