The landscape of hypertension management is facing a potential paradigm shift following the presentation of a groundbreaking artificial intelligence study at the ENDO 2026 conference in Chicago, Illinois. Researchers from the Mayo Clinic have unveiled a sophisticated machine learning model designed to identify patients at high risk for primary aldosteronism (PA), a frequently overlooked but treatable cause of secondary hypertension. By analyzing 30 years of routine electronic health records (EHR), the AI tool offers a scalable solution to a long-standing clinical challenge: the systemic underdiagnosis of a condition that significantly elevates the risk of life-threatening cardiovascular events.
Primary aldosteronism, often referred to as Conn’s Syndrome in its specific forms, occurs when the adrenal glands—small, triangular organs situated atop each kidney—produce an excessive amount of the hormone aldosterone. This hormonal imbalance leads to the retention of sodium and the depletion of potassium, causing a surge in blood volume and, consequently, severe high blood pressure. While primary hypertension (essential hypertension) is often managed through lifestyle changes and standard medications, primary aldosteronism requires specific targeted therapies, such as mineralocorticoid receptor antagonists or surgical intervention, to mitigate its unique physiological damage.
The Silent Epidemic of Primary Aldosteronism
Despite being a leading cause of secondary hypertension, primary aldosteronism remains largely unrecognized in primary care settings. Dr. Frank Lee, the lead researcher from the Mayo Clinic in Rochester, Minnesota, emphasized during the ENDO 2026 presentation that the true prevalence of the condition is likely much higher than historical data suggests. Current estimates indicate that up to 20 percent of all patients diagnosed with hypertension may actually be suffering from primary aldosteronism.
The clinical stakes of missing this diagnosis are exceptionally high. Patients with PA do not merely have high blood pressure; they experience a disproportionately higher risk of cardiovascular morbidity compared to those with essential hypertension. The chronic overexposure to aldosterone exerts toxic effects on the heart and kidneys, leading to increased rates of stroke, coronary artery disease, atrial fibrillation, and heart failure. Furthermore, the renal impact can result in chronic kidney disease, which often progresses more rapidly than in cases of standard hypertension.
A Longitudinal Approach to AI Training
The strength of the Mayo Clinic study lies in its use of the Mayo Clinic Platform, a federated and privacy-preserving infrastructure that allowed researchers to access multimodal clinical data spanning from 1986 to 2025. This 30-year dataset provided a robust foundation for the AI’s learning phase, utilizing de-identified information from more than 22,000 patients.
The researchers employed a gradient-boosting machine learning library known as XGBoost (Extreme Gradient Boosting). This architecture is renowned in the data science community for its efficiency and predictive power, particularly when dealing with structured tabular data found in medical records. The model was trained to analyze a complex web of variables, including:
- Patient demographics (age and gender).
- Historical ICD (International Classification of Diseases) codes related to hypertension and hypokalemia (low potassium).
- Longitudinal systolic blood pressure measurements.
- Serum potassium levels recorded over decades.
- Prescription histories, specifically focusing on antihypertensive medications and potassium supplements.
By processing these variables, the XGBoost model was able to identify subtle patterns that often elude human clinicians during routine visits. Crucially, the model was designed to predict a patient’s risk for primary aldosteronism up to 12 months before a formal clinical diagnosis was traditionally made.
Validating the Model: High Sensitivity and Clinical Feasibility
After the training phase, the AI was tested against a massive cohort of 225,887 adults with hypertension. The results demonstrated that an AI-driven approach to screening is not only feasible but highly effective at identifying the "needles in the haystack."
In the testing phase, researchers adjusted the model’s threshold to prioritize sensitivity—ensuring that as many potential cases as possible were flagged. At this setting, the AI correctly identified more than 90 percent of confirmed primary aldosteronism cases. The "miss rate" was less than 10 percent, a significant improvement over current manual screening rates, which some studies suggest capture fewer than 3 percent of eligible patients in standard clinical practice.
Under these parameters, the model flagged approximately two-thirds of the study participants as candidates for further diagnostic work-up. While this represents a large portion of the hypertensive population, Dr. Lee argued that this selective screening approach is far more efficient than the current "wait and see" method, which often results in patients only being tested after they have already suffered a stroke or heart failure.
"Clinicians have been challenged to screen primary aldosteronism effectively due to the complexity of existing diagnostic protocols," Dr. Lee stated. "The tool developed by our team could offer a solution based on routine information already available in a patient’s medical records, removing the need for initial specialized testing and allowing for earlier intervention."
Contextualizing the 2025 Clinical Practice Guidelines
The timing of the Mayo Clinic study is particularly relevant given the recent updates to medical standards. In 2025, the Endocrine Society released its updated "Primary Aldosteronism: An Endocrine Society Clinical Practice Guideline." This document issued a clarion call for more widespread and aggressive screening for the condition, noting that the medical community has historically been too conservative in its diagnostic approach.
The 2025 guidelines highlighted the "screening gap"—the massive discrepancy between the number of people who meet the criteria for PA testing and the number of people who actually receive it. Traditionally, screening was reserved for patients with resistant hypertension or those with spontaneous hypokalemia. However, the new guidelines suggest that a much broader swath of the hypertensive population should be evaluated, especially those with grade 2 or 3 hypertension.
The AI model presented at ENDO 2026 provides the technological "engine" needed to fulfill the 2025 guidelines. By automating the identification process within the EHR, health systems can implement the Endocrine Society’s recommendations without placing an undue cognitive burden on primary care physicians who are already stretched thin.
Chronology of the Research and Implementation
The journey toward this AI breakthrough reflects a decades-long evolution in both endocrinology and data science:
- 1986–2010: Collection of longitudinal data begins at Mayo Clinic, capturing the long-term progression of hypertensive patients before the widespread adoption of modern AI.
- 2015–2020: The rise of big data in healthcare leads to the creation of the Mayo Clinic Platform, enabling researchers to utilize federated data sets while maintaining strict patient privacy.
- 2023–2024: Development of the XGBoost screening algorithm begins, focusing on identifying the specific biomarkers and medication patterns associated with aldosterone excess.
- 2025: The Endocrine Society updates clinical guidelines, providing the medical justification for broader screening.
- June 2026: Dr. Frank Lee presents the study’s findings at ENDO 2026, demonstrating that the AI model can successfully identify 90% of cases a year in advance.
Broader Impact and Economic Implications
The implications of this study extend beyond individual patient outcomes to the broader healthcare economy. Primary aldosteronism is often a "hidden" driver of healthcare costs. When the condition goes undiagnosed, patients often require multiple expensive antihypertensive medications that only partially control their blood pressure. Furthermore, the costs associated with treating the complications of PA—such as long-term disability following a stroke or the management of end-stage renal disease—are astronomical.
By using AI to facilitate early diagnosis, healthcare systems can shift toward a preventive model. Early detection allows for the use of targeted mineralocorticoid receptor antagonists (like spironolactone or eplerenone) or, in cases of unilateral adrenal adenomas, a surgical cure via adrenalectomy. These interventions can normalize blood pressure and, in many cases, reverse the damage to the cardiovascular system.
Moreover, the "federated" nature of the Mayo Clinic Platform used in this study suggests a blueprint for future medical research. By using a privacy-preserving infrastructure, researchers can train AI models on massive, diverse datasets without moving sensitive patient information out of its original secure environment. This addresses one of the primary hurdles in AI development: the tension between data access and patient confidentiality.
Expert Reactions and Future Directions
While the medical community has reacted with optimism to the ENDO 2026 presentation, experts note that the next step involves prospective validation. While the retrospective analysis of 30 years of data is compelling, clinical trials will be needed to see how the AI performs in "real-time" primary care environments.
Logistical questions also remain regarding the "two-thirds" flag rate. If an AI flags 66% of hypertensive patients for screening, clinics must be prepared for the subsequent diagnostic steps, which include the aldosterone-to-renin ratio (ARR) blood test, followed by confirmatory suppression tests and potentially adrenal vein sampling.
However, the consensus among specialists is that the current status quo—where 90% of PA cases remain undiagnosed—is no longer acceptable. The Mayo Clinic’s AI model represents a significant step toward a future where "precision screening" becomes the standard of care. By leveraging the power of 30 years of medical history, this technology promises to transform primary aldosteronism from a hidden threat into a manageable, and often curable, condition.
