The medical community is witnessing a potential paradigm shift in the management of secondary hypertension following the unveiling of a sophisticated artificial intelligence model designed to identify primary aldosteronism through routine clinical data. Presented at the ENDO 2026 annual meeting in Chicago, Illinois, the research leverages three decades of electronic health records (EHR) to address one of the most significant yet overlooked contributors to cardiovascular disease. Developed by a team of researchers at the Mayo Clinic, the model offers a proactive solution to the systemic under-diagnosis of primary aldosteronism, a condition that affects millions but remains largely undetected by conventional screening protocols.
Primary aldosteronism (PA), also known as Conn’s syndrome, occurs when the adrenal glands produce an excess of the hormone aldosterone. This hormonal imbalance triggers the body to retain sodium and lose potassium, leading to significant increases in blood pressure. Unlike primary hypertension—which is often managed through lifestyle changes and generic antihypertensive medications—primary aldosteronism requires specific targeted therapies, such as mineralocorticoid receptor antagonists or surgical intervention. When left untreated, the condition significantly elevates the risk of life-threatening complications, including stroke, coronary artery disease, atrial fibrillation, heart failure, and chronic renal disease.
The Diagnostic Gap and Clinical Urgency
The urgency of this AI-driven approach is underscored by the staggering prevalence of the condition. According to Dr. Frank Lee, the study’s lead researcher from the Mayo Clinic in Rochester, Minnesota, it is estimated that up to 20 percent of all patients diagnosed with hypertension may actually be suffering from primary aldosteronism. Despite this high prevalence, the medical community has historically struggled with screening rates. Traditional diagnostic methods involve complex blood tests to measure the aldosterone-to-renin ratio (ARR), which are often seen as cumbersome, expensive, and subject to interference from common blood pressure medications.
In 2025, the Endocrine Society released updated clinical practice guidelines titled "Primary Aldosteronism: An Endocrine Society Clinical Practice Guideline." These guidelines issued a clarion call for more widespread and aggressive screening, citing the disproportionate cardiovascular burden carried by patients with PA compared to those with essential hypertension. However, implementing these guidelines at the primary care level has proven difficult due to "screening fatigue" and the complexities of initial work-ups. The AI model presented at ENDO 2026 aims to bridge this gap by automating the identification process using data that is already available in the patient’s medical file.
Methodology and the Power of Big Data
The development of the screening tool was made possible through the Mayo Clinic Platform, a sophisticated, privacy-preserving infrastructure that utilizes federated learning and multimodal clinical data. The researchers analyzed de-identified data from more than 22,000 patients collected over a 39-year period, spanning from 1986 to 2025. This longitudinal dataset provided a rich tapestry of clinical history, allowing the AI to learn the subtle, often ignored precursors to a formal PA diagnosis.
The team utilized an XGBoost (Extreme Gradient Boosting) architecture, a powerful machine learning library known for its speed and performance in predictive modeling. The model was trained to analyze a diverse set of variables, including:
- Patient age and gender.
- ICD-coded diagnoses related to hypertension and hypokalemia (low potassium).
- Historical systolic blood pressure measurements.
- Serum potassium levels over time.
- Prescription histories for antihypertensive medications and potassium supplements.
After the initial training phase, the model was rigorously tested on a massive cohort of 225,887 adults with diagnosed hypertension. The primary objective was to determine if the AI could predict which patients were at risk for primary aldosteronism up to 12 months before a traditional clinical diagnosis was made.
Performance Metrics and Clinical Feasibility
The results of the study indicate that an AI-based approach to screening is not only feasible but highly effective. By adjusting the threshold to identify individuals at low risk, the model successfully flagged more than 90 percent of confirmed primary aldosteronism cases. Crucially, the model maintained a high level of sensitivity, missing fewer than 10 percent of cases in the test group.
In a practical clinical setting, the model identified approximately two-thirds of the study participants as candidates for further screening. While this may seem like a high volume of patients, Dr. Lee emphasized that the goal is to cast a wide enough net to capture the "missing millions" who currently cycle through ineffective blood pressure treatments while their cardiovascular systems sustain progressive damage from excess aldosterone.
"Clinicians have been challenged to screen primary aldosteronism effectively for decades," Dr. Lee stated during the presentation. "The tool developed by our team could offer a solution based on routine information available in a patient’s medical records. By identifying these patients a year before they might otherwise be diagnosed, we can intervene earlier, optimize their medication regimens, and potentially save lives."
A Timeline of Progress in Endocrine Screening
The journey toward this AI breakthrough has been decades in the making. The timeline below illustrates the evolution of primary aldosteronism recognition and the technological milestones that led to the 2026 presentation:
- 1986–2010: Collection of early EHR data at the Mayo Clinic begins, though digital integration is in its infancy.
- 2010–2020: The rise of big data in healthcare allows for the consolidation of longitudinal patient records.
- 2020–2024: Development of the Mayo Clinic Platform, enabling secure, federated access to multi-site clinical data.
- 2025: The Endocrine Society updates its clinical guidelines, emphasizing the critical need for expanded PA screening due to the link between aldosterone and systemic organ damage.
- 2025–2026: Researchers apply the XGBoost architecture to the 30-year dataset to refine predictive algorithms.
- June 2026: The AI model is officially presented at ENDO 2026, demonstrating a 90% success rate in identifying at-risk patients 12 months in advance.
Economic and Systemic Implications
The implications of integrating such a tool into standard EHR systems are profound. From a healthcare economics perspective, the early diagnosis of primary aldosteronism could lead to significant cost savings. Patients with undiagnosed PA are frequent users of healthcare resources, often requiring multiple antihypertensive drugs, emergency room visits for hypertensive crises, and long-term care for the resulting heart or kidney failure.
By identifying these patients early, healthcare systems can move toward a "precision medicine" approach for hypertension. Instead of the standard trial-and-error method of prescribing ACE inhibitors or calcium channel blockers, clinicians can move directly to mineralocorticoid receptor antagonists (like spironolactone or eplerenone) or evaluate the patient for adrenalectomy (surgical removal of the affected gland), which can be curative.
Furthermore, the use of "routine information" means that the screening process does not require additional blood draws or patient visits at the initial stage. The AI operates in the background, analyzing existing potassium levels and blood pressure trends, and then alerts the physician through the EHR interface. This "passive screening" model addresses the primary barrier to PA diagnosis: the lack of clinical suspicion among non-specialists.
Expert Analysis and Future Directions
Medical analysts suggest that the success of this model could pave the way for similar AI applications across other endocrine disorders that are frequently under-diagnosed, such as Cushing’s syndrome or pheochromocytoma. The ability of machine learning to detect patterns in potassium fluctuations and blood pressure variability—patterns that might be too subtle for a human clinician to notice during a brief office visit—marks a new era in diagnostic accuracy.
However, the transition from research to clinical practice will require careful navigation. While the AI model identified two out of three hypertensive patients for work-up, the medical community must ensure that the subsequent diagnostic infrastructure—specialized labs and adrenal imaging—can handle the increased volume. There is also the matter of "false positives," though in the context of a high-risk condition like PA, experts argue that a high sensitivity (catching as many cases as possible) is more valuable than high specificity (avoiding false alarms).
The next steps for the Mayo Clinic team involve validating the model in diverse populations outside of their own health system to ensure the algorithm’s generalizability. Integration into commercial EHR platforms like Epic or Cerner would be the final hurdle, potentially turning this research into a standard-of-care tool for primary care physicians worldwide.
As the ENDO 2026 conference concludes, the consensus among attendees is that the marriage of 30 years of clinical history with cutting-edge machine learning has provided a powerful new weapon against cardiovascular disease. By turning "dead data" in medical records into actionable insights, the medical community is closer than ever to unmasking the silent epidemic of primary aldosteronism. For the millions of patients currently struggling with "uncontrollable" high blood pressure, this AI model represents more than just a technological achievement—it represents a path toward a healthier, longer life.
