In a groundbreaking development at the intersection of artificial intelligence and endocrinology, researchers from the Mayo Clinic have unveiled a sophisticated AI-driven screening model designed to transform the detection of primary aldosteronism (PA), a frequently overlooked but critical contributor to hypertension and subsequent cardiovascular disease. This advance emerges from a rigorous analysis of three decades’ worth of electronic health record (EHR) data and offers a promising new pathway to identify individuals at risk well before symptoms escalate, potentially preventing severe health complications.
Primary aldosteronism arises due to excessive secretion of aldosterone by the adrenal glands, small but vital structures perched atop the kidneys. Aldosterone plays a key physiological role in regulating sodium and potassium balance, thereby influencing blood pressure homeostasis. However, when produced in surplus, it disrupts this delicate electrolyte equilibrium, leading to persistent hypertension that is often refractory to standard treatment. This condition markedly heightens the probability of various cardiovascular events, including stroke, coronary artery disease, atrial fibrillation, heart failure, and renal impairment.
Despite its significant health impacts, the precise prevalence of primary aldosteronism remains elusive, though expert estimates suggest it may affect up to 20% of hypertensive patients. This under-recognition is partly due to limitations in current screening frameworks, which rely on clinical suspicion and specialized testing that are not universally applied in routine medical practice. Consequently, many PA cases remain undiagnosed, leaving patients vulnerable to unchecked disease progression.
Recognizing this critical gap, the recent study led by Dr. Frank Lee of the Mayo Clinic leverages the formidable computational power of artificial intelligence to enhance screening accuracy and coverage. By harnessing a comprehensive, de-identified dataset of over 22,000 patients collected between 1986 and 2025 via the Mayo Clinic Platform—a secure, federated infrastructure combining diverse clinical data modalities—the team constructed a machine learning model capable of discerning subtle, yet telltale patterns indicative of PA risk.
The algorithm employed, based on the XGBoost framework, incorporated a multi-dimensional array of clinical variables, including demographic factors (age, gender), diagnostic codes for hypertension and hypokalemia, systolic blood pressure readings, serum potassium levels, and records of antihypertensive or potassium supplementation prescriptions. By training on this extensive data pool, the model learned complex associations and temporal signals that precede clinical confirmation of primary aldosteronism, enabling prediction up to a year before traditional diagnosis.
Testing this innovative AI model on a broader hypertensive cohort of 225,887 adults yielded highly encouraging results. At thresholds optimized for high sensitivity, the tool successfully identified over 90% of confirmed PA cases, while maintaining a false-negative rate below 10%. Notably, this balance was achieved while designating roughly two-thirds of the hypertensive population as candidates warranting further clinical evaluation—a pragmatic proportion that could feasibly integrate into existing healthcare workflows.
This breakthrough is particularly significant given the therapeutic implications of primary aldosteronism detection. Unlike essential hypertension, PA is often amenable to targeted interventions such as mineralocorticoid receptor antagonists or surgical adrenalectomy when appropriate. Early and accurate identification thereby has the potential not only to improve patient outcomes by mitigating cardiovascular risk but also to reduce the substantial economic burden associated with untreated hypertension-related complications.
The Endocrine Society’s latest Clinical Practice Guideline, published in 2025, underscores the urgency for broader screening initiatives for primary aldosteronism, highlighting the condition’s outsized role in driving cardiovascular morbidity. The AI model developed by Lee and colleagues not only aligns with but could operationally advance these guideline recommendations by providing clinicians with an accessible, data-driven decision support tool embedded in routine care settings.
Importantly, this study exemplifies the transformative possibilities of integrating AI within real-world clinical data ecosystems. The utilization of the Mayo Clinic Platform enabled leveraging diverse and longitudinal EHR entries while preserving patient privacy through a federated architecture. This methodological sophistication ensures that predictive models remain robust, generalizable, and ethically sound, addressing key challenges in medical AI deployment.
Moreover, this AI-based approach presents an adaptable template for tackling other underdiagnosed conditions where subtle clinical signatures are masked within vast datasets. By automating risk stratification and flagging high-priority cases, healthcare systems can optimize resource allocation and foster precision medicine paradigms that are proactive rather than reactive.
As Dr. Lee indicated, with the model’s ability to detect two out of every three previously unscreened hypertensive patients who may harbor PA, clinicians gain a powerful ally in overcoming conventional screening barriers. The integration of such AI tools can streamline workflows, prompt earlier diagnostic evaluations, and ultimately save lives by preventing the downstream consequences of missed diagnoses.
The implications of this research extend beyond endocrinology, illustrating how advanced machine learning methodologies can revamp approaches to chronic disease management. By combining clinical insight with technological innovation, the frontier of personalized health care is rapidly expanding, promising a future in which data-driven strategies enhance screening efficiencies and deliver timely interventions tailored to individual risk profiles.
As primary aldosteronism remains a leading, yet often silent, cause of hypertension worldwide, the deployment of AI-enhanced screening represents a critical leap forward in public health efforts. Continued refinement of such models, integration into electronic health infrastructure, and prospective validation in diverse populations will be essential to realize their full potential and reshape the diagnostic landscape.
In summary, the Mayo Clinic’s pioneering application of XGBoost AI modeling to extensive real-world EHR data offers a scalable, precise, and resource-efficient mechanism to identify primary aldosteronism risk well in advance of clinical confirmation. This technological stride aligns with the Endocrine Society’s call for expansive screening and heralds a new era where machine learning catalyzes improved detection, treatment, and outcomes for hypertension’s most insidious subtypes.
Subject of Research: Artificial Intelligence-based screening for primary aldosteronism using electronic health records.
Article Title: AI-Driven Screening Model Enhances Early Detection of Primary Aldosteronism to Combat Hypertension-Linked Cardiovascular Risks
News Publication Date: ENDO 2026 Annual Meeting (Presentation Date: February 24, 2026)
Web References:
Mayo Clinic Platform
Endocrine Society Clinical Practice Guideline on Primary Aldosteronism (2025)
Keywords: Primary aldosteronism, hypertension, artificial intelligence, machine learning, XGBoost, electronic health records, cardiovascular disease, screening, endocrinology, Mayo Clinic Platform
Tags: adrenal gland aldosterone regulationAI hypertension detection modelAI in cardiovascular disease preventionAI predictive model for hypertension riskAI-driven endocrinology diagnosticscardiovascular risk from primary aldosteronismearly detection of hypertension causeselectronic health record analysis hypertensionhypertension refractory to treatmentMayo Clinic AI research hypertensionprimary aldosteronism screening AIunderdiagnosed hypertension causes
