The convergence of multiomics profiling and artificial intelligence (AI) is transforming the detection and subtyping of endocrine hypertension (EH), with emerging models demonstrating diagnostic accuracy comparable to invasive standards in selected settings, according to a narrative review published in the Expert Review of Cardiovascular Therapy.
EH accounts for approximately 5% to 10% of all hypertension cases and up to 30% in tertiary referral centers, yet it remains substantially underdiagnosed. Primary aldosteronism (PA) alone may account for about 10% of hypertension overall, and prevalence may be 3- to 5-fold higher than the traditionally cited 5% to 10%. Despite this, fewer than 1% of hypertensive patients undergo screening for PA in routine practice.
“This review matters because endocrine hypertension remains one of the few potentially curable forms of hypertension, yet it is frequently underdiagnosed due to complex and often invasive diagnostic pathways,” corresponding author Marcio José Concepción-Zavaleta, MD, an endocrinologist in the School of Medicine at Universidad Científica del Sur, Lima, told GI & Hepatology News. “By integrating multiomics technologies with artificial intelligence, our review outlines a paradigm shift from traditional phenotype-based evaluation toward data-driven, precision diagnostics. Earlier detection, improved subtype differentiation, and more personalized treatment strategies could significantly reduce cardiovascular risk and improve long-term outcomes for patients with hormonally mediated hypertension.”
Methods and Scope
The researchers conducted a narrative review of literature published from January 2000 through July 2025 in PubMed, Scopus, and Web of Science. Included studies addressed genomics, transcriptomics, proteomics, metabolomics, and AI-based tools in the diagnosis and management of EH, including PA, pheochromocytoma/paraganglioma (PPGL), Cushing syndrome (CS), and thyroid-related hypertension.
The review compiles data on the prevalence of these conditions, the underlying biological mechanisms, the performance of current and emerging diagnostic tests, and the practical challenges of integrating new technologies into routine care. Because this was a narrative review rather than a systematic review or meta-analysis, the quality of the included studies was not formally assessed, which the authors acknowledge may have introduced selection bias.
Molecular Subtyping and Effect Sizes
In PA, data cited in the review note that somatic mutations in KCNJ5 occur in ≥40% of aldosterone-producing adenomas overall, with reported ranges from approximately 12% to 80% depending on the cohort. CACNA1D mutations occur in roughly 10% to more than 20% of cases; ATP1A1 and ATP2B3 mutations are less common. CTNNB1 mutations account for approximately 2% to 5%. These alterations converge on calcium-dependent activation of CYP11B2, driving autonomous aldosterone production.
Multiomics analyses have also identified circulating microRNA signatures that distinguish unilateral from bilateral PA, including hsa-miR-193b-5p. Integration of CYP11B2 immunohistochemistry with next-generation sequencing expands the detectable mutational spectrum compared with Sanger sequencing.
For PPGL, plasma free or 24-hour urinary fractionated metanephrines measured by liquid chromatography–tandem mass spectrometry achieve approximately 97% sensitivity and 91% specificity. According to cited data, the incidence is estimated at 2 to 8 cases per million annually, with prevalence among hypertensive patients ranging from 0.1% to 0.6%. Approximately 3.8% of adrenal incidentalomas harbor pheochromocytoma in pooled analyses.
In CS, hypertension affects 70% to 85% of adults with overt disease. Osteopenia or osteoporosis is reported in up to 80% of cases, and diabetes in 35% to 50%. Up to 15% of patients demonstrate cyclic hypercortisolism in retrospective cohorts.
Thyroid disorders account for approximately 1% of hypertension cases. Subclinical hyperthyroidism progresses to overt disease in about 8% of patients at 1 year and 26% at 5 years.
AI and Steroidomics
Machine learning models integrating clinical, biochemical, radiomic, and mass spectrometry–based steroidomic data cited in the review have shown high diagnostic performance for identifying PA and predicting unilateral disease. In some proof-of-concept studies, AI-based steroid profiling and imaging models approached the diagnostic accuracy of adrenal venous sampling, the current reference standard for lateralization.
Similarly, multimodal language–vision models are being explored in computational pathology and imaging to support differential diagnosis and case interpretation, although these applications remain experimental.
Implementation Barriers
Although the results are promising, translating them into real-world practice is difficult. The main challenges include scattered data, privacy rules, poor data-sharing between institutions, and limited digital skills among clinicians. The authors also highlight the “black box” problem in complex AI systems, underscoring the need for explainable AI frameworks.
“Within a decade, endocrine hypertension management will likely evolve from static, phenotype-based diagnosis to dynamic, data-driven systems medicine,” they wrote.
Disclosure: The researchers reported having no financial disclosures.
Interview with Dr. Concepción-Zavalet
GI & Hepatology News invited corresponding author Marcio José Concepción-Zavaleta, MD, an endocrinologist in the School of Medicine at Universidad Científica del Sur, Lima, Perú, to comment on the work.
When you had all the data in front of you, was there a finding, or perhaps more than one, that surprised you?
Dr. Concepción-Zavaleta: What surprised us most was how consistently AI-driven models demonstrated diagnostic performance approaching that of invasive gold-standard procedures, particularly in conditions such as primary aldosteronism. Additionally, emerging multiomic signatures appear capable of identifying subtle biological alterations before overt biochemical abnormalities become clinically evident, suggesting that future detection of endocrine hypertension could occur much earlier than is currently possible.
How might the findings influence clinical practice?
Dr. Concepción-Zavaleta: Our findings suggest that artificial intelligence and multiomic profiling could progressively become complementary clinical decision–support tools, helping physicians stratify risk, select appropriate diagnostic pathways, and reduce unnecessary invasive testing. Over time, this may lead to more streamlined, cost-effective, and individualized diagnostic algorithms, ultimately transforming endocrine hypertension management into a dynamic, precision-based model rather than a static diagnostic process.
Is there anything else you'd like to say about this work?
Dr. Concepción-Zavaleta: Beyond technological innovation, we emphasize that successful implementation will require equitable access to digital infrastructure, standardized data integration, clinician training in AI literacy, and robust ethical and regulatory frameworks. We envision not a replacement of clinical expertise, but a collaborative human–machine model in which AI enhances clinical reasoning and supports more precise, patient-centered care.
