A parsimonious hormone-based triage model for selected endocrine causes of hypertension: a prospective single-center cohort study

Technical Summary: A Parsimonious Hormone-Based Triage Model for Selected Endocrine Causes of Hypertension

Problem Statement
Endocrine causes of secondary hypertension, specifically primary aldosteronism, pheochromocytoma/paraganglioma (PPGL), and Cushing syndrome, carry significant cardiovascular risks and require distinct, disease-specific diagnostic pathways. While current guidelines recommend specific screening protocols for each condition (e.g., aldosterone/renin for primary aldosteronism, metanephrines for PPGL), clinicians often face the practical challenge of prioritizing which patients undergoing a general hypertension work-up require extensive, resource-intensive endocrine evaluation. A nonspecific baseline hormone panel cannot serve as a standalone diagnostic test, yet a reproducible, parsimonious combination of baseline markers could potentially aid in triaging patients for confirmatory testing.

Methodology
This prospective, single-center cohort study was conducted at Shenzhen University of Advanced Technology General Hospital between July 2023 and January 2025. The study enrolled adults (≥18 years) referred for hypertension evaluation who met specific inclusion criteria (documented hypertension, complete baseline data, and completion of the reference-standard pathway).

• Candidate Predictors: Five baseline hormone measurements were collected prior to final adjudication: adrenocorticotropic hormone (ACTH), plasma renin activity (PRA), angiotensin II, aldosterone, and cortisol. These were natural-log transformed and standardized.
• Reference Standard: The outcome was the presence of at least one confirmed endocrine cause (primary aldosteronism, PPGL, or Cushing syndrome) determined via disease-specific diagnostic pathways (e.g., saline infusion/captopril challenge for aldosteronism, metanephrines for PPGL, concordant cortisol tests for Cushing syndrome).
• Model Development: The cohort (n=1,502) was split 70:30 into development and validation sets. Nested logistic regression models were compared sequentially:
  1. Aldosterone alone.
  2. Aldosterone + PRA.
  3. Aldosterone + PRA + Cortisol.
  4. Addition of Angiotensin II.
  5. Addition of ACTH.
     An exploratory nonlinear five-marker model using spline-transformed predictors was also evaluated.
• Performance Metrics: Models were assessed using the Area Under the Receiver Operating Characteristic Curve (AUC), Brier score, calibration (intercept and slope), sensitivity, specificity, predictive values, and likelihood ratios.

Key Results

• Cohort Characteristics: Of 5,986 screened participants, 1,502 completed the reference-standard pathway. 412 participants (27.4%) had a confirmed endocrine cause (304 primary aldosteronism, 48 PPGL, 60 Cushing syndrome).
• Model Comparison:
  • Aldosterone alone yielded an AUC of 0.814.
  • Adding PRA significantly improved performance (AUC 0.866).
  • Adding Cortisol further improved the model (AUC 0.889).
  • Adding Angiotensin II and ACTH did not materially improve the linear model (AUC remained 0.889; p-values for comparison were non-significant).
• Parsimonious Model Performance: The three-marker model (Aldosterone, PRA, Cortisol) achieved an AUC of 0.889 (95% CI 0.852–0.922) in the validation cohort. At a probability threshold of 0.339, it demonstrated:
  • Sensitivity: 72.6%
  • Specificity: 86.9%
  • Positive Predictive Value (PPV): 67.7%
  • Negative Predictive Value (NPV): 89.3%
  • Brier Score: 0.109
• Exploratory Findings: A nonlinear five-marker model achieved a slightly higher AUC of 0.900 with higher sensitivity (85.5%) but lower specificity (80.1%). Angiotensin II showed the largest numerical contribution in the PPGL subgroup, though this was based on only 13 cases and is considered hypothesis-generating.

Significance and Claims
The paper claims that a three-marker model (aldosterone, PRA, and cortisol) retains the vast majority of the discrimination power of a full five-marker panel. The authors position this model not as a standalone diagnostic tool or a replacement for established screening guidelines (such as the 2025 Endocrine Society guidelines for primary aldosteronism), but as a workflow-oriented triage tool.

The model's primary utility is to help prioritize patients for disease-specific confirmatory evaluation within the hypertension work-up. The authors explicitly state that the model should not be used to withhold guideline-recommended screening or to rule out disease, given the residual false-negative rate (sensitivity of 72.6%).

Limitations and Future Directions
The authors acknowledge several limitations:

1. Single-center design: Transportability to other settings is unknown.
2. Partial verification bias: Only participants who completed the reference-standard pathway were included, which may introduce referral bias.
3. Composite endpoint: The outcome combined three biologically distinct disorders, with primary aldosteronism dominating the sample.
4. Medication interference: Despite sensitivity analyses, residual effects of medications (e.g., ACE inhibitors, diuretics) and preanalytical variables remain a source of uncertainty.
5. Lack of external validation: The study relies on internal split-sample validation.

The paper concludes that independent external validation and assessment of partial-verification bias are required before any clinical implementation. The findings support further evaluation rather than immediate adoption.