Abstract
Background
Arterial hypertension can be considered a global epidemic, and despite increased disease awareness and the availability of effective treatment options, global control rates remain unsatisfactory. Early identification of endocrine hypertension (EHT), including primary aldosteronism (PA), pheochromocytoma/paraganglioma (PPGL), and Cushing syndrome (CS), is essential for patient-tailored management, as these conditions are potentially curable and require disease-specific treatment strategies. However, current diagnostic pathways are complex and resource-intensive.
We previously demonstrated promising results using targeted metabolomics (TM) to distinguish primary hypertension (PHT) from EHT in a retrospective cohort. The aim of this prospective multicenter study was to further examine the diagnostic performance and reproducibility of our previous findings.
Methods
In this prospective study, patients with PHT and EHT were recruited from seven European centers within the ENSAT-HT consortium. Targeted metabolomics profiling was performed on plasma samples using liquid chromatography–mass spectrometry. Discriminatory metabolites were identified using two complementary analytical strategies: a conventional statistical approach based on sequential univariate and multivariate analyses and a machine learning approach using random forest models.
Results
From July 2016 till March 2020 a total of 1,050 patients were recruited (474 PHT: mean age 49 years, 46.2% female; 576 EHT: mean age 50 years, 41.4% female), including 495 PA, 38 PPGL, and 43 CS patients.
Among 155 metabolites and 22 metabolic indices analyzed, the conventional statistical approach identified 66 discriminatory metabolites (14 of 31 previously identified) and 14 metabolic indices (4 of 7 previously identified). The machine learning approach identified 19 discriminatory metabolites (4 of 15 previously identified) and 9 metabolic indices (3 of 8 previously identified).
The discriminative performance of the top 15 metabolites (C10, C12, C14:1, C18:2, C2, lysoPCaC17:0, PCaaC32:2, PCaaC34:1, PCaaC34:2, PCaaC34:3, PCaaC36:1, PCaaC36:2, PCaaC36:3, Serotonin, Taurine) identified by the conventional approach (AUC 0.764) and the 19 metabolites (Ac-Orn, C10, C12, C14:1, C18:1, C18:2, C2, Glu, lysoPCaC17:0, PCaa32:2, PCaa34:1, PCaa34:2, PCaaC34:3, PCaaC36:1, PCaaC36:2, PCaaC36:3, PCaeC36:3, Serotonine, Taurine) identified by the machine learning approach (AUC 0.72) was comparable between methods but lower than observed in the previous retrospective study (AUC 0.86 and 0.83, respectively).
Conclusions
Targeted metabolomics demonstrated robust discriminatory performance in this large prospective multicenter cohort of hypertensive patients. However, only a subset of previously identified metabolites and metabolic indices was reproducibly confirmed. Understanding the sources of these differences will be essential before targeted metabolomics can be implemented in routine clinical practice.
Arterial hypertension can be considered a global epidemic, and despite increased disease awareness and the availability of effective treatment options, global control rates remain unsatisfactory. Early identification of endocrine hypertension (EHT), including primary aldosteronism (PA), pheochromocytoma/paraganglioma (PPGL), and Cushing syndrome (CS), is essential for patient-tailored management, as these conditions are potentially curable and require disease-specific treatment strategies. However, current diagnostic pathways are complex and resource-intensive.
We previously demonstrated promising results using targeted metabolomics (TM) to distinguish primary hypertension (PHT) from EHT in a retrospective cohort. The aim of this prospective multicenter study was to further examine the diagnostic performance and reproducibility of our previous findings.
Methods
In this prospective study, patients with PHT and EHT were recruited from seven European centers within the ENSAT-HT consortium. Targeted metabolomics profiling was performed on plasma samples using liquid chromatography–mass spectrometry. Discriminatory metabolites were identified using two complementary analytical strategies: a conventional statistical approach based on sequential univariate and multivariate analyses and a machine learning approach using random forest models.
Results
From July 2016 till March 2020 a total of 1,050 patients were recruited (474 PHT: mean age 49 years, 46.2% female; 576 EHT: mean age 50 years, 41.4% female), including 495 PA, 38 PPGL, and 43 CS patients.
Among 155 metabolites and 22 metabolic indices analyzed, the conventional statistical approach identified 66 discriminatory metabolites (14 of 31 previously identified) and 14 metabolic indices (4 of 7 previously identified). The machine learning approach identified 19 discriminatory metabolites (4 of 15 previously identified) and 9 metabolic indices (3 of 8 previously identified).
The discriminative performance of the top 15 metabolites (C10, C12, C14:1, C18:2, C2, lysoPCaC17:0, PCaaC32:2, PCaaC34:1, PCaaC34:2, PCaaC34:3, PCaaC36:1, PCaaC36:2, PCaaC36:3, Serotonin, Taurine) identified by the conventional approach (AUC 0.764) and the 19 metabolites (Ac-Orn, C10, C12, C14:1, C18:1, C18:2, C2, Glu, lysoPCaC17:0, PCaa32:2, PCaa34:1, PCaa34:2, PCaaC34:3, PCaaC36:1, PCaaC36:2, PCaaC36:3, PCaeC36:3, Serotonine, Taurine) identified by the machine learning approach (AUC 0.72) was comparable between methods but lower than observed in the previous retrospective study (AUC 0.86 and 0.83, respectively).
Conclusions
Targeted metabolomics demonstrated robust discriminatory performance in this large prospective multicenter cohort of hypertensive patients. However, only a subset of previously identified metabolites and metabolic indices was reproducibly confirmed. Understanding the sources of these differences will be essential before targeted metabolomics can be implemented in routine clinical practice.
| Original language | English |
|---|---|
| Publication status | Published - 9 May 2026 |
| Event | 28th European Congress of Endocrinology - Prague Congress Centre, Prague, Czech Republic Duration: 9 May 2026 → 12 May 2026 https://www.ese-hormones.org/education-and-training/european-congress-of-endocrinology/ece2026/ |
Conference
| Conference | 28th European Congress of Endocrinology |
|---|---|
| Abbreviated title | ECE 2026 |
| Country/Territory | Czech Republic |
| City | Prague |
| Period | 9/05/26 → 12/05/26 |
| Internet address |
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 3 Good Health and Well-being
Keywords
- Targeted metabolomics
- Endocrine hypertension
- Machine learning (ML)
Fingerprint
Dive into the research topics of 'Targeted Metabolomics to Distinguish Primary from Endocrine Hypertension: Results from the Prospective ENSAT-HT Study'. Together they form a unique fingerprint.Projects
- 1 Finished
-
Omics-Based Strategies for Improved Diagnosis and Treatment of Endocrine Hypertension (ENSAT-HT) (Joint with Inserm, University of Torino, University of Padua, University of Glasgow, University of Birmingham, Radboud University Medical Centre, SleekIT Limited and Inserm Transfert)
Connell, J. (Investigator), Doney, A. (Investigator), Jefferson, E. (Investigator) & Zhou, K. (Investigator)
COMMISSION OF THE EUROPEAN COMMUNITIES
1/05/15 → 31/12/21
Project: Research
Equipment
Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver