Machine learning for classification of hypertension subtypes using multi-omics: A multi-centre, retrospective,...

Area of Science:

• Cardiovascular Medicine
• Biomarkers
• Machine Learning

Background:

• Arterial hypertension is a significant cardiovascular risk factor.
• Identifying secondary hypertension is crucial for preventing complications.
• There is a need for simplified diagnostic tests to detect secondary hypertension.

Purpose of the Study:

• To classify subtypes of endocrine hypertension (EHT) using machine learning and multi-omics analysis.
• To develop a diagnostic tool for differentiating EHT from primary hypertension (PHT).

Main Methods:

• Utilized machine learning (ML) on multi-omics (MOmics) data from plasma and urine samples.
• Analyzed 409 features including miRNAs, metabolites, and steroids from hypertensive patients and controls.
• Employed feature reduction, ML classifiers, and class balancing for robust classification.

Main Results:

• Achieved ~92% balanced accuracy in distinguishing four conditions (PA, PPGL, CS, PHT) using 57 MOmics features.
• Discriminated EHT from PHT with 0.96 AUC, 90% sensitivity, and ~86% specificity using 37 MOmics features.
• Identified specific miRNAs and metabolites as key discriminating biomarkers.

Conclusions:

• Developed an ML pipeline for differentiating EHT subtypes from PHT using multi-omics data.
• This approach advances diagnostic capabilities for EHT, increasing testing throughput.
• Accelerates the administration of appropriate treatment for EHT patients.