A new machine learning-based prediction model for subtype diagnosis in primary aldosteronism.

Area of Science:

• Endocrinology
• Medical Informatics
• Artificial Intelligence

Background:

• Primary aldosteronism (PA) has two subtypes: unilateral (UPA) and bilateral (BPA).
• Accurate differentiation between UPA and BPA is crucial for effective patient management.
• Current gold-standard diagnostic methods like adrenal venous sampling (AVS) have limitations.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for diagnosing PA subtypes.
• To identify key clinical parameters for predicting UPA versus BPA.
• To explore AI-driven tools for improving PA diagnosis and reducing reliance on invasive procedures.

Main Methods:

• Extracted data from the Dryad database, including ten clinical parameters.
• Developed ML models using a random forest classifier to predict PA subtypes.
• Validated the optimal model on an independent external dataset.

Main Results:

• The optimal ML model achieved high accuracy (90.0% testing, 81.4% validation) and AUC (0.938 testing, 0.887 validation).
• Key predictors included post-saline infusion test PAC, ARR, and post-captopril challenge ARR.
• The model demonstrated strong performance in distinguishing between UPA and BPA.

Conclusions:

• A novel ML-based predictive model can accurately diagnose PA subtypes using readily available clinical data.
• This AI approach offers a promising, non-invasive alternative to CT imaging and AVS.
• Future AI models could enhance clinical decision-making for PA patients.