Validating Machine Learning Models Against the Saline Test Gold Standard for Primary Aldosteronism Diagnosis

Area of Science:

• Endocrinology
• Medical Informatics
• Artificial Intelligence

Background:

• Primary aldosteronism (PA) diagnosis is challenged by the saline infusion test's cumbersome and time-consuming nature.
• Lack of uniform protocols for the gold standard test necessitates more efficient diagnostic methods.
• Developing advanced diagnostic tools is crucial for timely and accurate PA detection in hypertensive patients.

Purpose of the Study:

• To develop and validate machine learning (ML) models for predicting primary aldosteronism (PA).
• To compare the diagnostic performance of ML models against the traditional saline infusion test.
• To enhance diagnostic efficiency and standardization for PA in hypertensive East-Asian populations.

Main Methods:

• Utilized patient data from three distinct cohorts: TAIPAI, CONPASS, and a South Korean cohort.
• Employed Random Forest, XGBoost, and deep learning techniques to identify key predictive features for PA.
• Evaluated model performance using metrics such as accuracy, sensitivity, and specificity.

Main Results:

• The Random Forest model achieved an accuracy of 0.673 (95% CI: 0.640-0.707).
• Machine learning models significantly outperformed baseline models in predicting primary aldosteronism.
• Identified key predictive features contributing to the models' diagnostic accuracy.

Conclusions:

• Machine learning models show superior performance in predicting primary aldosteronism compared to traditional methods.
• The developed models offer a potentially more efficient and standardized approach to PA diagnosis.
• Further validation in diverse populations is recommended to enhance the generalizability of ML models for PA detection.