A hyperaldosteronism subtypes predictive model using ensemble learning.

Area of Science:

• Endocrinology
• Medical Informatics
• Machine Learning

Background:

• Aldosterone-producing adenoma (APA) is a significant cause of primary aldosteronism (PA).
• Accurate and timely diagnosis of APA is crucial for effective patient management and treatment.
• Current diagnostic pathways can be complex, potentially delaying care for curable APA patients.

Purpose of the Study:

• To develop and validate a machine-learning algorithm for predicting APA probabilities.
• To identify key clinical features that are most predictive of APA.
• To create a tool that supports PA diagnosis in primary care settings.

Main Methods:

• Retrospective analysis of the Japan Rare/Intractable Adrenal Diseases Study dataset (2006-2019).
• Development of an ensemble-learning model (ELM) combining seven machine-learning programs.
• Utilized 46 screening features and 13 confirmatory test features for model development.
• External validation of the developed algorithm.

Main Results:

• The ensemble-learning model demonstrated high predictive accuracy.
• Screening model achieved an average AUC of 0.899; confirmatory test model achieved 0.913.
• External validation showed an AUC of 0.964 for the screening model.
• Key predictors included serum potassium levels, plasma aldosterone concentration, and aldosterone-to-renin ratio.

Conclusions:

• The developed machine-learning algorithm accurately predicts APA using clinical findings.
• This novel algorithm can assist primary care physicians in diagnosing PA and identifying APA.
• The tool has the potential to improve patient outcomes by facilitating earlier diagnosis and treatment of APA.