Advanced machine learning-based screening for primary aldosteronism with plasma steroids, potassium, and renin

Area of Science:

• Endocrinology
• Medical Diagnostics
• Artificial Intelligence in Medicine

Background:

• Current screening tests for primary aldosteronism (PA) have limited accuracy, especially when patients are on antihypertensive medications.
• Accurate PA screening is crucial for timely diagnosis and management of this common cause of secondary hypertension.

Purpose of the Study:

• To develop and validate machine learning models for improved PA screening using plasma steroid and electrolyte levels.
• To compare the diagnostic performance of these models against the traditional aldosterone-to-renin ratio (ARR).

Main Methods:

• Utilized three patient datasets (N=1380) with and without PA.
• Developed feedforward neural network (FNN) models incorporating plasma steroids, potassium, and renin.
• Evaluated model accuracy with and without antihypertensive medication washout.

Main Results:

• A renin-independent FNN model including steroids and potassium demonstrated superior diagnostic accuracy.
• Optimized renin-independent models achieved higher areas under the receiver-operating-characteristic curve (0.948-0.954) compared to the ARR (0.839).
• These models maintained accuracy regardless of medication status and reduced false positives by 53-72% at high sensitivity.

Conclusions:

• Machine learning models, particularly renin-independent ones using steroids and potassium, offer a more effective screening strategy for PA.
• These advanced models reduce the need for medication washout and improve diagnostic accuracy over the ARR.
• This approach promises more efficient and accurate identification of patients with primary aldosteronism.