Do machine learning models based on plasma steroids and potassium improve diagnostic accuracy for primary aldosteronism compared to the traditional aldosterone-to-renin ratio?
Renin-independent machine learning models using plasma steroids and potassium significantly improve screening accuracy for primary aldosteronism compared to the traditional aldosterone-to-renin ratio, while minimizing the need for antihypertensive medication washout.
Abstract Commonly used screening tests for primary aldosteronism (PA) provide suboptimal diagnostic accuracy, particularly with antihypertensive medication use. This study utilized three datasets totaling 1380 patients with and without PA to develop machine learning models for screening based on plasma steroids, potassium, and renin. A feedforward neural network (FNN) model with steroids and potassium improved diagnostic accuracy compared to models without potassium. Inclusion of renin negligibly improved accuracy. The FNN and other renin-independent models showed similar accuracy before and after antihypertensive medication washout, whereas renin-dependent models exhibited poorer accuracy without medication washout. Three further optimized renin-independent models outperformed the aldosterone-to-renin ratio (ARR) for screening according to areas under receiver-operating-characteristic curves of 0.948–0.954 versus 0.839 for the ARR. Those models minimize need for medication washout and, at cut-offs for optimal 90–95% diagnostic sensitivity, reduce false positives by 53–72% to more effectively screen for PA than with the ARR.
Zhang et al. (Wed,) studied this question.