A machine learning decision tree integrates plasma steroid hormone profiles measured by LC-MS/MS for rapid and precise etiological diagnosis of congenital disorders of adrenal steroidogenesis (CDAS).
The model achieved >97% accuracy with high sensitivity and specificity across multiple CDAS subtypes in a large development cohort and was independently validated.
Key discriminatory steroids identified include 11-deoxycortisol, 17-hydroxyprogesterone, 21-deoxycortisol, and corticosterone, which enable subtype differentiation with biological coherence.
This approach offers a clinically interpretable, scalable tool that can enhance pediatric endocrine diagnostics and aid early targeted intervention.
