Using machine learning to discriminate non-classical 21-hydroxylase deficiency from polycystic ovary syndrome: an external validation study
Authors' conclusion
Machine learning combined with basal LC-MS/MS steroid profiling enables accurate identification of NC21OHD without dynamic stimulation.
Study context
- Population
- Women with polycystic ovary syndrome (PCOS) or NC21OHD; training set Pitié-Salpêtrière Hospital, Paris (263 PCOS, 59 NC21OHD); 2 independent validation cohorts, Saint-Antoine Hospital, Paris, and Bordeaux University Hospital (104 PCOS, 21 NC21OHD)
- Participants
- 447 women3 centres
- Comparison
- Basal LC-MS/MS steroid profiling machine learning model to discriminate NC21OHD from PCOS
- Outcome
- Diagnostic performance in distinguishing NC21OHD from PCOS based on baseline steroid signatures
Methodological detailstricentric, external validation, machine learning (orthogonal partial least squares discriminant analysis), internal cross-validation
- Tricentric
- External validation
- Machine learning (orthogonal partial least squares discriminant analysis)
- Internal cross-validation
Performance
As reported in the abstract
Discrimination of NC21OHD from PCOS in external validation cohorts
| Test or model | Sensitivity | Specificity | Accuracy |
|---|---|---|---|
| 20-steroid model | — not stated in the abstract | — not stated in the abstract | 100% |
| Simplified 3- and 6-steroid models | ∼90% | ∼97% | — not stated in the abstract |
Sensitivity
20-steroid model— not stated in the abstract
Simplified 3- and 6-steroid models∼90%
Specificity
20-steroid model— not stated in the abstract
Simplified 3- and 6-steroid models∼97%
Accuracy
20-steroid model100%
Simplified 3- and 6-steroid models— not stated in the abstract
— not stated in the abstract
Further findings
- 21-deoxycortisone, 21-deoxycortisol and 17-hydroxyprogesterone were the most discriminant metabolites
Original publication
Using machine learning to discriminate non-classical 21-hydroxylase deficiency from polycystic ovary syndrome: an external validation study. Eur J Endocrinol. Published 7 July 2026. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →