Combining metabolomics and machine learning models as a tool to distinguish non-classic 21-hydroxylase deficiency from polycystic ovary syndrome without adrenocorticotropic hormone testing - Archive ouverte HAL Accéder directement au contenu
Article Dans Une Revue Human Reproduction Année : 2023

Combining metabolomics and machine learning models as a tool to distinguish non-classic 21-hydroxylase deficiency from polycystic ovary syndrome without adrenocorticotropic hormone testing

Résumé

Abstract STUDY QUESTION Can a combination of metabolomic signature and machine learning (ML) models distinguish nonclassic 21-hydroxylase deficiency (NC21OHD) from polycystic ovary syndrome (PCOS) without adrenocorticotrophic hormone (ACTH) testing? SUMMARY ANSWER A single sampling methodology may be an alternative to the dynamic ACTH test in order to exclude the diagnosis of NC21OHD in the presence of a clinical hyperandrogenic presentation at any time of the menstrual cycle. WHAT IS KNOWN ALREADY The clinical presentation of patients with NC21OHD is similar with that for other disorders of androgen excess. Currently, cosyntropin stimulation remains the gold standard diagnosis of NC21OHD. STUDY DESIGN, SIZE, DURATION The study was designed using a bicentric recruitment: an internal training set included 19 women with NC21OHD and 19 controls used for developing the model; a test set included 17 NC21OHD, 72 controls and 266 PCOS patients used to evaluate the performance of the diagnostic strategy thanks to an ML approach. PARTICIPANTS/MATERIALS, SETTING, METHODS Fifteen steroid species were measured in serum by liquid chromatography–mass spectrometry (LC-MS/MS). This set of 15 steroids (defined as ‘steroidome’) used to map the steroid biosynthesis pathway was the input for our models. MAIN RESULTS AND THE ROLE OF CHANCE From a single sample, modeling involving metabolic pathway mapping by profiling 15 circulating steroids allowed us to identify perfectly NC21OHD from a confounding PCOS population. The constructed model using baseline LC-MS/MS-acquired steroid fingerprinting successfully excluded all 17 NC21OHDs (sensitivity and specificity of 100%) from 266 PCOS from an external testing cohort of originally 549 women, without the use of ACTH testing. Blood sampling timing during the menstrual cycle phase did not impact the efficiency of our model. LIMITATIONS, REASONS FOR CAUTION The main limitations were the use of a restricted and fully prospective cohort as well as an analytical issue, as not all laboratories are equipped with mass spectrometers able to routinely measure this panel of 15 steroids. Moreover, the robustness of our model needs to be established with a larger prospective study for definitive validation in clinical practice. WIDER IMPLICATIONS OF THE FINDINGS This tool makes it possible to propose a new semiology for the management of hyperandrogenism. The model presents better diagnostic performances compared to the current reference strategy. The management of patients may be facilitated by limiting the use of ACTH tests. Finally, the modeling process allows a classification of steroid contributions to rationalize the biomarker approach and highlight some underlying pathophysiological mechanisms. STUDY FUNDING/COMPETING INTEREST(S) This study was supported by ‘Agence Française de Lutte contre le dopage’ and DIM Région Ile de France. This study was supported by the French institutional PHRC 2010-AOR10032 funding source and APHP. All authors declare no competing financial interests. TRIAL REGISTRATION NUMBER N/A.
Fichier principal
Vignette du fichier
HUMREP-22-0749.R2_Proof_hi - HAL.pdf (1.57 Mo) Télécharger le fichier
Origine : Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03996582 , version 1 (20-02-2023)

Identifiants

Citer

Guillaume Bachelot, Anne Bachelot, Marion Bonnier, Joe-Elie Salem, Dominique Farabos, et al.. Combining metabolomics and machine learning models as a tool to distinguish non-classic 21-hydroxylase deficiency from polycystic ovary syndrome without adrenocorticotropic hormone testing. Human Reproduction, 2023, 38 (2), pp.266-276. ⟨10.1093/humrep/deac254⟩. ⟨hal-03996582⟩
163 Consultations
213 Téléchargements

Altmetric

Partager

Gmail Facebook X LinkedIn More