EUVIMEDEUVIMED entdecken
Crossref

L26/P-816 Using machine learning (ML), anti-mullerian hormone (AMH) is a key predictor of IVF live birth (LB) outcomes

M. Yao, E.T. Nguyen, A. Herweck, E.C. Feinberg

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.1143 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von M. Yao und 3 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question Does AMH predict IVF LB outcomes prior to IVF cycle start (preIVF), and is there an AMH threshold below which no LBs occurred? Summary answer Using ML, AMH and female age (Age) are a priori LB predictors, with varying relative importance by geographic regional specificity of the data set. What is known already Age, AMH and IVF LB have complex, non-linear relationships. While AMH has been accepted as a strong predictor of oocyte count, its role in predicting IVF LB has been supported and challenged. Clinicians expect a strong IVF LB predictor to identify patients with near-zero probability of success; no threshold has been established. It is unclear if AMH has prognostic redundancy with clinical diagnoses such as polycystic ovarian syndrome (PCOS), diminished ovarian reserve (DOR), and endometriosis (ENDO). Resolving these controversies is crucial for patient counselling, management, and IVF access expansion. We tested the hypothesis that AMH is a preIVF LB predictor. Study design, size, duration Retrospective cohort design was applied to IVF cycles from 13 states and District of Columbia, from Society of Assisted Reproductive Technology Clinic Outcome Reporting System, USA. Dataset 1 (DS1) comprised 189,653 IVF cycles (2019-2021) from 145,680 patients with Age<42, first IVF cycle in 2019-2021. Localised datasets were created for 5 states: DS-CA, DS-IL, DS-MA, DS-NC, DS-TX. We excluded treatments pertaining to donor oocyte, donor embryo, gestational carrier, fertility preservation, or natural cycle IVF. Participants/materials, setting, methods ML preIVF models were trained using 5-fold cross-validation and gradient boosted machine on pretreatment variables or Age-control, to predict per-cycle IVF LB outcomes based on linked embryo transfers from 2019-2022, for each dataset. Each model was replicated 10 times to compute mean model metrics. Models were assessed using area-under-the-curve of receiver-operating-curve (AUC), AUC improvement (over Age-control), posterior log odds ratio compared to Age-control (PLORA). Relative importance was used to quantify each predictor’s contribution. Main results and the role of chance The overall LB rates for DS1, DS-CA, DS-IL, DS-MA, DS-NC, DS-TX, were 46.9%, 49.9%, 43.0%, 46.8%, 54.8%, 50.4%, respectively. ML preIVF LB models achieved the following AUC ± standard deviation, AUC improvement and PLORA. DS1: 0.7153±0.0001, 9.7%, 4563; DS-CA: 0.7112±0.0002, 9.5%, 718: DS-IL: 0.7322±0.0002, 11.0%, 479; DS-MA: 0.7297±0.0004, 13.2%, 463; DS-NC:0.7255±0.0007, 11.6%, 220; DS-TX: 0.7122±0.0003, 9.5%, 550. Despite having smaller sample sizes, the state models showed comparable and excellent model performance. Age and AMH were the top 2 predictors in all the models. However, the relative importance of Age and AMH varied across models, ranging 28.8% to 46.3% for Age, and 25.3% to 50.6% for AMH. Specifically, AMH had the second highest relative importance in all preIVF models except for the preIVF models for DS-IL and DS-MA, where AMH was the top-ranked IVF LB predictor. preIVF LB prognosis required ML-multivariate model as no AMH threshold predicted near-zero LB probability for patients Age<40. The preIVF model for DS1 used 8 predictors (Age, AMH, DOR, had prior IVF elsewhere, had prior IVF at current center, BMI, D3FSH), whereas the state models used 12-14 predictors including predictors not used by the DS1 model: non-PCOS ovulatory disorders, PCOS, teratozoospermia, unexplained infertility, and infertility duration. Limitations, reasons for caution Inter-model variations of predictors and their relative importance warrant further investigation and may be associated with varying patterns of clinical profiles, IVF access, clinical management, AMH usage, and data entry. The reported ML preIVF models were designed for hypothesis testing; clinic-specific validation should be performed prior to clinical usage. Wider implications of the findings AMH can predict IVF LB when used with a model with good performance metrics. Predictors and relative importance vary among state models, underscoring the importance of geographic specificity to improve personalisation and local relevance of IVF prognosis. Trial registration number No

Zitieren

APA

Yao, M., Nguyen, E., Herweck, A., & Feinberg, E. (2026). L26/P-816 Using machine learning (ML), anti-mullerian hormone (AMH) is a key predictor of IVF live birth (LB) outcomes. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.1143

Vancouver

Yao M, Nguyen E, Herweck A, Feinberg E. L26/P-816 Using machine learning (ML), anti-mullerian hormone (AMH) is a key predictor of IVF live birth (LB) outcomes. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.1143

BibTeX
@article{Yao2026,
  author = {Yao, M. and Nguyen, E.T. and Herweck, A. and Feinberg, E.C.},
  title = {L26/P-816 Using machine learning (ML), anti-mullerian hormone (AMH) is a key predictor of IVF live birth (LB) outcomes},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.1143}
}
Kostenlos starten →