L26/P-179 Validating a machine learning model to predict live birth after vitrified donor oocyte IVF
Human Reproduction · 01.07.2026 · 41 (Supplement_1)
DOI: 10.1093/humrep/deag083.515 ↗Kurz gefasst
Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von V. Pataia und 7 weiteren Autor:innen.
Abstract (Originalsprache)
Study question Can a machine learning (ML) model incorporating donor and recipient variables predict live birth after donor oocyte IVF more accurately than age-based counselling? Summary answer ML models outperformed age-based models, with usable embryo number being most influential, followed by embryo transfer history, recipient age, donor age, and donor AMH. What is known already IVF using donor oocytes achieves higher live birth rates than autologous IVF for older women, yet outcomes can vary. Predicting IVF outcomes is inherently challenging due to the many interacting biological and clinical variables. Patient counselling is typically based on age or average clinic success rates, as traditional regression methods are limited by the complex, non-linear relationships between donor and recipient predictors and IVF live birth outcomes. ML approaches have improved prediction of IVF outcomes using autologous oocytes, yet ML prediction of live birth following donor oocyte IVF remains largely unexplored. Study design, size, duration This retrospective cohort study analysed 3,036 embryo transfer (ET) attempts from 1,958 recipients using vitrified donor oocytes from 1,135 donors between 2017 and 2022. Gradient boosted machine models were developed and validated using five-fold cross-validation to predict live birth per ET and compared with control models using recipient age alone (RA), donor age alone (DA) or recipient and donor ages (RDA). Participants/materials, setting, methods Data were obtained from a UK-regulated vitrified donor oocyte programme. Anonymised donor and recipient clinical variables were linked per ET. Pre-treatment and pre-ET models were trained using variables available prior to treatment or ET, respectively, to predict live birth per ET. Top models were compared with age-based control models using receiver operating characteristic area under the curve (AUC), precision–recall AUC, and F1 scores. Predictor contributions were assessed using relative importance metrics. Main results and the role of chance The pre-ET model outperformed the best age-based control model, RDA, based on a 14.9% improvement in AUC (0.62 vs. 0.54) and a 12.3% improvement in precision–recall AUC (0.50 vs. 0.45), indicating improved discrimination and ability to predict live birth events, respectively. In the live birth per ET ≥ 51% range, the pre-treatment and pre-ET models had higher F1 scores than controls, indicating superior overall ability to minimize false positives and false negatives compared with controls. For example, at live birth per ET ≥ 58%, the F1 scores of the pre-treatment (0.26) and pre-ET models (0.45) were higher than those of RA (0), DA (0) and RDA (0.07). In the pre-ET model, the relative importance of total number of usable embryos was the highest (45%), followed by recipient ET number (16%), recipient age (11%), donor age (11%), and donor AMH (9%). Observed live birth rates were 43% after the first ET decreasing to 34–36% in subsequent transfers. Of the 1336 recipients (68% of 1958) who received up to 3 ETs, the cumulative live birth rate was 87.7%, highlighting the importance of treatment continuation and the potential value of more reliable and individualised prognostic counselling beyond population averages. Limitations, reasons for caution The study was conducted within a limited number of UK centres using vitrified donor oocytes and may reflect centre-specific practices and selection criteria. Live birth per ET was modelled rather than cumulative live birth rates, as recipients may receive oocytes from different donors and/or varying numbers of oocytes. Wider implications of the findings ML models integrating donor and recipient factors can improve individualised counselling in donor oocyte IVF. Quantifying prognosis beyond age alone may help manage patient expectations, support shared decision-making, and identify areas where additional biological markers could further enhance prediction accuracy. Trial registration number No
Zitieren
APA
Pataia, V., Yao, M., Nguyen, E., Wolska, M., Shah, T., Nair, S., Macklon, N., & Ahuja, K. (2026). L26/P-179 Validating a machine learning model to predict live birth after vitrified donor oocyte IVF. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.515
Vancouver
Pataia V, Yao M, Nguyen E, Wolska M, Shah T, Nair S, et al. L26/P-179 Validating a machine learning model to predict live birth after vitrified donor oocyte IVF. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.515
BibTeX
@article{Pataia2026,
author = {Pataia, V. and Yao, M.W.M. and Nguyen, E.T. and Wolska, M. and Shah, T. and Nair, S. and Macklon, N. and Ahuja, K.},
title = {L26/P-179 Validating a machine learning model to predict live birth after vitrified donor oocyte IVF},
journal = {Human Reproduction},
year = {2026},
volume = {41},
number = {Supplement_1},
doi = {10.1093/humrep/deag083.515}
}
EUVIMED