L26/P-878 An explainable machine learning-based pre-treatment prediction models for the first IVF/ICSI outcomes
Human Reproduction · 01.07.2026 · 41 (Supplement_1)
DOI: 10.1093/humrep/deag083.1202 ↗Kurz gefasst
Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von M. Zhang und 2 weiteren Autor:innen.
Abstract (Originalsprache)
Study question Can we develop a high accuracy pre-treatment prediction model to estimate individualized possibility of a live birth? Summary answer We developed a highly accurate pre-treatment prediction model to estimate individualized possibility of a live birth and implemented it as an online clinical tool. What is known already Predictive models based on pre-treatment parameters have demonstrated insufficient accuracy in studies of various diseases, posing challenges to clinical applications. In assisted reproductive technology, it is difficult to predict an individual’s live birth probability prior to treatment owing to considerable clinical heterogeneity among couples. This study developed a novel pre-treatment model to provide individualized, highly accurate, and interpretable predictions for the first IVF/ICSI outcomes. Study design, size, duration A total of 23,107 patients underwent the first oocyte pick-up cycles between June 2015 and June 2020 at the Clinical Center. Participants/materials, setting, methods Six machine learning models were implemented to build predictive models for performance comparisons. Based on different outcomes, two pre-treatment prediction models were developed: a pre-treatment model for predicting cumulative live birth rates (CLBR model); and a pre-treatment model for predicting live birth rates after the first embryo transfer (FLBR model). Model performance was compared in terms of discrimination, calibration, and clinical utility. SHapley Additive exPlanations (SHAP) interpreted the importance of predictors in the best models. Main results and the role of chance XGBoost demonstrated the best performance for both the CLBR model (AUC: 0.75, Brier: 0.13) and the FLBR model (AUC: 0.67, Brier: 0.22), and provided higher net clinical benefit. SHAP-based visualization quantified the impacts of individual parameter, and explained each feature’s contribution to the predicted probability. SHAP analysis revealed that for the CLBR model, key predictive factors included female age, AMH, AFC, basal FSH, duration of infertility, female BMI, thin endometrium, basal LH, basal E2, and tubal factor. For the FLBR model, the key predictors were female age, AFC, AMH, basal LH, thin endometrium, male age, female BMI, duration of infertility, basal E2, basal FSH, and oligomenorrhea. Meanwhile, we confirmed 33.6 years old as the critical maternal age threshold beyond which CLBRs and FLBRs declined dramatically. Subgroup analysis showed that CLBR model had strong discriminatory ability in most subgroups, particularly for Poseidon group 3 (AUC=0.85), Poseidon group 4 (AUC=0.98), and ≥35 years old (AUC=0.85). And FLBR model had moderate discriminatory ability in the frozen embryo transfer group (AUC=0.71). An online tool was developed to assist clinicians and patients in predicting individualized pregnancy outcomes. Limitations, reasons for caution The study is limited by its single-centre design and may not be representative elsewhere. Wider implications of the findings We have developed a highly accurate pre-treatment prediction model and an online tool to predict the likelihood of live birth for patients considering their first IVF/ICSI cycle. This tool is promising to alleviate patient stress during IVF/ICSI treatment, while also facilitating clinical decision-making and enhancing the effectiveness of doctor-patient communication. Trial registration number No
Zitieren
APA
Zhang, M., Yan, M., & Yutian, Z. (2026). L26/P-878 An explainable machine learning-based pre-treatment prediction models for the first IVF/ICSI outcomes. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.1202
Vancouver
Zhang M, Yan M, Yutian Z. L26/P-878 An explainable machine learning-based pre-treatment prediction models for the first IVF/ICSI outcomes. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.1202
BibTeX
@article{Zhang2026,
author = {Zhang, M. and Yan, M. and Yutian, Z.},
title = {L26/P-878 An explainable machine learning-based pre-treatment prediction models for the first IVF/ICSI outcomes},
journal = {Human Reproduction},
year = {2026},
volume = {41},
number = {Supplement_1},
doi = {10.1093/humrep/deag083.1202}
}
EUVIMED