EUVIMEDEUVIMED entdecken
Crossref

L26/P-759 Machine learning prediction of usable blastocyst yield in hp-hMG-only ICSI cycles: A multicenter retrospective analysis

A. Luz, N. Itzhak, R. Hourvitz, M. Youngster, A. Hochberg, M. Baum, E. Maman, D. Chuderland Ben Arie, M. Yu, W. Zhou, K. Barletta, E. Hariton und 2 weitere

Human Reproduction · 01.07.2026 · 41 (Supplement_1)

DOI: 10.1093/humrep/deag083.1086 ↗

Kurz gefasst

Fachartikel in der Zeitschrift Human Reproduction, online erschienen am 08.07.2026. Verfasst von A. Luz und 13 weiteren Autor:innen.

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Study question Can a machine-learning model accurately predict the number of usable blastocysts in hp-hMG-only ICSI cycles based solely on data available on retrieval day? Summary answer The model can predict usable blastocyst’s number with reliable accuracy (R2=0.50, MAE=1.35), providing a useful tool for personalized patient counselling and clinical decision support. What is known already Usable blastocyst yield is a key metric driving patient counseling, laboratory workload planning, and clinical decision-making. However, yield varies substantially between cycles and depends on multiple variables. While most ART-related AI models focus on embryo ranking or pregnancy prediction, few tools forecast the total usable blastocyst cohort. A quantitative model predicting usable blastocyst count from routine retrieval-day data would assist in effectively managing patient expectations, optimizing resource allocation, and benchmarking performance across settings. Study design, size, duration This retrospective multicenter cohort study analyzed hp-hMG-only ART cycles from multiple clinics in Asia, Europe, and North America. The model was developed on 3,955 cycles, out of which 974 had full blastocyst information. Evaluation was performed on a random independent test set of 272 cycles with full outcome information. The primary outcome was the number of usable blastocysts, with predictive performance assessed using the coefficient of determination (R2) and Mean Absolute Error (MAE). Participants/materials, setting, methods The study included hp-hMG-only cycles with complete records for total oocytes, M2 oocytes, 2PN fertilization, and usable blastocyst yield. An XGBoost regression framework was employed utilizing patient baseline characteristics, ovarian reserve markers and treatment parameters. To simulate real-time retrieval-day prediction—in which fertilization results are not yet known, the algorithm first predicted the number of 2PN, which was then fed into a second model that forecast the total number of usable blastocysts. Main results and the role of chance In the independent test cohort, the model demonstrated robust overall performance when predicting the number of usable blastocysts. (R2=0.50, MAE=1.35). The mean number of blastocysts in the test set was 2.56. Additional subgroup analysis by maternal age was performed. Due to small sample sizes of young patients (<30; n = 16) or old patients (>45; n = 3), the subgroup analysis focused on four core age ranges: 30–34 (n = 40), 35–39 (n = 81), 40–42 (n = 82), and 43–45 (n = 50). Predictive error (MAE) decreased as age increased: 1.66 (ages 30–34), 1.42 (35–39), and 1.14 (40–42) and 0.95 (43–45). The model accurately tracked the biological decline in blastocyst formation. The absolute difference in yield ratios was consistent at 0.02-0.04 across all analyzed groups, reflecting a slight but stable underestimation. Sensitivity analysis using broader inclusion criteria (n = 345, including cycles without complete MII data) confirmed these patterns with similar agreement (absolute difference 0.01–0.03). Limitations, reasons for caution Limitations include the retrospective design and restriction to hp-hMG-only cycles, which limits generalizability to other drugs. Additionally, differences in laboratory techniques between clinics may affect accuracy. Prospective validation is required to confirm these findings. Wider implications of the findings Our model achieves strong performance in hp-hMG monotherapy cycles, confirming that clinical outcomes are consistent and primarily driven by age. This supports post-retrieval counseling and benchmarking. Broader external validation across stimulation regimens and prospective evaluation are warranted before clinical deployment. Trial registration number No

Zitieren

APA

Luz, A., Itzhak, N., Hourvitz, R., Youngster, M., Hochberg, A., Baum, M., Maman, E., Arie, D. C. B., Yu, M., Zhou, W., Barletta, K., Hariton, E., Polyzos, N. P., & Hourvitz, A. (2026). L26/P-759 Machine learning prediction of usable blastocyst yield in hp-hMG-only ICSI cycles: A multicenter retrospective analysis. Human Reproduction, 41(Supplement_1). https://doi.org/10.1093/humrep/deag083.1086

Vancouver

Luz A, Itzhak N, Hourvitz R, Youngster M, Hochberg A, Baum M, et al. L26/P-759 Machine learning prediction of usable blastocyst yield in hp-hMG-only ICSI cycles: A multicenter retrospective analysis. Human Reproduction. 2026;41(Supplement_1). doi:10.1093/humrep/deag083.1086

BibTeX
@article{Luz2026,
  author = {Luz, A. and Itzhak, N. and Hourvitz, R. and Youngster, M. and Hochberg, A. and Baum, M. and Maman, E. and Arie, D. Chuderland Ben and Yu, M. and Zhou, W. and Barletta, K. and Hariton, E. and Polyzos, N. P. and Hourvitz, A.},
  title = {L26/P-759 Machine learning prediction of usable blastocyst yield in hp-hMG-only ICSI cycles: A multicenter retrospective analysis},
  journal = {Human Reproduction},
  year = {2026},
  volume = {41},
  number = {Supplement_1},
  doi = {10.1093/humrep/deag083.1086}
}
Kostenlos starten →