Development and internal validation of a prediction model for clinical pregnancy in GnRH antagonist cycles: a retrospective cohort study.

BackgroundIn clinical practice, the management of luteinizing hormone (LH) levels during controlled ovarian hyperstimulation (COH) with gonadotropin releasing hormone antagonist (GnRH-ant) protocols presents a significant challenge that can influence in-vitro fertilization (IVF) outcomes. This complex issue requires comprehensive consideration of multiple interrelated factors, including patient age, ovarian response, and other hormonal parameters. Currently, no consensus has been established regarding the optimal approach to integrating these variables and determining their weights to achieve the most favorable pregnancy outcomes.ObjectiveThis study aimed to identify key determinants of IVF outcomes and to develop predictive models for transferable embryo yield, cumulative pregnancy, and live birth in assisted reproductive technology (ART).Study designThis retrospective cohort study enrolled 570 patients who underwent the GnRH-ant protocol between January 2020 and January 2025. All eligible patients were randomly divided into a training set and a validation set. The Boruta algorithm and LASSO regression were applied to identify key clinical predictors for the number of transferable embryos, cumulative pregnancy, and live birth, respectively. Because the predictor set associated with pregnancy showed the greatest overlap with those for available embryo and delivery, a multivariable logistic regression model was built with pregnancy as the primary outcome, and a nomogram was constructed. The model's discrimination and calibration were assessed in the training set and evaluated in an internal split-sample validation set.ResultsA total of 570 patients were included in the analysis. Among 21 candidate variables, six features-age, antral follicle count (AFC), the baseline follicle-stimulating hormone (FSH) and LH on the 2nd or 3th day of menstruation, estradiol (E2) level on the day of human chorionic gonadotropin (HCG) administration and LH alterations-were consistently identified as significant predictors. A nomogram incorporating these factors was developed. The model yielded AUCs of 0.715 (95% CI, 0.658-0.771) in the training set and 0.662 (95% CI, 0.565-0.759) in the validation cohort. Calibration curves demonstrated agreement between predicted and observed clinical pregnancy probabilities.ConclusionLH alterations during controlled ovarian hyperstimulation was associated with clinical outcomes in GnRH antagonist cycles, and may serve as a candidate dynamic marker for further investigation.