Predicting cumulative live birth after IVF with growth hormone supplementation: the role of pretreatment duration and machine learning-based models

Aim To investigate the association between growth hormone (GH) pretreatment duration and cumulative live birth (CLB), identify key prognostic factors, and develop predictive models using logistic regression (LR) and machine learning (ML) algorithms among women undergoing in vitro fertilization (IVF) with GH supplementation. Methods We conducted a retrospective cohort study of women undergoing IVF/intracytoplasmic sperm injection (ICSI) with GH between February 2015 and June 2023 at a single tertiary center. Five models were developed: LR, random forest (RF), extreme gradient boosting (XGBoost), light gradient boosting machine (LightGBM), and support vector machine (SVM). Performance was assessed with cross-validation and bootstrap resampling. AUROC and Brier scores evaluated discrimination and calibration. Results Among 1,782 women, 763 (42.82%) achieved at least one live birth. Maternal age, anti-Müllerian hormone (AMH), antral follicle count (AFC), the assisted reproductive technology (ART) cycle number, and GH duration were significant predictors. LR performed best (AUROC 0.782; Brier 0.187), showing strong calibration and interpretability. ML models had moderate discrimination but inferior calibration. AMH, AFC, and GH duration were key predictors across all methods. Conclusions Both logistic regression and machine learning effectively predicted CLB after GH-pretreated IVF, but LR provided favorable discrimination, calibration, and interpretability. GH pretreatment ≥2 months was independently associated with higher odds of cumulative live birth.