Interpretable machine learning model for predicting MII oocyte retrieved following controlled ovarian stimulation: a retrospective cohort study of 24,976 IVF/ICSI cycles

Comprehensive Summary

In this study by Zhang et. al., a machine learning model was created for predicting the number of MII oocytes after controlled ovarian stimulation (COS) for patients experiencing in vitro fertilization or intracytoplasmic sperm injection following embryo transfer (IVF/ICSI-ET). A retrospective study was conducted with 24,976 infertile women who had undergone IVF or ICSI-ET. To assess clinical predictors, patient information regarding demographic characteristics, laboratory test results, and treatment parameters was retrieved. The outcome was the number of oocytes in the second meiotic division of the cycle (MII) that were retrieved per cycle. Eight machine learning models were developed. The most robust predictors for the number of MII oocytes were antral follicle count, follicle-stimulating hormone (FSH) and estradiol on initiation day, estradiol and the number of follicles ≥14mm on human chorionic gonadotropin (hCG) day, and age. Optimal predictive accuracy was demonstrated by the multilayer perceptron regression (MLPR) model, with the lowest RMSE (3.675), MAE (2.702), and the highest R2 (0.714). The number of follicles ≥14mm on hCG day was the most impactful predictor.

Outcomes and Implications

As the rise in infertility rates necessitates assistive reproductive technology (ART) usage, the steps required to successfully perform ART procedures, including COS, must be refined. The MLPR predictive model for the number of MII oocytes is beneficial because it assesses the risk of ovarian hyperstimulation. The model was developed into an online calculator for clinicians that predicts mature oocyte yield using 6 routinely-collected laboratory variables, which enhances clinical implementation and supports individualized counseling.