Integrated Prediction System for Individualized Ovarian Stimulation and Ovarian Hyperstimulation Syndrome Prevention:...

Area of Science:

• Reproductive endocrinology and infertility research.
• Application of artificial intelligence in clinical decision support systems.
• Biostatistics and predictive modeling in healthcare.

Background:

• Accurate prediction of ovarian response and optimal follicle-stimulating hormone (FSH) dosage are crucial for effective ovarian stimulation.
• Current methods lack a comprehensive model for simultaneously predicting oocyte yield and ovarian hyperstimulation syndrome (OHSS) risk.

Purpose of the Study:

• To establish an integrated predictive model for forecasting the number of oocytes retrieved (NOR) and assessing early-onset moderate-to-severe OHSS risk.
• To guide optimal starting doses of FSH in individualized ovarian stimulation protocols.

Main Methods:

• Prognostic study involving patients undergoing their first ovarian stimulation cycles.
• Development and validation of machine learning models (11 for NOR, 11 for OHSS) using large internal and external datasets.
• Application of Shapley additive explanation for variable interpretation and development of a web-based prediction tool (InOvaSGuide).

Main Results:

• Gradient boosting regressor achieved high performance for NOR prediction (R² ≈ 0.79).
• LightGBM model showed superior performance for OHSS prediction (AUC ≈ 0.73-0.76).
• Key predictors identified include FSH starting dose to BMI ratio for NOR and baseline antral follicle count for OHSS.

Conclusions:

• An integrated framework for predicting NOR and OHSS risk across varying FSH doses has been developed.
• The predictive models were implemented in a user-friendly online tool, InOvaSGuide.
• Further prospective evaluation is recommended prior to clinical implementation.