Supporting first FSH dosage for ovarian stimulation with machine learning

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• In Vitro Fertilization (IVF)

Background:

• Determining the optimal initial dose of follicle-stimulating hormone (FSH) is crucial for successful ovarian stimulation in IVF.
• Current methods rely on clinician experience, which can lead to suboptimal dosing and variable outcomes.

Purpose of the Study:

• To develop and validate a machine learning (ML) model for accurately predicting the optimal first dose of FSH in women undergoing their first IVF cycle.
• To compare the performance of the ML model against clinician-prescribed doses.

Main Methods:

• An observational study included 2713 patients for model development (2011-2019) and 774 for validation (2020-2021) from five reproductive centers.
• Predictor variables included patient age, body mass index (BMI), anti-Müllerian hormone (AMH), antral follicle count (AFC), and prior live births.
• Model performance was assessed using a score based on retrieved mature oocytes (MII) relative to the FSH dose.

Main Results:

• The ML model achieved a mean performance score of 0.87 in the development phase and 0.89 in the validation phase.
• These scores were significantly higher than those for clinician-prescribed doses (0.83 development, 0.84 validation; P < 0.001 for both).
• The model demonstrated superior performance compared to standard clinical practice in optimizing FSH dosage.

Conclusions:

• The developed machine learning model accurately identifies the optimal first FSH dose for ovarian stimulation.
• This AI tool can serve as a valuable training aid for new clinicians and a quality control measure for experienced practitioners.
• Implementing this model has the potential to enhance IVF treatment efficacy and patient outcomes.