An artificial intelligence-based approach for selecting the optimal day for triggering in antagonist protocol cycles

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Ovarian Stimulation Protocols

Background:

• Antagonist protocol cycles are widely used in assisted reproductive technologies.
• Optimizing trigger day selection is crucial for maximizing oocyte yield.
• Current trigger day selection relies on physician experience, which can be variable.

Purpose of the Study:

• To develop and evaluate a machine-learning model for suggesting optimal trigger days in antagonist cycles.
• To maximize the number of retrieved total and mature (metaphase II) oocytes.
• To compare model-suggested trigger days with physician-selected days.

Main Methods:

• Retrospective cohort study of 9622 antagonist cycles (2018-2022).
• An XGBoost machine-learning algorithm was trained to predict optimal trigger days.
• Model performance was evaluated on 'Freeze-all oocytes', 'Fertilize-all', and 'ICSI-only' test sets.

Main Results:

• The model suggested 1, 2, or 3 trigger day options based on predicted outcomes.
• Concordant trigger days (model vs. physician) showed significant increases in oocytes and embryos retrieved across all test groups (P < 0.001).
• Average increases included 4.8 total oocytes and 3.4 MII oocytes in 'freeze-all' cycles.

Conclusions:

• Machine learning can optimize trigger day selection in antagonist cycles.
• This AI-driven approach may enhance oocyte retrieval and improve cycle outcomes.
• Implementation can lead to more standardized, patient-specific protocols and better physician decision-making.