An interpretable machine learning model for predicting the optimal day of trigger during ovarian stimulation

Area of Science:

• Reproductive Medicine
• Artificial Intelligence in Healthcare
• Data Science in Clinical Practice

Background:

• Optimizing ovulation induction protocols is crucial for successful in vitro fertilization (IVF).
• Precise timing of the human chorionic gonadotropin (hCG) trigger shot is critical for maximizing oocyte maturation and subsequent embryo development.
• Deviations from optimal trigger timing can lead to suboptimal retrieval of mature oocytes and reduced embryo quality.

Purpose of the Study:

• To develop an interpretable machine learning (ML) model to optimize the day of trigger administration in IVF cycles.
• To identify the impact of early or late trigger timing on key IVF outcomes.
• To enhance the predictability of IVF cycle outcomes through data-driven insights.

Main Methods:

• Retrospective analysis of 30,278 autologous IVF cycles from three US centers (2014-2020).
• Development of interpretable ML models using linear regression, incorporating follicle counts and estradiol levels.
• Propensity score matching was employed to compare outcomes between early, on-time, and late trigger groups.

Main Results:

• The ML model identified potential early and late triggers in 48.7% and 13.8% of cycles, respectively.
• Patients with early triggers had significantly fewer mature oocytes (MII), fertilized oocytes (2PNs), and usable blastocysts compared to on-time triggers.
• Patients with late triggers also experienced a reduction in MII oocytes, 2PNs, and usable blastocysts compared to the on-time group.

Conclusions:

• An interpretable ML model can effectively optimize trigger timing in IVF.
• Suboptimal trigger timing (both early and late) negatively impacts IVF cycle yield and quality.
• Implementing this ML model holds potential for improving outcomes in a significant number of IVF patients.