An interpretable machine learning model for individualized gonadotrophin starting dose selection during ovarian...

Area of Science:

• Reproductive endocrinology and infertility
• Machine learning in healthcare
• Assisted reproductive technology

Background:

• Optimizing gonadotrophin dosing in IVF is crucial for maximizing oocyte yield and embryo development.
• Current methods for dose selection may not be individualized, potentially leading to suboptimal outcomes or overtreatment.
• Machine learning offers a novel approach to personalize treatment protocols.

Purpose of the Study:

• To develop and validate an interpretable machine learning model for optimizing starting gonadotrophin (follicle-stimulating hormone, FSH) dosage in autologous IVF cycles.
• To assess the impact of ML-guided FSH dose selection on key laboratory outcomes: mature oocytes (metaphase II [MII]), fertilized oocytes (2 pronuclear [2PN]), and usable blastocysts.
• To evaluate the potential for FSH dose reduction with ML-guided selection.

Main Methods:

• Retrospective analysis of 18,591 autologous IVF cycles from 2014-2020 across three US centers.
• Development of an interpretable K-nearest neighbours (KNN) machine learning model to generate individual patient dose-response curves.
• Classification of cycles into 'dose-responsive' and 'flat-responsive' based on their generated curves.
• Propensity score matching to compare outcomes between optimal and non-optimal FSH dosing strategies within each response group.

Main Results:

• 30% of cycles were identified as dose-responsive and 64% as flat-responsive.
• In dose-responsive cycles, optimal FSH dosing yielded 1.5 more MII oocytes, 1.2 more 2PN embryos, and 0.6 more usable blastocysts, using 10 IU less starting and 195 IU less total FSH.
• In flat-responsive cycles, low starting FSH doses yielded 0.3 more MII oocytes, 0.3 more 2PN embryos, and 0.2 more usable blastocysts, using 149 IU less starting and 1375 IU less total FSH.

Conclusions:

• An interpretable machine learning model can effectively guide starting FSH dose selection in IVF.
• ML-guided dosing optimizes laboratory outcomes, including MII oocytes, 2PN embryos, and usable blastocysts.
• This approach allows for significant reductions in both starting and total FSH administered, improving efficiency and potentially reducing costs.