Scholarlink

Background <p>Selecting an appropriate gonadotropin starting dose in GnRH antagonist protocols remains challenging due to substantial inter-individual variability in ovarian response and the need to balance treatment efficacy with safety.</p> Methods <p>In this retrospective cohort study, women undergoing IVF/ICSI with GnRH antagonist protocols were included. Machine learning models, including XGBoost and Random Forest, were developed to predict oocyte yield and moderate-to-severe ovarian hyperstimulation syndrome (OHSS) using baseline clinical and hormonal variables. A multi-objective optimization framework incorporating a composite score (oocyte yield − 5.0 × OHSS probability) was applied to derive individualized starting doses, balancing efficacy against safety.</p> Results <p>A total of 770 cycles were included in the final analysis. The model demonstrated predictive performance for oocyte yield (R² = 0.523, RMSE = 4.75) and moderate discrimination for moderate-to-severe OHSS (AUC = 0.597). Feature importance analysis identified AMH, AFC, and age as the most influential predictors. Dose-response simulations revealed substantial inter-individual variability in optimal dosing, with higher ovarian reserve patients requiring lower starting doses to achieve favorable outcomes while minimizing OHSS risk.</p> Conclusions <p>This study presents a machine learning-based multi-objective optimization framework for individualized gonadotropin dosing, offering clinically actionable decision support and advancing precision medicine in assisted reproduction.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A machine learning-based framework for predicting ovarian response and OHSS risk in GnRH antagonist IVF/ICSI cycles: a multi-objective optimization approach

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年7月20日

查看全文
  • Junbiao Mao,
  • Shuhong Luo,
  • Junling Wang,
  • Ben Yuan

摘要

Background

Selecting an appropriate gonadotropin starting dose in GnRH antagonist protocols remains challenging due to substantial inter-individual variability in ovarian response and the need to balance treatment efficacy with safety.

Methods

In this retrospective cohort study, women undergoing IVF/ICSI with GnRH antagonist protocols were included. Machine learning models, including XGBoost and Random Forest, were developed to predict oocyte yield and moderate-to-severe ovarian hyperstimulation syndrome (OHSS) using baseline clinical and hormonal variables. A multi-objective optimization framework incorporating a composite score (oocyte yield − 5.0 × OHSS probability) was applied to derive individualized starting doses, balancing efficacy against safety.

Results

A total of 770 cycles were included in the final analysis. The model demonstrated predictive performance for oocyte yield (R² = 0.523, RMSE = 4.75) and moderate discrimination for moderate-to-severe OHSS (AUC = 0.597). Feature importance analysis identified AMH, AFC, and age as the most influential predictors. Dose-response simulations revealed substantial inter-individual variability in optimal dosing, with higher ovarian reserve patients requiring lower starting doses to achieve favorable outcomes while minimizing OHSS risk.

Conclusions

This study presents a machine learning-based multi-objective optimization framework for individualized gonadotropin dosing, offering clinically actionable decision support and advancing precision medicine in assisted reproduction.