Answers are grounded in this paper and cite their sources.
This study aimed to construct a machine learning-based predictive model to determine the initial gonadotropin (Gn) dose in antagonist protocols. The GnRH antagonist protocol has become increasingly prevalent in IVF treatment due to its shorter duration, reduced risk of ovarian hyperstimulation syndrome (OHSS), and improved patient convenience compared to the long agonist protocol. However, determining the optimal initial Gn dose in antagonist protocols presents unique challenges due to the shorter suppression period and more rapid follicular recruitment. Inappropriate initial dosing can lead to severe clinical consequences, including ovarian hyperstimulation syndrome (OHSS) with excessive dosing, or poor ovarian response and cycle cancellation with insufficient dosing. Current formula-based approaches and nomograms relying on simple linear regression show limited predictive accuracy, particularly for complex cases requiring dose modifications. A retrospective analysis was conducted using data from 1,970 women undergoing antagonist protocols at Peking University People’s Hospital between October 2016 and December 2021. Two advanced machine learning models were selected after preliminary evaluation of multiple algorithms: Light Gradient Boosting Machine (LGBM), a powerful ensemble learning method using gradient boosting on decision trees, and Multi-Layer Perceptron (MLP), a neural network with multiple layers. These models were chosen for their superior performance in handling nonlinear relationships and high-dimensional medical data compared to traditional linear regression, random forest, and support vector regression approaches. LGBM demonstrated superior overall performance with test set MAE of 53.25 IU and R² of 0.527, significantly outperforming the baseline formula (MAE 61.03 IU, R² 0.298). MLP achieved the lowest mean relative error (0.212) compared to baseline (0.271). In complex cases requiring dose modifications, LGBM achieved MAE of 67.53 IU, substantially lower than the clinical adjustment error of 87.07 IU, demonstrating the model’s ability to provide more accurate dosing recommendations than clinician estimates in challenging scenarios. Machine learning models, particularly LGBM, offer substantially greater predictive accuracy than conventional formula-based approaches for determining initial Gn dose in antagonist protocols. These findings support the integration of machine learning-based decision support systems into clinical practice to enhance dose optimization, potentially improving IVF outcomes, reducing complications, and decreasing treatment costs. Future research should focus on prospective multicenter validation, development of user-friendly clinical interfaces, and exploration of advanced deep learning architectures to further enhance predictive performance. Not applicable.
No takes yet. Share an insight, caveat, or question.
Wu et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: