Predictive Modeling of Live Birth Outcomes Following Freeze-All FET Cycles
Overview
This study developed predictive models for live birth outcomes after freeze-all FET cycles. The CatBoost model demonstrated the highest predictive performance in the second retained transfer-record subgroup.
Background
Infertility affects a significant portion of the population, and freeze-all FET cycles are increasingly utilized in assisted reproductive technologies. Previous studies have often pooled data from first and subsequent transfers, potentially obscuring important differences in outcomes.
Data Highlights
Model
Overall Cohort AUC
First-Transfer AUC
Second-Transfer AUC
CatBoost + T6
0.704
0.812
N/A
CatBoost + T10
N/A
N/A
0.825
Key Findings
Transfer-order subgroup modeling improved prediction accuracy for live birth outcomes.
CatBoost + T6 model achieved an AUC of 0.704 for the overall cohort.
CatBoost + T10 model achieved an AUC of 0.825 for the second retained transfer-record subgroup.
SHAP analysis identified female age and ovarian-reserve indicators as key predictors.
Basal progesterone and interval days were significant in the second retained transfer-record subgroup.
Clinical Implications
The findings suggest that clinicians should consider transfer order when predicting live birth outcomes after freeze-all FET cycles. Utilizing machine learning models can enhance the accuracy of these predictions and support individualized patient counseling.
Conclusion
Transfer-order-specific modeling provides a nuanced understanding of live birth outcomes in freeze-all FET cycles.
