Clinical Report: Development of a Comprehensive Prediction Model for Identifying LGA Infants
Overview
This study developed a machine learning model to predict large-for-gestational-age (LGA) births following IVF/ICSI, achieving an AUC of 0.7003. Key predictors identified include embryo transfer strategy and parental characteristics.
Background
Large-for-gestational-age (LGA) infants pose significant risks during delivery and long-term health issues. This research aims to enhance predictive capabilities for LGA outcomes in infants conceived via IVF/ICSI.
Data Highlights
The study analyzed 17,741 singleton live births resulting from IVF/ICSI, categorizing them as appropriate for gestational age (AGA) or LGA.
Key Findings
The XGBoost model achieved an AUC of 0.7003, outperforming logistic regression's AUC of 0.6445.
Calibration analysis indicated a lower Brier score for the XGBoost model (0.2040) compared to logistic regression (0.2295).
SHAP analysis identified key predictors including embryo transfer strategy and maternal anthropometric characteristics.
Parental factors, particularly paternal characteristics, were also significant predictors of LGA outcomes.
Further prospective external validation is necessary.
Clinical Implications
The model's findings indicate identified predictors of LGA in IVF/ICSI-conceived infants, requiring external validation before clinical implementation.
Conclusion
The developed machine learning model shows moderate discriminative ability for predicting LGA risk in IVF/ICSI-conceived infants, with various parental and treatment-related factors identified.
