Machine learning algorithms in constructing prediction models for assisted reproductive technology (ART) related live...

Area of Science:

• Reproductive Medicine
• Biostatistics
• Machine Learning in Healthcare

Background:

• Existing models for predicting live birth after assisted reproductive technology (ART) have limitations.
• There is a lack of suitable predictive models for Chinese populations undergoing ART.

Purpose of the Study:

• To develop and validate a predictive model for live birth outcomes in Chinese couples undergoing in vitro fertilization (IVF).
• To identify key predictors of live birth in this specific demographic.

Main Methods:

• Retrospective study of 11,938 couples undergoing IVF (2015-2022) in Southwest China.
• Four machine learning algorithms (Random Forest, XGBoost, LightGBM, Binary Logistic Regression) were employed.
• Model performance was assessed using cross-validation and bootstrap methods.

Main Results:

• Seven key predictors identified: maternal age, infertility duration, basal FSH, progressive sperm motility, progesterone (P), estradiol (E2), and luteinizing hormone (LH) on HCG day.
• Both Random Forest and Logistic Regression models showed optimal performance (AUROC ~0.67).
• Maternal age, P on HCG day, and E2 on HCG day were the most significant predictors.

Conclusions:

• The logistic regression model is recommended for predicting live birth after IVF due to its simplicity and performance.
• This model provides a valuable tool for Chinese couples undergoing ART.
• The identified predictors can inform clinical decision-making and patient counseling.