Development and validation of risk prediction models for large for gestational age infants using logistic regression...

Area of Science:

• Maternal-fetal medicine
• Perinatal health
• Predictive modeling in obstetrics

Background:

• Large for gestational age (LGA) poses significant risks to maternal and infant health.
• Establishing accurate prediction models for LGA is crucial for timely intervention.
• Existing methods for LGA prediction require enhancement for early-stage identification.

Purpose of the Study:

• To develop and validate predictive models for identifying pregnancies at high risk of delivering a large for gestational age (LGA) infant.
• To compare the performance of logistic regression and machine learning algorithms in predicting LGA.
• To provide tools for early screening and management of LGA pregnancies.

Main Methods:

• Utilized data from a cohort of 1285 Chinese pregnant women.
• Defined LGA based on the 90th percentile of birth weight for gestational age and sex.
• Employed logistic regression, decision tree, and random forest algorithms for model development and validation.
• Classified gestational diabetes mellitus (GDM) into subtypes based on insulin sensitivity and secretion.

Main Results:

• Developed three distinct LGA risk prediction models.
• Logistic regression model achieved an Area Under the Curve (AUC) of 0.760 (training) and 0.748 (validation).
• Machine learning models demonstrated superior performance: Decision Tree (AUC 0.813 training, 0.779 validation) and Random Forest (AUC 0.854 training, 0.808 validation).
• Models incorporated clinical indicators, lipid profiles, and GDM subtypes.

Conclusions:

• Successfully established and validated three predictive models for LGA risk.
• The developed models exhibit good predictive power for screening high-risk pregnancies in early third trimester.
• These models can guide the implementation of early prevention strategies for adverse pregnancy outcomes like LGA.