Development and Validation of Machine-Learning-Based Prediction Models for Thyroid Diseases During Pregnancy.

Area of Science:

• Obstetrics and Gynecology
• Endocrinology
• Medical Informatics

Background:

• International guidelines advocate for early thyroid disease screening in pregnancy based on risk factors.
• Accurate prediction of thyroid disease during pregnancy is challenging due to the complexity of these risk factors.

Purpose of the Study:

• To develop and compare multiple machine-learning models for predicting thyroid disease in pregnant women.
• To identify key clinical features associated with thyroid disease during pregnancy.

Main Methods:

• A retrospective analysis of 5461 pregnant women was conducted.
• The Boruta algorithm identified nine significant features, including age, weight, and medical history.
• Eight machine learning models were developed and evaluated, with the random forest (RF) model showing the highest performance.

Main Results:

• The random forest model achieved high accuracy (0.98-0.99) and excellent discrimination (AUC ROC 0.998-0.999) on both training and test sets.
• Key predictors included age, pre-pregnancy weight, gravidity, parity, hypertensive disorders, scarred uterus, and autoimmune disease.
• The model demonstrated strong consistency and generalizability in predicting thyroid disease.

Conclusions:

• The random forest model is highly effective for predicting thyroid disease during pregnancy.
• This machine learning approach offers a promising tool for early identification and management of thyroid dysfunction in expectant mothers.