Related Experiment Videos
Development and validation of an interpretable machine learning model for predicting incident gestational
Liran Shen1, Ru Liu2, Yunbiao Zhang1
1Department of Medical Laboratory Center, Shanxian Central Hospital, Heze, China.
Frontiers in Medicine
|July 16, 2026
Summary
A machine learning model using clinical laboratory markers effectively predicts gestational hypothyroidism (GHT). This approach offers improved early identification compared to traditional risk factors, aiding clinical decision-making.
Area of Science:
- Biomedical research
- Clinical laboratory science
- Machine learning in healthcare
Background:
- Traditional risk factors have limited accuracy for early gestational hypothyroidism (GHT) detection.
- Predictive models using clinical laboratory markers for GHT are underexplored.
Purpose of the Study:
- To develop and validate a machine learning model for predicting incident gestational hypothyroidism (GHT) using clinical laboratory markers.
- To identify key laboratory predictors for GHT.
Main Methods:
- Retrospective observational study of 407 pregnant women.
- Feature selection using LASSO, Boruta, and Random Forest algorithms.
- Development and comparison of 12 machine learning models, with LightGBM selected for its performance and interpretability via SHAP values.
Main Results:
- Nine predictors were identified: zinc, iron, copper, calcium, vitamin D, vitamin E, albumin, alanine aminotransferase, and alkaline phosphatase.
- The LightGBM model achieved an AUC of 0.916, demonstrating high accuracy, PPV, and specificity.
- SHAP analysis highlighted zinc, alkaline phosphatase, and albumin as significant contributors to GHT prediction.
Conclusions:
- A LightGBM model utilizing clinical laboratory markers provides a balanced and effective approach for predicting incident GHT.
- Further external prospective validation is recommended before clinical implementation of the model.