Prediction Method of Gestational Diabetes Based on Electronic Medical Record Data

Area of Science:

• Medical Informatics
• Machine Learning in Healthcare
• Obstetrics and Gynecology

Background:

• Electronic medical records (EMRs) are increasingly used for auxiliary medical diagnosis.
• Predicting gestational diabetes mellitus (GDM) is crucial for maternal and fetal health.
• Existing prediction methods may not fully leverage EMR data.

Purpose of the Study:

• To develop and evaluate a predictive model for gestational diabetes using electronic medical record data.
• To assess the effectiveness of machine learning algorithms in GDM prediction.
• To identify key features within EMRs that improve GDM prediction accuracy.

Main Methods:

• Data preprocessing involved removing inaccurate records and balancing the dataset.
• Machine learning models, including Random Forest (RF) for feature importance and logistic regression for prediction, were employed.
• Feature selection using Weighted Bidirectional Feature Selection (WBFS) was applied.
• Model performance was evaluated using accuracy, F1-score, Area Under the Curve (AUC), Mean Absolute Error, and Root Mean Square Error.

Main Results:

• The logistic regression model achieved an accuracy of 0.809, an F1-score of 0.881, and an AUC of 0.825 when using WBFS-filtered features.
• This represents an approximate 12% improvement in predictive accuracy compared to models without feature selection.
• The RF model successfully ranked feature importance for classification results.

Conclusions:

• Electronic medical record data can effectively drive the accurate prediction of gestational diabetes.
• Feature selection techniques significantly enhance the performance of predictive models for GDM.
• Machine learning approaches offer a promising avenue for improving clinical diagnosis and patient outcomes in obstetrics.