LightGBM: an efficient and accurate method for predicting pregnancy diseases.

Area of Science:

• Reproductive Medicine
• Machine Learning in Healthcare
• Clinical Diagnostics

Background:

• Machine learning (ML) is increasingly used for disease prediction, yet accurate models for pregnancy complications remain limited.
• Existing ML models for gestational diseases often use simpler algorithms and have suboptimal prediction accuracy.
• Early prediction of pregnancy-induced hypertension (PIH) and intrahepatic cholestasis of pregnancy (ICP) is crucial for maternal health.

Purpose of the Study:

• To compare the performance of nine common machine learning methods for predicting PIH and ICP.
• To identify the most effective ML model for early detection of these two common pregnancy diseases.

Main Methods:

• A retrospective study analyzed data from 320 pregnancies between 10-19 weeks of gestation.
• Nine distinct machine learning algorithms were evaluated for their ability to predict subsequent PIH (149 cases) and ICP (250 cases).
• Model performance was assessed using eight standard evaluation metrics.

Main Results:

• The light Gradient Boosting Machine (lightGBM) demonstrated the highest predictive accuracy among the tested methods.
• lightGBM achieved an Area Under the Curve (AUC) of 81.72% for PIH prediction with 70.59% sensitivity.
• For ICP prediction, lightGBM achieved an AUC of 95.91% with a sensitivity of 97.91%.

Conclusions:

• The lightGBM model provides an effective tool for first-trimester screening of PIH and ICP.
• This ML model can aid clinicians in diagnosis and treatment planning for pregnant women.
• Further research should validate lightGBM's efficacy in daily practice and for other gestational diseases like pregnancy diabetes.