Prediction of post-delivery hemoglobin levels with machine learning algorithms

Area of Science:

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

Background:

• Predicting postpartum hemorrhage (PPH) before delivery is critical for timely interventions and improved patient outcomes.
• Existing methods may not fully capture the complexity of predicting PPH risk using readily available clinical data.
• Machine learning (ML) offers a potential avenue for developing more accurate predictive models.

Purpose of the Study:

• To utilize machine learning (ML) with pre-labor clinical data and laboratory measurements to predict postpartum hemoglobin (Hb) levels in uncomplicated singleton pregnancies.
• To identify key predictors of post-delivery Hb levels.
• To develop and validate an ML model for predicting indirect measures of PPH.

Main Methods:

• Retrospective analysis of delivery databases from two academic care centers, including 1974 women.
• Feature selection using Elastic Net regression and Random Forest algorithms to identify significant pre-delivery predictors.
• Training and evaluation of various ML algorithms, including artificial neural networks (ANN), to predict 24-hour post-delivery Hb levels.

Main Results:

• Key predictors for post-delivery Hb included parity, gestational age, pre-delivery hemoglobin, fibrinogen levels, and pre-labor platelet count.
• Artificial Neural Network (ANN) demonstrated the highest accuracy with a Root Mean Squared Error (RMSE) of 0.62.
• A web-based calculator was developed based on the ANN model: https://predictivecalculators.shinyapps.io/ANN-HB.

Conclusions:

• Machine learning models can accurately predict post-delivery hemoglobin levels, serving as indirect predictors of postpartum hemorrhage (PPH).
• The developed ML model and web-based calculator can be integrated into healthcare systems to support clinical decision-making.
• Further validation with diverse, population-based samples is recommended to enhance model generalizability.