Milk Composition Is Predictive of Low Milk Supply Using Machine Learning Approaches.

Area of Science:

• Biochemistry
• Data Science
• Pediatrics

Background:

• Low milk supply stems from genetic, endocrine, and removal frequency factors impacting mammary gland function.
• These factors can alter milk composition, affecting infant nutrition.
• Identifying low milk supply early is crucial for intervention.

Purpose of the Study:

• Investigate milk composition differences between mothers with low and normal milk supply.
• Develop predictive machine learning (ML) models for low milk supply identification.
• Integrate milk composition with maternal and infant characteristics for enhanced prediction.

Main Methods:

• Milk production measured via test-weigh method over 24 hours.
• Milk components analyzed in 58 low supply (<600 mL/24h) and 106 normal supply (≥600 mL/24h) mothers.
• ML algorithms, including deep learning and gradient boosting, used for model development.

Main Results:

• Deep learning and gradient boosting models showed superior performance.
• A comprehensive model with 14 milk components and other factors achieved 87.9% accuracy (AUC 0.917).
• A simplified clinical model retained 78.8% accuracy (AUC 0.794).

Conclusions:

• ML models can predict low milk supply with high accuracy.
• Combining milk composition with maternal/infant data offers a practical identification approach.
• Early identification facilitates timely interventions to support breastfeeding and infant nutrition.