Got (Optimal) Milk? Pooling Donations in Human Milk Banks with Machine Learning and Optimization.

This article focuses on a data-driven framework combining machine learning and optimization to improve the pooling of human donor milk at nonprofit milk banks lacking resources to measure macronutrient content. Developed in collaboration with the Rogers Hixon Ontario Human Milk Bank (RHOHMB), the framework predicts fat and protein levels in individual milk donations using routinely collected data and optimizes pooling decisions to meet clinical macronutrient targets while adhering to operational constraints. Simulation experiments demonstrate that this approach outperforms current heuristic pooling practices in pass rates and consistency, and a year-long trial implementation at RHOHMB showed a 31% increase in pools meeting fat and protein thresholds alongside a 60% reduction in recipe creation time. The study provides theoretical guarantees on solution quality despite prediction uncertainty and highlights the framework’s adaptability to other milk banks, contributing to healthcare operations management by illustrating the feasibility and benefits of integrating predictive analytics with optimization in clinical settings.