A team of researchers at the University of Toronto is leveraging machine learning to optimize the macronutrient content of pooled human donor milk recipes.
Researchers — led by timothy chanProfessor of. Department of Mechanical and Industrial Engineering Faculty of Applied Science and Engineering – Introducing data-driven optimization models In a new paper published in a magazine Manufacturing and system operations management.
Chan and his team worked with Mount Sinai Hospital. Rogers Hixon Ontario Breast Milk Bank — We provide donor milk to hospitalized premature and sick babies across Ontario — Deborah O’connorProfessor of. Department of Nutritional Science Currently enrolled at the Temerti Faculty of Medicine.
“For a variety of reasons, many hospitalized infants do not have an adequate supply of human milk. It helps and can save lives.” Sharon UngerShe is a neonatologist, medical director of the Rogers Hixon Ontario Breast Milk Bank, and a professor in the nutritional science department at Temerty Medicine. medicine.
“The new program developed by Dr. Chan helps ensure that each batch of human donor milk meets the protein and calorie needs of preterm infants.”
Currently, many milk banks, including Mount Sinai Milk Bank, rely on individual decisions when pooling donor milk. This poses a major challenge in producing consistent donor milk products with sufficient macronutrients for premature and sick babies in neonatal intensive care units.
“Creating these recipes without a defined method takes a lot of time,” says Chan, who is also U of T’s vice provost and associate provost for strategic initiatives.
“While some studies have shown that milk from donors with early postpartum life tends to be richer in protein, our approach better predicts the actual macronutrient content and helps milk bank personnel to We’ll help you make the right pool decisions.”
Given that milk banks are often non-profit organizations operating on lean budgets, a low-cost alternative to obtaining consistent, nutritionally balanced products could be useful across the sector. There is a gender.
A device known as a breast milk analyzer allows you to measure the exact macronutrient content of your breast milk. Each milk bank milk sample is expensive and requires extensive regulatory approval, so it is only used by half of all milk banks in North America. Furthermore, analyzing all donations is a labor- and resource-intensive and expensive task.
“Our data-driven framework circumvents the need for devices to analyze donor milk by using artificial intelligence models to predict the macronutrient content of each donation.” says Mr. Rachel WongShe received her master’s degree from U of T Engineering last year and is the principal investigator on this study.
“Furthermore, by using an optimization model to select which donations to pool together, we can increase consistency in the macronutrient content of donor dairy products.”
The multi-phase study included a one-year implementation trial at Rogers-Hixson Milk Bank designed to test whether an AI-informed model could help close the gap.
In the first phase, the researchers collected the data needed to create a machine learning model to predict the macronutrient content of the pooled recipes, and then determined the macronutrient requirements, i.e. We designed an optimization model to create recipes based on protein and protein levels. fat.
The team then created a simulation model to test the method before embarking on experiments in milk banks. The experiment took place over his 16-month period in 2021 and 2022.
“Our research was conducted during regular business hours at a milk bank, rather than in a controlled environment, so there were many unforeseen challenges we had to adapt to,” Wong says.
“During the COVID-19 pandemic, donation amounts have fluctuated based on state restrictions. During the lockdown period, the number and volume of donations increased to an unprecedented extent.
“We also had to adapt the AI decisions that had already been proposed to ensure compliance with the milk bank’s operational protocols.”
The final stage of the study began by observing milk bank operations for six months and measuring levels of fat, protein and bacteria in the pooled recipes.
Over the next six months, the milk bank used a data-driven optimization framework to create a pooled milk recipe. At the end of the year, researchers compared the optimized recipes to previous recipes to assess which recipes met macronutrient goals.
“We found that the recipes we pooled simultaneously met protein and fat criteria up to 75% more often, without compromising other factors such as increased bacteria,” says Chan. “And it took him 60% less time to make the recipe.”
The team’s optimized recipe also has benefits for preterm and sick babies. Premature or sick babies have underdeveloped digestive systems, so it’s especially important to make sure the milk they consume doesn’t contain too much protein or fat.
Chan’s team is currently working on expanding the study to measure other nutrients in human donor milk and see if the model can optimize them.research wins 2023 Pierce Scala Best Paper Award Award from the Institute of Operations Research and Management Sciences (INFORMS) and the Excellence in Quality and Safety Award from Sinai Health.
“Our ultimate goal is to show that our tool can be applied to other milk banks as well,” says Chan. “We want to design a system that can connect to hospital systems and optimize recipes in a sustainable way for milk bank staff.”
Wong said the entire team is grateful to everyone who made this project possible.
“We couldn’t do this without all of that.” “It’s the mothers who donate to the milk bank and the staff who work incredibly hard to provide donor milk to infants in Ontario and beyond,” she says.
“We hope this study provides a framework to help milk banks across North America increase the consistency of macronutrient content in donor dairy products. to see the downstream impact of improved infant growth and developmental outcomes.”
