Optimal Management of Donor Milk Banks

University Of Pittsburgh, Pittsburgh PA

Investigators

Abstract

Human breast milk is a dynamic, bioactive fluid that provides crucial nutritional and medicinal advantages to infants, especially those born prematurely. Breast milk contains hundreds of antibodies and other factors that protect babies from life-threatening neonatal infections and diseases. Many mothers, however, cannot provide adequate breast milk for their infants for a host of reasons including absent or insufficient lactation, the latter of which is especially acute for mothers of premature infants. Therefore, donated breast milk, collected and dispensed via non-profit milk banks, is the standard of care for infants whose mother's own milk is not a viable option. Despite the fact that over two million ounces are dispensed annually in the US, demand far outweighs supply. Moreover, milk banks involve many complex operational processes. Hence, this growing network of facilities is widely regarded as being in great need of support. This award aims to provide such support by optimizing the processing, storage, dispensing and prioritized allocation of this precious scarce resource. The outcomes of the research will benefit society through better medical outcomes, shorter hospital stays and saved lives. The specific objectives of this research project are to optimize how milk from different donors is blended together, how hospital orders and out-patient prescriptions are fulfilled and how recipients are prioritized. These decision-making processes will be formulated as large-scale multistage stochastic programs and will be calibrated using data from multiple sources. The blending decisions will be modeled as a multistage stochastic mixed-integer program, a class of problems that has a well-deserved reputation for being very difficult to solve, but exhibit appealing structure. Patient prioritization decisions will be modeled as a multistage stochastic convex program and an ambitious combination of these decision processes will be modeled as a multistage stochastic mixed-integer convex program. Overall, the main intellectual merits of the work lie in the timeliness and novelty of the application (i.e., milk bank operations) as well as the modeling and algorithmic challenges that will be overcome in solving the models.