“We’re giving them a structured way to think about the problem. We want to optimize the solution.”
The problem in this case is influenza, the flu, a commonplace, ever-mutating disease with symptoms ranging from sniffles to death. Providing the structure is Andrew J. Schaefer, the Noah G. Harding Chair and Professor in computational and applied mathematics (CAAM) at Rice University.
“Flu shot design has two critical aspects. First, the makeup of the shot influences the effectiveness of the vaccine. Second, the timing of the strain selections affects the production capacity. It’s a delicate balance,” Schaefer said.
The stakes are high. Annually since 2010, the U.S. Centers for Disease Control (CDC) estimates influenza in the U.S. has resulted in from 9.2 million to 60.8 million illnesses, between 140,000 and 710,000 hospitalizations and from 12,000 to 56,000 deaths.
Schaefer and his colleagues – Osman Y. Özaltın of North Carolina State University and Oleg A. Prokopyev of the University of Pittsburgh – have published their findings in the
INFORMS Journal on Computing published by the Institute for Operations Research and the Management Sciences: “Optimal Design of the Seasonal Influenza Vaccine with Manufacturing Autonomy.”
As Schaefer explains it, flu viruses mutate rapidly and often. To keep up, researchers each year update the composition of flu shots, which contain inactivated strains of the virus. The World Health Organization and the CDC recommend which strains to include in the shot based on global surveillance of the disease, and during February and March the U.S. Food and Drug Administration (FDA) decides on the final composition of the shot for the following flu season.
Vaccine manufacturers produce trivalent and quadrivalent shots, depending on the season’s predominant strains. The shot’s design directly affects the manufacturers’ profit margin and the vaccination benefits for the public.
Using powerful optimization methods, Schaefer and his colleagues sought to improve the accuracy of the decisions regarding which strains should be included in the vaccine. Waiting longer to determine the final composition of the vaccine, and observing which strains are emerging in other parts of the world, fine-tunes the accuracy of the selection, but results in an insufficient quantity of vaccine made in time for the start of flu season.
The model devised by Özaltın, Prokopyev and Schaefer considers multiple factors. “An important conclusion of our work,” Schaefer said, “is that decision-makers should balance prevalence, cross-immunity and ease of manufacturing. Our model gives a framework for understanding the trade-offs inherent in this important decision.”