Flu and similar respiratory diseases start and peak at different times in different geographical locations. In 2009 in the United States, the H1N1 flu first started in August in the Southeast, as schools opened, and then moved northward. H1N1 vaccine shipments did not start to arrive until October. By that time, the flu waves were almost completely concluded in the Southeast, but were just starting in many northern states.
The CDC allocation policy deployed vaccine doses to states in direct proportion to a state’s census population, independent of the status of the flu wave in each state. By that method, many vaccine doses were sent to the Southeast in October and November 2009, when they were much less effective, and many susceptible people in other regions — where the flu wave had not yet run its course — were without vaccines.
MIT researchers have treated vaccine allocation as a decision problem. Each week as new vaccine shipments arrive, decision makers need to determine where to direct the new doses to be most effective. A new paper addresses this problem, coauthored by Anna Teytelman Ph.D '12 and Richard Larson, the Mitsui Professor of Engineering Systems at MIT. Recently published online by the journal Service Science, the paper builds on previous research and derives an alternative way to deploy vaccine doses as they arrive during a pandemic.
The idea is to estimate the status of the various flu waves on a weekly basis, always updating the knowledge of flu progression in each region. Then the method uses mathematical models of the infectious disease to project forward in time, in order to estimate the number of potential infections that could be averted by allocating a given number of vaccine doses to each region. The derived “data-driven vaccine allocation algorithm,” in its pure form, then recommends allocating vaccine doses to the respective regions in a way that will prevent the maximum number of infections nationwide. The calculations are updated each week, as new data arrive, and new allocations are suggested. The authors suggest that the agency in charge of vaccine allocations might use a weighted hybrid approach, based partly on census populations and partly on this new modeling allocation technique.
Applying the allocation algorithm’s pure form within a simulation model to recreate the 2009 H1N1 flu, the authors estimated a 31 percent reduction in cases of the fluin contrast to a vaccine allocation policy that relies only on population. This corresponds to roughly 7 million flu-infected people who in 2009 would not have been infected had the proposed new scheme been in place.
The two competing objectives might be called “maximize equity” and “maximize effectiveness,” the first corresponding to allocation based on population and the second based on number of infections averted. One may visualize the two competing objectives using an old-fashioned fire bucket brigade set-up using buckets of water to fight ongoing fires in a number of contiguous neighborhood houses.
In the “effectiveness” approach, one seeks to apply the delivered water to burning houses in order to minimize the expected fire damage to the entire neighborhood. Each bucket of water, when delivered to the zone of the fires, is applied to that fire or fires so as to have the most beneficial effect in terms of minimizing the damage.
In the “equity” approach, water is deployed according to house size. That is, buckets of water are administered to houses in direct proportion house size only and independent of the status of the fire in each house — even if the fire had already been extinguished.
The savings to be achieved by the proposed new “effectiveness” method depend strongly on the timing of vaccine shipments. If, for example, 100 million vaccine doses arrive before any flu waves start, then the allocate-by-population method should work. If, however, the vaccines arrive after all the flu waves have run their course, then no allocation method would be successful.
The newly derived method is most advantageous when the vaccine shipments start to appear after the flu waves have started. With present flu vaccine manufacturing techniques in the U.S., any new (“novel”) infectious virus usually requires six to twelve months from identification and isolation of the virus until an immunization vaccine becomes available. Thus, it is likely — as in 2009 with the H1N1 swine flu pandemic — that vaccine shipments will start to occur only after the flu waves have started.
Research could help prevent millions of people from becoming infected with the next pandemic flu
Publication Date:
