The optimal composition of influenza vaccines subject to random production yields

The Vaccine and Related Biologic Products Advisory Committee meets at least once a year to decide the composition of seasonal influenza vaccine in the United States. Past evidence suggests that the committee could use a more systematic approach to incorporate observed information and to quantify the risks associated with different options. There are two key trade-offs involved in this decision. First, if the Committee decides to retain the current vaccine composition instead of updating to a new one, there is lower uncertainty in production yields, but the current vaccine could be less effective if a new virus strain spreads. Second, if the Committee decides early with less information, then manufacturers have more production time, but the reduced information increases the risk of choosing a wrong strain. We derive an optimal dynamic policy for this decision. Because of the greater uncertainty in production yields of new vaccines, the optimal thresholds are neither symmetric between retaining and updating the composition nor monotonic over time. We apply our model to past decisions using parameter values estimated from a historical case. Our analysis shows that the dynamic optimal policy can significantly improve social welfare.