Parameterizing a dynamic influenza model using longitudinal versus age-stratified case notifications yields different predictions of vaccine impacts

Dynamic transmission models of influenza are sometimes used in decision-making to identify which vaccination strategies might best reduce influenza-associated health burdens. Our goal was to use laboratory confirmed influenza cases to fit model parameters in an age-structured, two-type (influenza A/B) dynamic model of influenza. We compared the fitted model under two fitting methodologies: using longitudinal weekly case notification data versus using cross-sectional age-stratified cumulative case notification data. The longitudinal data came from a Canadian province (Ontario) whereas the cross-sectional data came from the national level (all of Canada). We find that the longitudinal fitting method provides best fitting parameter sets that have a higher variance between the respective parameters in each set than the cross-sectional cumulative case method. Model predictions- particularly for influenza A-are very different for the two fitting methodologies under hypothetical vaccination scenarios that expand coverage in either younger age classes or older age classes: the cross-sectional method predicts much larger decreases in total cases under expanded vaccine coverage than the longitudinal method. Also, the longitudinal method predicts that vaccinating younger age groups yields greater declines in total cases than vaccinating older age groups, whereas the cross-sectional method predicts the opposite. We conclude that model predictions of vaccination impacts under different strategies may differ at national versus provincial levels. Finally, we discuss whether using longitudinal versus cross-sectional data in model fitting may generate further differences in model predictions (above and beyond population-specific differences) and how such a hypothesis could be tested in future studies.