Influenza-associated mortality in the United States: Spatio-temporal patterns related to climate and virus subtype.

The seasonality of influenza in temperate countries is recognized, but regional differences in influenza-related mortality patterns are poorly understood. Such knowledge could enhance our understanding of influenza epidemiology and improve epidemic prediction and prevention. To investigate these patterns, the monthly percentage of deaths attributable to pneumonia and influenza among elderly aged ≥65 years in the continental United States, 1968-1998, was analyzed. Spatial autocorrelation was demonstrated (p = 0.001), with several western states forming a focus of high mortality. Divisive cluster analysis of state-level mortality identified broad patterns that generally corresponded to climate regions. In transmission seasons dominated by virus subtype A(H3N2), mortality was synchronized at a high and consistent level across the nation (mean correlation = 0.56, standard deviation [SD] = 0.134). However, when subtype A(H1N1) or type B dominated, the average synchrony was lower (mean correlation = 0.23, SD = 0.058) and weakened with distance. These novel analyses suggest large-scale environmental drivers might affect influenza mortality. Spatio-temporal mortality patterns were further studied in relation to global and regional climatic variables. La Nina conditions were associated with increased influenza mortality in six of nine climate regions, mostly during seasons dominated by subtype A(H3N2). Incorporating particular synoptic air masses into regression models significantly improved the ability to predict influenza mortality in the West, Northwest, and West North Central climate regions. Air mass transitions, not just air mass types, were important. Associations of La Nina and synoptic air masses with mortality hold promise for developing an influenza early warning system. The potential for estimating climatic effects on influenza A epidemic patterns was assessed using deterministic compartmental transmission modeling. Susceptible-Exposed-Infectious-Removed-Susceptible (SEIRS) models were created for two regions with four immunity levels each to represent viral drift and waning immunity. The two regions were given different internal dynamics (potentially attributable to climate). Population movement between regions was introduced, but transportation was never shown to overcome intrinsic dynamics. By giving immunity a more realistic and complex role than that of previous models, we were able to determine that long-term, within-region dynamics rather than population movement may account for temporal patterns of infection.Ph.D.Health and Environmental SciencesPublic healthUniversity of Michigan, Horace H. Rackham School of Graduate Studieshttp://deepblue.lib.umich.edu/bitstream/2027.42/125089/2/3186639.pd