Given substantial regional differences in absolute humidity across the US and our understanding of the relationship between absolute humidity and influenza, we may expect important differences in regional seasonal influenza activity. Here, we assessed cross-seasonal influenza activity by comparing counts of positive influenza A and B rapid test results during the influenza season versus summer baseline periods for the 2016/2017 and 2017/2018 influenza years. Our analysis indicates significant regional patterns in cross-seasonal influenza activity, with relatively fewer influenza cases during the influenza season compared to summertime baseline periods in humid areas of the US, particularly in Florida and Hawaii. The cross-seasonal ratios vary from year-to-year and influenza type, but the geographic patterning of the ratios is relatively consistent. Mixed-effects regression models indicated absolute humidity during the influenza season was the strongest predictor of cross-seasonal influenza activity, suggesting a relationship between absolute humidity and cross-seasonal influenza activity. There was also evidence that absolute humidity during the summer plays a role, as well. This analysis suggests that spatial variation in seasonal absolute humidity levels may generate important regional differences in seasonal influenza activity and dynamics in the US.

Influenza is a respiratory virus that causes significant morbidity and mortality throughout the world every year. Seasonal epidemics of influenza occur in the late fall and winter in the United States annually, but there are variations in its timing from year to year. Further, although the timing of epidemic waves in the United States are similar, there is variation between different populations. It is not well understood why these differences exist. Understanding the spatial and temporal variation in the timing of influenza is important because it shapes our understanding of preventive actions that can be taken to limit the spread of the virus. Past studies that have examined the timing of influenza have been limited by the fact that they have used influenza-like illness (ILI) as an indicator of influenza. ILI has traditionally been the conventional indicator of influenza because the illness does not present unique symptoms. As such, spatial and temporal variation in the relative timing of influenza A, influenza B, and ILI have not been investigated extensively. Additionally, there has been concern raised about implications of the imprecise nature of ILI in treating patients that it is believed may have influenza. This study addressed this gap and concern by utilizing influenza-specific data from clinics and hospitals throughout the United States to evaluate spatial variation in the timing of influenza across the United States in the 2016-17 influenza season. Results from influenza rapid tests were aggregated by urban area as a means of evaluating associations between epidemic timing and independent variables in different locations. The timing of influenza A and B epidemics was tested for spatial autocorrelation and incorporated in regression models to identify potential relationships between epidemic timing and several variables (including dew point, temperature, population, population density, and cumulative seasonal vaccination rates). Forward iv stepwise regression was then conducted to identify a set of variables that may be best suited to explain the timing of these milestones, and spatial lag regression was conducted to account for spatial autocorrelation in these variables. This analysis indicated that higher average dew point and temperature and greater population and population density were both associated with earlier epidemic beginnings and later epidemic endings, while higher cumulative seasonal vaccination rates were associated with earlier epidemic endings for influenza A and B. Forward stepwise regression yielded models that generally differed for each epidemic milestone and type of influenza, indicating that different sets of variables might be best suited to explain different milestones of epidemics. Spatial lag regression improved model fit for the forward stepwise models for which there was residual spatial autocorrelation. This is one of the first studies to evaluate the timing of different points within an epidemic. The techniques I used to study timing are well-suited for the study of futu...