Incorporating calendar effects to predict influenza seasonality in Milwaukee, Wisconsin

Abstract

Social outings can trigger influenza transmission, especially in children and elderly. In contrast, school closures are associated with reduced influenza incidence in school-aged children. While influenza surveillance modelling studies typically account for holidays and mass gatherings, age-specific effects of school breaks, sporting events and commonly celebrated observances are not fully explored. We examined the impact of school holidays, social events and religious observances for six age groups (all ages, ⩽4, 5–24, 25–44, 45–64, ⩾65 years) on four influenza outcomes (tests, positives, influenza A and influenza B) as reported by the City of Milwaukee Health Department Laboratory, Milwaukee, Wisconsin from 2004 to 2009. We characterised holiday effects by analysing average weekly counts in negative binomial regression models controlling for weather and seasonal incidence fluctuations. We estimated age-specific annual peak timing and compared influenza outcomes before, during and after school breaks. During the 118 university holiday weeks, average weekly tests were lower than in 140 school term weeks (5.93 vs. 11.99 cases/week, P < 0.005). The dampening of tests during Winter Break was evident in all ages and in those 5–24 years (RR = 0.31; 95% CI 0.22–0.41 vs. RR = 0.14; 95% CI 0.09–0.22, respectively). A significant increase in tests was observed during Spring Break in 45–64 years old adults (RR = 2.12; 95% CI 1.14–3.96). Milwaukee Public Schools holiday breaks showed similar amplification and dampening effects. Overall, calendar effects depend on the proximity and alignment of an individual holiday to age-specific and influenza outcome-specific peak timing. Better quantification of individual holiday effects, tailored to specific age groups, should improve influenza prevention measures.

An analecta of visualizations for foodborne illness trends and seasonality

“…While the use of NBHR models, specifically the inclusion of trigonometric harmonic oscillations, is similar to existing works on foodborne illnesses, these studies often incorporate harmonic oscillators only to adjust for or remove seasonal oscillations [47][48][49][50][51][52][53][54][55] . We have extended the use of harmonic terms and develop the tools to estimate peak timing and amplitude 8,29,30 . The developed δ-method provides a systematic calculation of confidence intervals for peak timing and amplitude estimates based on the results of harmonic regression models.…”

“…Measures of uncertainty enable formal testing and comparisons across diseases in the same location or locations for the same disease. In our previous works, we have demonstrated the broad utility of the δ-method and applications of peak timing and amplitude estimation in the context of epidemiological studies 6,29,[56][57][58][59][60][61][62] .…”

“…We estimated trend and seasonality characteristics using Negative Binomial Harmonic Regression (NBHR) models, which are commonly used to analyse count-based time series records with periodic fluctuations 29 – 31 . These models include harmonic terms representing sine and cosine functions, which allow us to fit periodic oscillations.…”

“…The regression parameters for these harmonic terms serve as a base for estimating important characteristics of seasonality: when the maximum rate occurs (peak timing) and the magnitude at that peak (amplitude). We calculated peak timing, amplitude, and their confidence intervals from NBHR model coefficients using the δ-method, which allow us to transform the regression coefficients of the model to seasonality characteristics based on the properties of the basic trigonometric functions (Supplementary Table S2 ) 29 , 30 . To estimate annualized seasonality characteristics, we applied a NBHR model for each study year and location with the length of the time series set to 12 to represent the months of the year.…”

Effects of Data Aggregation on Time Series Analysis of Seasonal Infections

“…The definition of a week (e.g., when does it start), of a month (how many days are in a month), and of a season (start date and duration) could significantly vary based on the temporal reference selected and study characteristics, such as geographic location. Furthermore, it is also important to consider during data aggregation additional local calendar characteristics, such as non-working days and holidays, which may impact data collection and analysis [ 28 ]. The effects of calendar characteristics during data aggregation are rarely discussed in epidemiological literature even though they have been shown to impact model results [ 29 , 30 ].…”