Seasonality of respiratory viruses and bacterial pathogens

Choe, Young June; Smit, Michael A.; Mermel, Leonard A.

The null hypothesis was no seasonal cross-correlation between respiratory viral activity, antibiotic prescriptions filled, weather variables and detection of different bacterial pathogens. We constructed a longitudinal incidence rate using a monthly dataset which included respiratory viruses detected, antibiotic prescriptions filled, and meteorological parameters (total precipitation and average temperature), as well as detected bacterial pathogens (C. difficile, MRSA, GNB, and S. pneumoniae) (Fig. 1) . A seasonal trend decomposition procedure, based on Locally Weighted Scatterplot Smoothing (STL), was conducted for each bacterial pathogen to assess for seasonality and trends associated with detection of respiratory viruses, antibiotic prescription, and meteorological parameters [13] . An additive decomposition model was used. To assess for a correlation between a time series and a given a number of lags, we measured the cross-correlation of X t and Y t + k for each month. Separate cross-correlation functions were applied to determine the specific bacterial pathogens and their highest correlation with detection of respiratory viruses, antibiotic prescriptions filled, or meteorological parameters on defined time lags. We calculated odds ratios (OR) with 95% confidence intervals, estimating risk of elevated incidence compared to annual average incidence. Due to the small numbers and multiple testing, the p-values should only be seen as descriptive measures. Analyses were performed using R (ver. 3.4.3; R Development Core Team, Vienna, Austria). Packages used were: forecast, TSA (time series analysis) [14] , and ASTA (applied statistical time series analysis) [15] .