Climate factors influence seasonal influenza activity in Bangkok, Thailand

This study had several limitations. Influenza surveillance was derived from one major hos- pital in western Bangkok and may not have represented the influenza activity of the entire city. Our surveillance was passive in nature because samples were submitted to us rather than actively sought, therefore we did not use the time of viral exposure for analysis. The over- whelming majority of the samples came from in-patient settings, so nosocomial influenza was not considered. Although we had influenza data prior to 2011, these were excluded due to fewer sample size and possible bias from the influenza pandemic of 2009. From our experi- ence, we presumed that only influenza A(H1N1)pdm09 and A(H3N2) viruses comprised the overwhelming majority of influenza A virus in circulation, but it is possible that other less fre- quent subtypes were missed. Due to latitude and climate differences, influenza incidence in northern and southern Thailand may be different and this would benefit from future studies. In addition, daily or weekly climate factor values, when available, would be more useful in improving modeling accuracy. Despite these limitations, our results suggest that certain cli- mate factors do influence influenza activity and further support the existing schedule for influ- enza vaccination prior to the rainy season. The time series analysis of influenza virus could potentially supplement a more accurate forecasting of future seasonal influenza activity when combined with an active epidemiological surveillance. Countries lacking the latter but share

similar geographical and climatological factors may benefit from the model-based disease predictions. This and other time series analysis, especially when performed on the daily climate factors, could potentially improve the foundation for enhanced modeling of disease seasonality. Supporting information S1 Fig. The original and log10 transformed time series plots of the monthly confirmed influenza-positive cases. Original time series (left panels) and transformed time series (right panels) are shown for influenza A(H1N1)pdm09 (A and B), influenza A(H3N2) (C and D), influenza B (E and F), and all influenza viruses (G and H). (TIF) S2 Fig. Plots of residuals of selected forecasting model, autocorrelation function (ACF), and the normal distribution of the residuals. The left column represents the residuals of uni- variate models, while the right column represents the residuals of multivariate models. (A and B) influenza A(H1N1)pdm09, (C and D) influenza A(H3N2), (E and F) influenza B, (G and H) all influenza viruses. (TIF) S3 Fig. Plots of residuals of a selected forecasting model, autocorrelation function (ACF), and partial autocorrelation function (PACF). The left column represents the residuals of univariate models, while the right column represents the residuals of multivariate models. (A and B) Influenza A(H1N1)pdm09 virus. (C and D) Influenza A(H3N2) virus. (E and F) Influ- enza B virus. (G and H) All influenza viruses. (TIF) S1 Table. Comparison of the mean meteorological variables between no influenza activity month and influenza-active month. (DOCX)

S2 Table. Cross-correlation between individual and all influenza viruses with climate fac- tors at different lag times. (DOCX) PLOS ONE Climate factors influence seasonal influenza activity in Bangkok, Thailand PLOS ONE | https://doi.org/10.1371/journal.pone.0239729 September 29, 2020 12 / 15