Climate factors influence seasonal influenza activity in Bangkok, Thailand

Laboratory testing of influenza virus From January 2010 to December 2018, 30,852 nasopharyngeal or throat swab samples from patients with influenza-like illness (ILI) from Bangpakok 9 International Hospital (latitude 13.75˚N and 100.55˚E longitude) in Bangkok were analyzed as part of an ongoing influenza surveillance [23, 28]. We defined ILI as fever >38˚C accompanied by cough and/or sore throat. Respiratory specimens arrived with limited patient information and included only age and gender. Samples were tested by using real-time reverse-transcription polymerase chain reaction to identify influenza A(H1N1)pdm09, A(H3N2) and influenza B virus as previously described [29]. Meteorological data Data on the monthly average temperature (˚C), rainfall (cm3), and relative humidity (%) from January 2010 to December 2018 were available from Wolfram Alpha database [27]. They were recorded at VTBD (Don Mueang International Airport) in Bangkok (latitude 13.91˚N and 100.6˚E longitude). Data analysis Monthly incidence of influenza was compiled based on the sample collection date. Percentages of monthly influenza virus-positive samples relative to all ILI samples each month were deter- mined. Univariate analysis of the meteorological variables including average temperature, rela- tive humidity and rainfall was done on data between influenza-active months and months without influenza activity using Mann-Whitney U test. An influenza-active month was defined as a month in which the proportion of the confirmed influenza cases was equal to or greater

than a fixed threshold of 10% [10, 30]. This was calculated by dividing the monthly confirmed influenza cases by each year’s total number of influenza virus-positive samples [31]. To examine the seasonality of influenza, we applied a generalized linear model (GLM) with Gaussian and Poisson distribution on the monthly values of climate variables and laboratory- confirmed influenza cases, respectively [32]. A seasonal model in the waveform with a period of 12 months was then established with the estimated parameters from the GLM [11, 33]. Cross-correlation analysis and forecasting model We initially examined the relationships between each influenza virus (sub)type and all influ- enza activity with respect to meteorological variables. The monthly confirmed cases of influ- enza A(H1N1)pdm09, influenza A(H3N2), influenza B, and all influenza viruses were considered as dependent variables. Monthly values of mean temperature, relative humidity, and accumulated rainfalls were considered as independent variables. Cross-correlation analy- sis was conducted to assess the relationship between virus incidence and various meteorologi- cal factors, and for relationships among influenza A(H1N1)pdm09, influenza A(H3N2), and influenza B virus with different lag times (defined as the difference between time of maximum climate factor values and time of maximum influenza incidence) of between 1 to 4 months [26]. We considered four forecast models for individual and all influenza viruses using depen- dent variables from influenza A(H1N1)pdm09 (model 1), influenza A(H3N2) (model 2), influ-

enza B (model 3), and all influenza viruses (model 4). For each model, we also considered potential association among different influenza viruses. Climate factors were considered as independent variables or proxy predictors in forecasting future influenza activity. Time series analysis was conduct to identify the best-fit model for each influenza (sub)type and for all PLOS ONE Climate factors influence seasonal influenza activity in Bangkok, Thailand PLOS ONE | https://doi.org/10.1371/journal.pone.0239729 September 29, 2020 3 / 15