Influence of age and body condition on astrovirus infection of bats in Singapore: An evolutionary and epidemiological analysis

Mendenhall, Ian H.; Skiles, Maggie M.; Neves, Erica Sena; Borthwick, Sophie A.; Low, Dolyce H.W.; Liang, Benjamin; Lee, Benjamin P.Y.-H.; Su, Yvonne C.F.; Smith, Gavin J.D.

Individual prevalence from pools was estimated using a frequentist approach that assumed a fixed pool size and a perfect detection and a Bayesian approach with a Gibbs sampler model with AusVet EpiTools [35] . Details of the criteria used in the analysis are available in the supplementary materials. These were compared to individual prevalence rates on four dates within the sampling period. Multiple logistic regression (MLogR) was used to examine the effects of species, catch/ roost location, sex, age, and bat body condition [36] on astrovirus infection. The scale of each predictor was determined by performing exploratory data analysis. A variance inflation factor (VIF) was calculated to assess multi-collinearity (inter-relatedness of the covariates). Hierarchical logistic models were tested to investigate significant predictors of astrovirus infection. Likelihood ratio tests (LRT) were generated to compare full, extended MLogR to more constrained models and assess whether the inclusion of interaction terms within a larger model were significant. Akaike Information Criterion (AIC) values were used to rank models and to determine the optimal set of predictors. Odds ratios were computed for the explanatory variables in the resulting model. Forward and backward stepwise selection based on AIC and a 5% pvalue cut-off were utilized as an aid to choose predictors for the final model. All covariates were considered for elimination. Based upon the statistical output of the stepwise processes, likelihood ratio tests outcomes, the variance inflation factor procedure, confounding assessments, cross tabulations of the outcome and variables, and prior knowledge regarding the subject, the final model was fit. The final model included 1) bat age and 2) bat species. The Pearson's chi-squared goodness-of-fit test (H 0 : the model M 0 fits) was used to determine the overall fit of the final model. The p-value was computed at P: 0.27, indicating a well-fit model. All analyses were carried out using Stata version 12.1 [37] .