2.6 Summary measures and statistical analysis
We estimated the overall pooled prevalence of NTS and the prevalence of antimicrobial resistance in poultry samples with a meta-analysis of proportions using the Freeman-Tukey double arcsine transformation with 95% exact confidence intervals (95% CI) (Chaidez-Ibarra et al., 2021). Due to expected heterogeneity across studies, we defined a prioria random effects model (DerSimonian & Kacker, 2007). To estimate the prevalence of NTS: first, we performed independent meta-analyses according to the type of sample (birds, products, and environmental) with subgroup analysis aggregating the studies by the level of sampling (individual or flock/collective), and second, we performed independent meta-analysis according to the type of sample aggregating the studies by country (considering both individual and flock/collective data). To estimate the prevalence of antimicrobial resistance, we performed independent meta-analyses according to the number of antibiotics (1, 2-3, or ≥4) to which the NTS isolates were resistant and aggregated the studies by the type of sample. The overall effect of the model was assessed with the z statistic assuming an effect size = 0, whereas the Cochran’s Q statistic (X 2 test) was used to assess significant heterogeneity across trials and theI 2 statistic was used to measure the proportion of variation in the effects caused by heterogeneity in the true effects rather than sampling error (Diaz et al., 2019; Romo-Barron et al., 2019). All analyses were performed on Stata 12 (StataCorp, TX, USA) and the graphs were constructed on Prism 9 (GraphPad Inc., CA, USA). In all the cases, we considered a value of p < 0.05 as significant.