2.4 Study selection process and data extraction
To select the studies to be included in the narrative synthesis, one reviewer first removed the duplicates automatically from the EndNote library and then revised the records manually. Next, the same reviewer performed the screening of all the remaining records by checking whether the publications were related to the reviewed subject, first based on the title, and second, based on the abstract. Once the screening was completed, the selected publications were retrieved in full text to assess their eligibility for final inclusion in the systematic review and meta-analysis. To account for this, two independent reviewers used a standardized questionnaire that was based on the eligibility criteria (Supplementary Table 2 ) and was piloted in 10% of the publications selected at random. Each time there was a discrepancy between the reviewers, a third reviewer served as referee until consensus was reached.
A single reviewer used a standardized questionnaire to extract data from all the selected publications. Prior to be used, the extraction sheet format was piloted in 10% of the studies selected at random. The extracted data included a) the main characteristics of the publication (first author, country, publication year, and study design), b) the main characteristics of the assessed population (poultry subgroup, sex, and age), c) the type of sample reported (birds, products and subproducts, or environmental), d) the diagnostic technique for NTS, and e) the reported outcomes (prevalence, serotyping, or antimicrobial resistance). For the publications that reported both data on the individual and flock/collective level, we selected for extraction only data at the individual level to avoid duplication of estimates for a single publication and because we found less variation in the estimations made at the individual when compared to the flock/collective level. Besides, for the publications that assessed more than one type of sample, each one of them was extracted separately and counted as one independent study within a single publication. All the extracted data were registered into an Excel spreadsheet (Microsoft, USA) and a codebook was created to facilitate its handling.