resistancebank.org, an open-access repository for surveys of antimicrobial resistance in animals

Antimicrobials have dramatically reduced infectious disease mortality, but overuse in human medicine and animal production has driven rising antimicrobial resistance (AMR). Approximately 73% of antimicrobials are used in animals for prevention, treatment, and growth promotion, raising concerns for animal health and livelihoods, and for transmission of resistant bacteria from animals to humans. Monitoring AMR in animals is crucial to coordinate international actions. While high-income countries maintain systematic surveillance (e.g., EFSA, NARMS, CIPARS), LMICs—where meat demand and antimicrobial use are rapidly increasing—largely lack such systems. In the short term, point-prevalence surveys (PPS) can inform policymaking. A 2019 systematic review synthesized PPS to generate a first global, sub-national map of AMR in food animals and to highlight hotspots and data gaps. However, systematic reviews are time-consuming, must be updated frequently, and may miss data due to language barriers, paywalls, and restricted access. An open-access, real-time platform can facilitate data sharing, strengthen evidence bases, and empower local researchers, thereby improving the visibility of LMIC research and informing resource allocation.

The paper situates resistancebank.org within existing surveillance and data-sharing efforts. High-income regions operate systematic AMR surveillance in animals (EFSA in Europe, NARMS in the US, CIPARS in Canada), whereas LMICs generally lack such systems despite increasing antimicrobial usage and animal production. A 2019 systematic review collated PPS on foodborne pathogens to map AMR at sub-national scales in LMICs, identifying hotspots and poorly surveyed areas. The authors note the rise of online data-sharing platforms across scientific domains and in epidemiology (e.g., HealthMap, Malaria Atlas Project). In human AMR, platforms such as the ECDC Surveillance Atlas and CDDEP ResistanceMap exist but typically aggregate data at country level, lack granular open-access data, and do not allow external uploads. Genomic AMR platforms (e.g., Microreact, Nextstrain) are complementary but focus on genetic determinants rather than phenotypic resistance in animals. This landscape underscores the need for an open, granular, and updatable platform for animal AMR in LMICs.

Database construction: A systematic literature search (January 2019) across PubMed, Scopus, and Web of Science identified PPS on AMR in food animals from LMICs, targeting four AGISAR-recommended indicator bacteria. Titles/abstracts were deduplicated and screened; exclusions included books, meta-analyses, reviews, and surveys of sick animals per AGISAR guidelines. Additional gray literature (papers, theses, conference proceedings) was collected via field visits to five Indian veterinary schools. From each study, AMR surveillance-relevant data were extracted: sampling size, animal hosts, bacterial species, sampling coordinates, antimicrobial susceptibility testing (AST) results, antimicrobials tested, and number of isolates. Resistance rates were aggregated by location/host/bacteria combination. For geospatial analysis, only AGISAR-recommended antimicrobials were used. Variable definitions align with Table 1 (e.g., ISO3, coordinates, dates, species, sample origin, pathogen, compound, ATC code, resistance percentage, breakpoints, guidelines, remarks). Geospatial prediction: A summary metric, P50 (proportion of antimicrobials tested with resistance >50%), previously modeled at 10×10 km resolution for all LMICs using PPS, underpins the platform’s AMR map. P50 maps are updated annually. Software implementation: The platform is built in R (shiny) with JavaScript/CSS for UI; deployed on shinyapps.io (configured for ~2,500 concurrent users). Spatial data are displayed with leaflet; AMR maps are served as raster tiles (PNG) at 10 zoom levels, generated in QGIS and hosted on GitHub Pages (~2.8 million tiles). Remote data storage uses Dropbox (rdrop2) and AWS S3 (aws.s3). Submissions upload CSVs to an online folder; upon human validation, data merge into the central database. Bibliographic metadata are fetched via ROpenSci tools (e.g., europepmc, rcrossref). Country reports are generated via parameterized R Markdown (PDF), using country-specific data and functions. Data submission workflow: Users can submit PPS via an online form or Excel template. Automated validation corrects typos against controlled vocabularies and flags irreparable errors (e.g., resistance >100%). DOI validation pulls bibliographic details from NCBI/PubMed. Submissions appear as temporary (light blue) markers pending human review, which includes checking resistance rates and breakpoints, and contacting authors if needed. Upon approval, outputs (database and country reports) update near-real-time.

- Coverage and scale: 1,285 PPS from LMICs (2000–2019), yielding 22,403 resistance rates. Surveys conducted in 72/135 LMICs. - Pathogens: Escherichia coli (n=9,206; 41.09%), non-typhoidal Salmonella spp. (n=7,080; 31.60%), Staphylococcus aureus (n=4,828; 21.55%), Campylobacter spp. (n=1,290; 5.76%). - Geographic distribution: 61.4% of PPS from Asia, 24.1% Africa, 14.5% Central and South America; 49.41% of PPS from India, China, Brazil, and Iran. - Hosts and sample origins: Poultry (38.35%), cattle (37.56%), pigs (15.28%), sheep (8.81%). Sample origins: food products (55.66%), living/slaughtered animals (43.82%), drag swabs (0.52%). - Antimicrobials: Resistance rates for 143 antimicrobials across 37 chemical families; 13,163 (59%) of resistance rates correspond to AGISAR-recommended drug-pathogen combinations. - Platform outputs: Interactive P50 AMR map at 10×10 km resolution; downloadable PPS database (CSV), P50 raster (TIF), and country reports (PDF) with associated country-level data subsets. - Functionality: Data filtering by country, species, sample origin, pathogen, AGISAR alignment, and WHO medical-importance classes; map marker intensity reflects average resistance for selected class. - Technical capacity: ~2,500 concurrent users supported; ~2.8 million pre-rendered raster tiles for efficient map navigation; near-real-time updates after validated submissions. - Data availability and code: All data at resistancebank.org; source code on GitHub (hegep-eth/resistancebank.org).

The platform consolidates dispersed PPS evidence on AMR in animals in LMICs into an accessible, continuously updated resource. It is designed as a surrogate—not a replacement—for systematic surveillance, summarizing current knowledge while enabling growth through community data contributions. Locally, it can guide epidemiological investigations in areas of concern; globally, it supports funders (e.g., Gates Foundation, Fleming Fund, FAO, WOAH) in prioritizing interventions, particularly in hotspots (e.g., P50 > 0.4) and in evaluating stewardship or alternatives (vaccines, probiotics). Compared with traditional studies, the platform offers open-access downloading and uploading, diverse outputs (granular survey data, maps, policy-oriented reports), near-real-time updates, and a focal point for the AMR animal research community. It complements existing human AMR platforms (ECDC Atlas, ResistanceMap), which typically lack high spatial resolution, open granular data, and user-upload features, and genomic platforms (Microreact, Nextstrain), by providing phenotypic resistance information with fine geospatial detail. These advances make resistancebank.org relevant for evidence-based resource allocation and for catalyzing broader, systematic surveillance development.

resistancebank.org centralizes PPS-derived AMR data in food animals from LMICs, delivering fine-scale maps, downloadable datasets, and country reports, and enabling community submissions with validation. The platform enhances transparency, supports policy prioritization, and empowers LMIC researchers by overcoming access and publication barriers. Future directions include expanding linguistic and geographic data collection networks, integrating additional PPS sources, increasing automation while maintaining quality control, and reducing computational barriers to enable more frequent (ideally near-real-time) geospatial map updates. The platform is intended to complement and accelerate the development of comprehensive, systematic AMR surveillance in animals.

- Heterogeneity and comparability: Event-based animal surveillance spans diverse sampling contexts (living vs. dead animals, food products, outbreak investigations, regulatory sampling), strategies (random vs. convenience), farming systems, isolate counts, AST methods (diffusion vs. dilution), and aggregation levels (population vs. isolate). These factors complicate harmonization and interpretation of resistance rates. The platform captures metadata (sampling scheme, guidelines/breakpoints, QC strains) to aid interpretation. - Summary metric (P50) constraints: P50 depends on the number and selection of drugs tested, which vary across laboratories and methods; some surveys screen second-line drugs (e.g., imipenem) on subsets of isolates, potentially biasing P50 (sub-sampling occurred in 34/1,940 P50 estimates). P50 counts compounds rather than classes; class-level rates are provided separately. P50 should guide high-level prioritization where systematic surveillance is limited; class-specific comparisons are preferred for public health decisions. - Data collection intensity and language coverage: Supplemental field collection occurred primarily in India; literature searches spanned six languages (English, Mandarin Chinese, Spanish, French, Portuguese, German), leaving gaps in other languages and regions. Broader partnerships are needed to improve coverage. - Update frequency: Computational cost currently prevents instantaneous global P50 map updates; updates are annual. Consequently, the platform is an imperfect surrogate for systematic surveillance but useful for short-term prioritization and for fostering long-term surveillance development.