Global surveillance of antimicrobial resistance in food animals using priority drugs maps

Antimicrobial resistance (AMR) in food animals poses a significant threat to animal and potentially human health. The majority of antimicrobials sold globally are used in food animals, often for treatment and non-therapeutic purposes like growth promotion and substituting for good hygiene practices. This widespread use has facilitated the intensification of animal farming, meeting growing global demand for animal protein. However, in LMICs, the prevalence of resistance has increased substantially between 2000 and 2020, with significant consequences for animal health and the potential for human health impacts. While high-income countries have systematic surveillance programs, LMICs largely rely on point prevalence surveys (PPS) which lack the detail needed for targeted interventions. Existing methods using PPS often rely on summary metrics like the percentage of antimicrobials with resistance exceeding 50% (P50), hindering action on individual antimicrobial classes. This study aimed to address this gap by mapping AMR prevalence for individual antimicrobial-bacteria combinations across LMICs, predicting future resistance trends, and identifying priority antimicrobials for surveillance based on local epidemiological patterns. The World Health Organization's list of Medically Important Antimicrobials (MIA) provides a starting point, but doesn't account for geographic variations in AMR within countries. Predicting which antimicrobials will surpass critical resistance levels is crucial for assessing risks and strengthening surveillance efforts. This study uses 1088 PPS to map, at a 10x10 kilometer resolution, the resistance prevalence to seven antimicrobials in *Escherichia coli* and nontyphoidal *Salmonella* in food animals across LMICs, combining these maps with environmental and anthropogenic covariates and patterns of co-resistance to predict which antimicrobials will most likely exceed critical resistance levels in the near future.

The literature review examined existing studies on antimicrobial resistance in food animals, focusing on the limitations of current surveillance methods in LMICs. Studies in high-income countries have established the importance of systematic surveillance for guiding policies and limiting the use of specific antimicrobial classes. However, LMICs generally lack such systems, relying instead on less comprehensive point prevalence surveys (PPS). These PPS provide valuable snapshots of AMR prevalence, but their limited scope and inconsistent reporting of antimicrobial-bacteria combinations has historically prevented detailed analyses and targeted interventions. The lack of disaggregated data on individual antimicrobial-bacteria combinations hinders effective strategies to combat AMR. Previous global estimates of AMR have often combined data across multiple pathogens, obscuring differences in geographic patterns for specific antimicrobials and bacteria. This study builds upon prior work, providing a more refined and geographically specific analysis of AMR in food animals within LMICs.

This study employed a five-step methodology. First, 1088 point prevalence surveys (PPS) on AMR in *E. coli* and nontyphoidal *Salmonella* from healthy food animals in LMICs between 2000 and 2019 were collected and extracted from four databases (PubMed, Scopus, ISI Web of Science, and China National Knowledge Infrastructure). Data extracted included resistance prevalence, antibiotic susceptibility testing (AST) methods, breakpoints, sample origin, geographic location, and time. Variations in breakpoints across surveys were adjusted to maximize comparability. The analysis focused on seven antimicrobials: tetracycline (TET), ampicillin (AMP), sulfamethoxazole-trimethoprim (SXT), chloramphenicol (CHL), ciprofloxacin (CIP), gentamicin (GEN), and cefotaxime (CTX). These were selected for their frequent reporting in surveys and importance in both veterinary and human medicine. Logistic regression models were used to estimate temporal trends of resistance prevalence between 2000 and 2019. The spatial distribution of resistance prevalence for each antimicrobial was mapped at a 10x10 kilometer resolution using Gaussian process stacked generalization, an ensemble approach combining three machine learning models (boosted regression trees, LASSO logistic regression, and feed-forward neural networks). Environmental and anthropogenic covariates, including antimicrobial use, animal population density, and temperature, were incorporated. To predict priority antimicrobials for surveillance, a separate model was developed using LASSO logistic regression. This model predicted the probability of each antimicrobial's resistance prevalence exceeding critical levels (10%, 25%, or 50%) in the future, based on local risk factors and patterns of co-resistance observed in the PPS. The uncertainty of the mapped predictions was quantified using the variance of the posterior predictive distribution and Monte Carlo simulations of imputed datasets. Spatial cross-validation was employed to assess the accuracy of the models. The entire process is graphically illustrated in Figure 5.

The analysis revealed several key findings. First, the mean prevalence of resistance, weighted by the number of samples, was highest for tetracycline (59% for *E. coli* and 54% for nontyphoidal *Salmonella*) and lowest for cefotaxime (33% and 19%). Significant temporal increases in resistance were observed for all antimicrobials except tetracycline. Prevalence varied by animal species, with poultry and pigs showing significant increases in resistance for several antimicrobials. Geospatial modeling identified hotspots of AMR in various regions, notably southern and eastern China, central Asia, northern India, northern Brazil, and Chile for *E. coli*. Northeastern China emerged as a hotspot for several antimicrobials in both *E. coli* and nontyphoidal *Salmonella*. The geographic distribution of AMR varied depending on the bacteria and antimicrobial class considered. For example, northeastern China was a hotspot for most antimicrobials except tetracyclines and penicillins, highlighting the need for targeted interventions in this region. Africa consistently showed lower AMR prevalence compared to other regions. Predictions of which antimicrobials would most likely exceed critical resistance levels (50%) in the future indicated that tetracycline or ampicillin were the priority antimicrobials in 78% of locations in Africa and South America, while ampicillin or sulfamethoxazole-trimethoprim were priorities in 77% of Asian locations. The uncertainty associated with predictions was generally low, except in some specific areas. The estimated time until resistance prevalence exceeded 50% varied by antimicrobial, with shorter times for tetracyclines, penicillins, and cephalosporins (less than 7 years on average). Co-resistance between antimicrobials was significant, with the strongest correlations observed between sulfamethoxazole-trimethoprim and chloramphenicol, and sulfamethoxazole-trimethoprim and tetracycline.

The findings highlight the geographic heterogeneity of AMR in food animals across LMICs and offer insights into future trends. The identified hotspots underscore the need for targeted interventions to reduce antimicrobial use and enhance biosecurity in these regions. The predictions of priority antimicrobials for surveillance are particularly valuable, providing a framework for resource allocation and focused monitoring. The observed co-resistance patterns emphasize the need for a holistic approach to AMR management, addressing multiple antimicrobial classes simultaneously. The relatively short estimated time until resistance prevalence surpasses critical levels for some antimicrobials necessitates prompt action. Differences in AMR prevalence between animal species suggest that interventions should be tailored to specific animal production systems. The study's findings align with previous global estimates of AMR, but the disaggregation by antimicrobial-bacteria combinations provides more precise information for targeted interventions. The study's results provide crucial information for informing national and regional AMR policies and strategies.

This study provides the first global maps of AMR in food animals across LMICs, disaggregated by antimicrobial-bacteria combinations and predicting future trends. The identified AMR hotspots and the prediction of priority antimicrobials for surveillance offer a crucial framework for targeted interventions and resource allocation. Future research should focus on refining these predictions with more granular data, incorporating country-specific regulations on antimicrobial use, and investigating the underlying mechanisms of co-resistance. This study highlights the urgent need for a coordinated global effort to combat AMR in food animals, focusing on responsible antimicrobial use and enhanced biosecurity measures.

The study acknowledges several limitations. First, the predictive maps and the imputation of missing resistance prevalence introduce uncertainty. The limited number of surveys reporting resistance for individual antimicrobial-bacteria combinations restricted the analysis to seven drugs and two bacteria. The pooling of surveys from all years into a single analysis limited the ability to conduct robust spatio-temporal modeling. The predictions are based on the assumption that resistance prevalence will continue to increase at a similar rate as in the past 20 years. The lack of a systematic inventory of country-specific regulations on antimicrobial use prevented their explicit inclusion as covariates. The choice of 50% as a threshold for defining priority antimicrobials is somewhat subjective. Finally, the study focused on commensal *E. coli* and nontyphoidal *Salmonella* from healthy animals, and used human clinical breakpoints for determining resistance phenotypes, which may not fully capture the situation for all pathogens and animal species.