Joining the Resistance Against Preventable Antimicrobial Use

Summary

All promotions can be followed live via this link: 
​https://www.uu.nl/en/organisation/utrecht-university-hall/schedule

BACKGROUND
In the early 20th century, the introduction of antimicrobials in healthcare revolutionized the treatment of bacterial infections, significantly improving clinical outcomes and increasing life expectancy. After World War II, the demand for animal proteins increased, and antimicrobials were used not only for therapeutic purposes (i.e., to treat infections in animals), but also to enhance production efficiency. By 1948, mass medication of large groups of animals through feed or drinking water was implemented and within a year, antimicrobials were being used at sublethal doses as antibiotic growth promoters (AGPs). Due to the lack of regulations, AGP use became common practice. The widespread use of antimicrobials (i.e., as AGPs, therapeutically, prophylactically and metaphylacticly) enabled intensive livestock farming at higher stocking densities with less labor and costs. As antimicrobials became central to industrial farming, concerns emerged as early as the mid-1960s about antimicrobial residues in animal-derived products and the environment, as well as the rise and spread of antimicrobial resistance (AMR) due to excessive AMU. Indiscriminate AMU led to substantial emergence of AMR among bacterial populations, posing life-threatening risks globally. In response, Europe implemented a complete ban of AGPs in 2006, followed by the prohibition of routine preventive use of antimicrobials in 2022. However, these two policies remain unrealized in many parts of the world. The Netherlands began targeted AMU reduction efforts in 2009, and since then a 78% reduction in AMU in Dutch livestock has been achieved due to a public private AMU reduction programme. Key measures include the mandatory reduction of 70% in AMU, antimicrobial stewardship, monitoring and benchmarking systems, prohibition of routine use, promotion of individual over group treatments and the recent introduction of acceptable AMU thresholds. Despite these successes, further AMU reduction remains a priority.

Achieving greater independence from AMU requires transitioning and improving farming systems. This has become evident in veal calf production, weaned piglets and conventional broilers where AMU reduction has stagnated in recent years, and many farms exceed acceptable AMU thresholds. The link between rearing conditions and animal health has been recognized since the origin of animal farming itself. Extensive research has explored resilient and robust livestock production systems, focusing on internal and external biosecurity and animal welfare. However, current intervention strategies often rely on expert opinions and personal experiences, rather than quantifiable, farm-specific evidence, making selection of optimal solutions subjective. Epidemiological studies over the past decades attempted to identify practices associated with AMU, but their results often lack direct applicability for farm-specific interventions. To address this challenge, exploring and operationalizing the relationship between farming practices and (diseases requiring) AMU at the individual farm level is essential. Advances in machine-learning (ML) algorithms now enable these types of analyses, offering a data-driven approach for developing tailored AMU reduction strategies.

RESEARCH AIMS
The aim of this PhD thesis was twofold. The first aim was to investigate which husbandry practices were associated with AMU in broilers, weaned piglets and veal calves in the Netherlands, and with diseases requiring AMU in the latter two species. Husbandry practices included all (infra)structural and managerial characteristics of farms, including available facilities, housing conditions, animal care, nutrition, animal health and welfare management. Broilers, weaned piglets, and veal calves were the focus of this thesis because they are among the main consumers of antimicrobials and key suppliers of food of animal origin. The analyses were performed using both previously and newly collected data, applying classical alongside machine-learning (ML) statistical methods, such as Random Forest (RF) models. These ML methods enabled the estimation of individual farm-level effects and complex interactions between predictors. The second aim of this thesis was to translate the results of the risk factor analyses into a tool able to generate customized AMU reduction plans for a given farm by operationalizing the husbandry practices associated with AMU or diseases requiring AMU. This tool was built as a free online application called AMUlet (Antimicrobials Usage Lowering Estimation Tool) designed to assist veterinarians and farmers in formulating AMU reduction plans. The algorithm behind AMUlet draws information from the statistical ML model to assess a farm’s strengths and weaknesses. It simulates counterfactual scenarios by evaluating all possible combinations of modifiable farm characteristics to predict changes in AMU levels or disease probability under different sets of interventions. Additionally, it incorporates a cost-benefit analysis to support informed decision-making.

MAIN RESULTS
In Chapter 2, a literature review on risk factors for AMU in pigs, broilers and veal calves was performed to map the current state of research and identify gaps in knowledge. This review showed that internal and external biosecurity are vital components for limiting AMU, as well as the importance of micro-climate conditions, especially in calves. However, the number of studies available on pigs and calves was relatively low, and research on broilers was even more limited. Meta-analyses of these studies were hampered by the lack of harmonization in measurement methods and categorization of farm characteristics, AMU and health-related data. In addition, all studies used models that describe effects at the population level and only one study used machine- learning approaches (i.e., Partial Least Squares (PLS)). Consequently, while this review highlighted the presence of potential interventions, it also identified significant constraints in quantifying their impact at the individual farm level. In addition, the earliest study included in the review dated back to 2007. Since then, research has evolved, incorporating higher variety and more detailed risk factors. Moreover, caution is necessary when extrapolating findings from older studies, as the livestock sectors undergo continuous changes that may influence the relevance and applicability of past results.

In Chapter 3, previously collected longitudinal data from 2012 in the Netherlands were analyzed to assess how different farm characteristics could explain AMU in Dutch pig and veal calf farms (N =36 and 51, respectively). Moreover, these analyses provided a comparative perspective on how the results from conventional variable- selection methods in regression analysis, such as backward elimination, differed from those obtained using PLS. The advantages of the latter approach were apparent, as it provided a more complete overview of how the different variables were associated with AMU. As these data were more than a decade old, the focus of the study was mainly methodological and not so much on the biological implications of the results, as livestock systems may have changed substantially since the time of data collection.

In Chapter 4, previously collected cross-sectional data on broilers from nine European countries gathered within the Ecology from Farm to Fork Of microbial drug Resistance and Transmission (EFFORT) project in 2014 were analyzed using RF and recursive feature elimination (RFE) for variable selection. These data contained information on AMU levels and internal and external biosecurity practices from 181 conventional broiler farms. Here it was observed that a farm’s AMU was primarily dependent on its size and the number of rounds per year, yet the effects of these variables were conditioned on the biosecurity practices applied, such as fully disconnecting the water system when cleaning. In addition, k-means clustering and prototypes of the proximity matrix from the RF showed how farms clustered together based on similarity of practices, and what factors defined high or low AMU levels in each cluster. Next, using the results from the final RF model, the alpha version of the AMUlet tool for broilers was developed, which is available in https://foresty.shinyapps.io/CIAOCIAO_Broilers/.

In Chapters 5 and 6, the same approach as in Chapter 4 was applied to develop the AMUlet tool for weaned piglets by analyzing newly collected data from 154 Dutch pig farms in 2019. However, here the full causal pathway, which included risk factors causing diseases leading to AMU, was taken into account. Extensive information on farm characteristics, AMU and disease indication for group antimicrobial treatments was collected through a digital questionnaire filled out by the veterinarians. The goal was to identify those farm characteristics affecting disease presence and AMU. As these two outcomes lay in the same causal pathway, the analysis was performed separately in the two chapters. In Chapter 5, it is shown that those diseases responsible for AMU were mainly respiratory and musculoskeletal/neurological conditions. A selection of variables associated with these diseases from the set of farm characteristics was performed using the binomial RF algorithm. In Chapter 6, the same analytical approach was used for variable selection, but the outcome here was AMU. Overall, variables such as late weaning, a lower percentage of slatted floor area and free-sow systems during lactation were protective against both diseases and AMU. In the AMUlet for pigs (https:⁠/⁠/foresty.shinyapps.io/CIAOCIAO_Pigs/), each model was used separately to provide the most suitable scenarios at the individual farm level for the respective outcomes.

In Chapter 7, similar analyses as in Chapters 5 and 6 were performed on newly collected cross-sectional data from 36 Dutch rosé veal calf starter farms in 2021. However, given the relatively small sample size, the PLS method was preferred along with a custom variable selection algorithm that resembled the RFE from Chapter 4. The main diseases within this production system were respiratory conditions, which were present in almost all farms. Farm characteristics associated with respiratory diseases and AMU were related to sorting of animals, micro-climate conditions, weaning methods and dairy density of the province. Based on two PLS models, the AMUlet for veal calves (https:⁠/⁠/foresty.shinyapps.io/CIAOCIAO_Veal/) was built, providing estimates at the population level.

CONCLUSIONS
Transitioning to a future where AMR no longer poses a threat requires the global elimination of AGPs and discontinuation of routine preventive AMU. However, achieving further reduction in therapeutic AMU will be challenging without thoughtful and evidence- based considerations of the reasons behind and the methods used in livestock farming. A key component towards this future is the development of tailored AMU reduction plans for individual farms, with reliable estimates of their expected impact. AMUlet is designed to support, but not replace, the decision-making process of veterinarians and farmers when designing AMU reduction strategies. Furthermore, the capabilities of machine- learning algorithms enable a more precise characterization of a farm’s situation, which allows to develop customized AMU reduction plans. Overall, this thesis provides both knowledge and tools to help achieve current and future AMU reduction goals. Moreover, the analytical framework developed here is adaptable to other data and settings. As concluded in the different chapters, risk factors for AMU vary between livestock sectors, but the main pillars of biosecurity, micro-climate, and animal welfare remain essential for reducing it. Finally, the availability of high-quality data is of paramount importance for conducting meaningful statistical analyses with practical applications. Given the continuous evolution of livestock farming practices, ongoing research in this field is essential to drive a sustainable paradigm shift in livestock production systems.