Antimicrobial resistance (AMR) is a global issue, and the current pattern and reduction in its harmful consequences to the biosphere's health (1, 2) require concerted actions from the human, animal, agricultural, and environmental sectors (3). A key measure for AMR mitigation is reducing antimicrobial use (AMU), especially “Critically Important Antimicrobials” (CIAs) for human health (4–6). The livestock industry is predicted to be responsible for 70% of the global AMU by 2030 (2, 7) and the possible relationship between AMU in animal production, and the development of AMR has been highlighted (8). Despite the association between AMR in livestock and humans, there is uncertainty about its magnitude (9–11).

Measuring AMU is fundamental for monitoring and reduction of AMU. Key indicators for understanding the patterns include (a) monitoring trends over time; (b) comparisons between different populations (e.g., types of production, species, or countries); (c) benchmarking; and (d) study of associations between AMU and AMR (12). To date, no standardised AMU measurement fulfils all these objectives. Thus, suitable measurement(s) must be determined based on a trade-off between the set goals and data at hand. Data resolution, comprehensiveness, and stability over time are important for the assessment of exposure and comparison of AMU within and between populations (12, 13).

All the practicalities around collection and reporting, regardless of the chosen resolution (e.g., animal, herd, or country level), have a high impact on AMU measurements. In turn, this affects the transparency and comparability of figures obtained. Despite this, only a few studies have addressed the qualitative and quantitative biases of AMU reporting systems (12, 14–16).

In Sweden, AMU in animals is only allowed on veterinary prescription, and veterinary drugs are sold exclusively through registered pharmacies. Sweden has had a leading role in the reduction of AMU, as well as the quantification and reporting of AMU statistics in animals and humans via the Swedres-Svarm reports (17). These reports are based on national sales data, which are not always the same as the amount prescribed or used in a country.

Although no recent effort has been made to evaluate AMU at herd level in Swedish dairy herds (18), the tools to do so are available through official records of veterinary treatments. The “Djursjukdata DAWA” is owned by the Swedish Board of Agriculture (in Swedish: Jordbruksverket). This is the oldest data collection system for veterinary treatments in Sweden, including antimicrobial use at herd level, initially compiled via paper records and launched at a national level in 1984. Currently, a computerised system covers all food-producing animals and horses, although 30% of the data is still fed to the system in paper format. Moreover, this system provides health information to Växa, the biggest dairy Levy group in Sweden that aims to monitor and improve the productivity, health and welfare of Swedish dairy cows (19). Växa's national database thus opens the window to understanding AMU in the context of detailed herd characteristics not available in the DAWA system.

Based on the EU rules for organic farming, dairy herds are expected to maintain a restrictive AMU (20). In practice, these rules limit the number of treatments per animal/year, and depending on the member state, might include the extension of drug withdrawal periods or the promotion of alternative medicines. Since conventional herds do not have to abide by such rules, it is often assumed that AMU would be higher in these herds. Although some studies have confirmed that organic herds have a lower AMU than conventional herds (5, 21–25), this research area remains unexplored in Sweden. The requirements may also impact the role of prescribing veterinarians under the two systems and, thus, also the AMU recordings. Hence, a comparison between AMU in organic and conventional dairy herds provides a case for AMU data evaluation.

This study aimed to provide a qualitative and quantitative assessment of how data source affects AMU reports, including treatment characteristics and characteristics of the population treated and at risk within a herd. Additionally, we wanted to quantify, describe, and compare the AMU in organic and conventional Swedish dairy herds and explore if AMU recording differs between the systems.

To effectively address the situation of AMU and AMR, herd-level data are needed. To our knowledge, this is the first study to provide a full description of AMU regardless of medical indication and split by production type (i.e., organic and conventional herds) in a Swedish dairy context.

The BIN method captured more drug use than SBA and VXA. The first type of discrepancy found was in the number of herds with corresponding reports of AMU across all datasets. The lowest percentage of discrepancies was found between SBA and VXA reports, especially for the injectable forms. Yet, the highest percentage of discrepancies was found between BIN and VXA datasets, especially for intramammary tubes for dry-cow treatments. The second type of discrepancy was in the amount of AMU reported among datasets. Here, Bland–Altman analyses indicated an overall trend of AMU underreporting in the SBA and VXA datasets compared to the reports from the BIN dataset. In most cases, the limits of agreement were large but not beyond what would be considered “clinically acceptable” (33), as in most cases, the disagreement represented <1 treatment course per animal. Consequently, datasets could be used interchangeably. However, as large discrepancies were found for intramammary drug forms, any metrics should take this into account.

A major strength of the BIN dataset is that “overreporting” of AMU is unlikely. This could occur if farm staff discarded outdated or unused drugs, or if half-empty packages/bottles/tubes were reported as fully used. We reduced that risk by making sure that all recorded packages were either empty or reported as the amount used if a half-empty package/bottle was found. In Sweden, the amount of the drug prescribed and dispensed to a farm or individual should match the volume/amount necessary to cover the treatment. Any leftovers must be safely discarded, preventing antimicrobial hoarding or imprudent handling of waste (34). Adherence to this could not be confirmed on the visiting farms. Yet, finding partially used bottles/packages in the BIN might suggest that staff on-farm indeed discard leftovers as required.

However, when SBA or VXA report higher AMU than in the BIN, we have little room to know if actual “underreporting” in the BIN occurred, i.e., if farm staff forgot (intentionally or not) to put the empty packaging in the BIN. In our study, we could establish real, yet probably unintentional, underreporting in the BIN for intrauterine drugs. We found no intrauterine drug packages in the BIN; however, a few such reports were found in the other datasets. Upon discussion with veterinary practitioners, it was understood that the few intrauterine treatments that take place on a farm are performed by the veterinarian on the spot. Hence, the veterinarian routinely discards the gloves, and drug packaging used together without the involvement of the staff on-farm, and the packaging does not reach the bin. This behaviour may also partially explain the overall trend of BIN underreporting AMU for intramammary tubes for lactating cows. These treatments can also be carried out on the spot by the veterinarian, and again, the discarded containers would not necessarily reach the bin.

Several studies have compared AMU based on drug packaging collections (i.e., BIN method) against other data sources, mainly farmers' reports (14, 16, 35). Similar to our study, the authors found that the BIN method/dataset outperforms other data sources. These studies attributed the success of the method to the convenience for staff in reporting AMU with the simple act of discarding packages, saving them from the burden of collating information. In our study, the prescribing veterinarians provided the data for the compared datasets. So, in essence, SBA and VXA reflect prescriptions, while BIN captures actual use. Veterinarians in the research team (KS, NF, and SSL) recognise that AMU reporting is an administrative burden. Depending on the tools available, records can be either collated manually (i.e., pen and paper), scanned and sent to SBA to be uploaded electronically, or transferred manually into a web portal. Alternatively, records could be transferred electronically directly to the SBA database. Transferal could be done on the spot as an individual record or as a bulk of prescriptions later on. Thus, the ability to correctly collate the information is highly depending on the clinical record-keeping. Indeed, upon revision of working datasets ahead of calculations, many treatment records had to be removed as not enough details were present to determine if they were AMU. Poor quality in reporting could explain why some herds failed to present records in SBA or VXA datasets when treatment was performed based on what was found in the BIN dataset. Another explanation could be that the full veterinary prescription was issued before the study period and hence did not appear in the databases. However, the actual use of the drug occurred during the study period, as shown in the BIN.

Nevertheless, the observed discrepancies are consistently larger for intramammary forms, primarily when related to dry-cow treatments (77 to 85% of missing records in either SBA or VXA datasets compared to BIN). Thus, contrary to what is expected, some prescriptions do not get reported as required. In Sweden, current recommendations condemn the storage of antimicrobials (e.g., hoarding of antimicrobial leftovers). Moreover, blanket treatments or preventive AMU is not allowed; selective dry cow therapy is permitted only in individual animals, and cows will only get treated after a diagnosis is made (36, 37). Thus, the lack of reports in SBA and VXA datasets compared to the BIN might be an indication of some practice deviations. If so, further understanding of the quality, magnitude, and drivers of this are needed.

Based on our results, it could be said that BIN provided more information on AMU at herd level for short observation periods. Yet, it is unsustainable for long periods and challenging to scale up on a national basis. On the other hand, our study confirms that SBA and VXA could be used interchangeably. Yet, VXA offers the advantage of being the only standalone dataset for obtaining herd-level AMU metrics in the recommended unit, since time-at-risk can be calculated, over a long time, and scaled to a national level. Nonetheless, underreporting of intramammary drug forms needs to be adequately addressed for SBA and consequently for VXA. Here, we suggest the application of a biannual screening of a random selection of herds using the BIN methodology. This exercise would allow formalizing the monitoring and validation of results captured by SBA and the VXA dataset.

With a focus on BIN reports, we found that the studied farms had an average treatment incidence and average course treatment of 0.43 DDD_vet/animal/year and 0.22 DCD_vet/animal/year, respectively. Procaine benzylpenicillin (QJ01CE09) was the preferred (92%) antimicrobial to be used in the reported treatment entries, where more than half (56%) were related to udder problems. This is only the second time in more than 20 years that AMU at herd level is published for dairy cattle in Sweden (38, 39). Direct comparison with these and work in other countries where treatment incidences are reported (23, 40, 41) are constrained mainly by variations in sources of data, calculation methodology, and study design (12).

Despite the challenge of direct comparisons between studies, current AMU in Swedish dairy herds is much lower compared to previous reports for Sweden (6.4 DDDcow/1,000 cow-days for injectable; 3.45 DDDcow/1,000 cow-days intramammary drugs) (38). Equally, it is also low in an international comparison of the median number of doses reported (interquartile range 5.5–13.6 DDDcow/1,000 cow-days) (40). Moreover, results agree with latest reports by EMA (42, 43). Such reports use the “population correction unit” (PCU) based on livestock demographics to estimate the total weight of the livestock population in each country to then compare AMU across the EU in mg/PCU. The report indicates that Sweden is the EU member state with the lowest AMU and after Iceland and Norway the third lowest within all European countries (43).

Only 16 ATC_vet codes, representing six antimicrobial classes, were found across the datasets. The list is smaller than that previously reported in Sweden (39, 44) or found in other international reports (40). For example, our study found no reports of macrolides. This use was common 20 years ago and continues to be so in some countries (40). We found no ATC_vet codes representing antimicrobial groups like cephalosporins, amphenicols, lincosamides, or pleuromutilins. Injectable (QJ01CE09, 92% farms) and intramammary penicillins (QJ51CE09, 32% farms) were the most used antimicrobials followed by tetracyclines (QJ01AA06, 17% farms), albeit with a low DDD_vet/animal/year value compared to that of penicillins. Internationally, the most reported antimicrobial groups in dairy cattle are penicillins, and third-generation cephalosporins (40) and highest-priority critically important antibiotic (HPCIA) treatments of mastitis could range from 10 to 80% depending on the veterinary practice (44). In our study, udder health was also the main reason behind the observed AMU. Yet, the HPCIA treatments were low, and cephalosporin use was not recorded at all. HPCIA category 1 includes macrolides, certain penicillins, and tetracyclines, while category 2 includes ampicillins, aminoglycosides, quinolones, 3rd- and 4th-generation cephalosporins, and polymyxins (e.g., colistin).

Moreover, our study also found large variations of HPCIA treatment percentages across farms. Yet, that includes many herds with no HPCIA treatments for category 2. Furthermore, the HPCIA reports for category 2 were mainly due to the use of aminoglycosides or quinolones but not to cephalosporins as in other countries. Sweden has a long-standing history of strengthening its policy recommendations across the human and veterinary sectors to reduce the AMR burden (43, 45). The efforts made have had a definite impact on the reduction as well as the pattern of AMU (17).

Previous studies comparing organic and conventional dairy herds under Swedish conditions found a marginal difference in udder health and reproductive performance, implying equally good animal health in both systems (46–48). Our study presented marginal differences between organic and conventional farms. It should be noted, however, that the small differences are due to an equally low AMU in conventional Swedish farms. Organic herds had a small advantage in that they had a lower mortality and less lameness and mastitis cases. Yet, the bulk milk somatic cell counts were higher for organic farms. Differences had been reported elsewhere between organic and conventional herds (23, 24), indicating that organic herds had a substantially lower level of AMU, but the difference in management between organic and conventional herds is typically larger outside Sweden (49, 50).

Organic herds had numerically lower treatment incidence and a lower number of treatment courses regardless of dataset or treatment form. Yet, at a closer look on the BIN dataset and DCD_vet/animal/year metric, an interaction between the production system and drug administration form was found. Organic herds had a significant lower AMU for injectables and intramammary lactating cow treatment forms. Yet, organic and conventional farms had a similar AMU for intramammary dry-cow treatment forms.

A potential explanation for the AMU difference between organic and conventional herds for particular drug forms, as suggested by others (23, 24), could be the underlying marginal difference in udder health between production systems in Sweden. Yet, evidence also exists that multiple factors beyond production type characteristics drive the choice of therapy and related management practices, such as farmers' and veterinarians' beliefs and social pressure (51, 52). A qualitative enquiry in the study herds could not relate any particular AMU perception to production type. Still, the authors highlight some behavioural discord among the farmers and veterinarians around udder treatments (53). This, in combination with our findings of underreporting of dry-cow intramammary tubes, suggests that these behaviours are a product of a different set of beliefs around the treatment of cows at dry-off and the practicalities of prescription and administration across herds, but this requires further investigation.

We conclude that herd-level assessment of AMU is possible using reported prescriptions in combination with herd data, such as in the VXA database, with the caveat that intramammary drugs may be misrepresented. The BIN methodology offered a more comprehensive alternative to gain an account of AMU at herd level. However, as this method is labour- and resource-demanding, it is difficult to apply routinely at a national scale. Thus, for Sweden, we suggest the application of the BIN methodology on a random selection of herds to monitor and validate results captured by SBA and VXA. We also suggest improving the understanding and reduction of underreporting of intramammary drug forms.

This study provides a detailed contemporary account of AMU in Swedish dairy herds. It confirms the reported low AMU based on sales data. Moreover, we conclude that AMU differs between organic and conventional farms for particular drug forms that may partly be driven by the marginal difference in disease prevalence. The use and reporting of dry-cow intramammary drug forms requires further investigation.

The datasets for this article are not publicly available because of privacy and confidentiality reasons. Requests to access the datasets should be directed to the corresponding author, Jordbruksverket (jordbruksverket@jordbruksverket.se) and Växa Sverige (info@vxa.se), respectively.

Ethical review and approval was not required for the animal study because the study design adhered to the good scientific practice guidelines set out by Swedish legislation and did not include direct animal testing. Thus, no further ethical evaluation or permit was required for its execution. Written informed consent was obtained from the owners for the participation of their animals in this study.

GO was in charge of the analysis and wrote the manuscript. KS was in charge of the data acquisition and made a substantial contribution to the conception and revision of the document. SS and NF provided valuable expertise in the conceptualisation of the idea and writing process. UE assisted in the conceptualisation of the idea, analytical process, and revision of the manuscript. All the authors have read and approved the manuscript.

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.