Sales data as a measure of antibiotics usage: Concepts, examples and discussion of influencing factors

Abstract Monitoring and surveillance of antimicrobial usage in animals is a public health concern and different methods are currently discussed widely in public, science and politics. The objective of the paper is to present the available methods of monitoring and to discuss possible differences in the assessment of the antibiotics treatment. Sales data are expressed as the average amount of substance per animal or kg live weight (e.g. population‐corrected unit, PCU). The number of Defined Daily Doses (nDDDvet) is calculated by extrapolating sales data with average animal weights and defined drug doses to a number of treatments theoretically applied to animals. In contrast, the number of Used Daily Doses (nUDDvet) displays the actual number of treatments which have been applied. As sales data are relatively easily obtained, they are frequently used. However, their results are influenced by the composition of the population and by the dose of the substances. As both may vary strongly between countries, direct comparison of sales data between countries may be misleading. The concept of analysing sales data is shown by exemplarily using the methods in the ESVAC report 2015. The presentation of usage data in terms of nDDDvet or of nUDDvet increases the comparability of the data from different countries or time periods. Furthermore, fluoroquinolones and third‐/fourth‐generation cephalosporins which, among other substances, bare a potential risk for human health, are used at low doses. Hence, their use contributes to a sales reduction while contrasting the guidelines of prudent use. nDDDvet or nUDDvet have the ability to better reflect the treatment frequency and thus to better link antibiotics use to public health concerns. Quantification of antibiotics should assist to focus on prudent use of antimicrobials to reduce the burden of resistant bacteria and, thus, enhance public health, animal health and animal welfare.

calculated by dividing the sum of sold active substances by any population unit, for example, number of animals or kg biomass produced.
Many countries report sales data and related units of measurement such as mg/biomass regularly (Bager et al., 2015;Belgian Veterinary Surveillance of Antibacterial Consumption National consumption report, ; The Public Health Agency of Sweden & National Veterinary Institute, 2014;Veterinary Medicines Directorate, 2012). Sales data are also used in projects and reported in the respective publications (Bondt, Jensen, Puister-Jansen, & van Geijlswijk, 2013;Carmo et al., 2017;Menéndez González, Steiner, Gassner, & Regula, 2010).
The project European Surveillance of Veterinary Antimicrobial Consumption (ESVAC) of the European Medicines Agency (EMA) was established in 2009 and has been reporting sales data in European countries since. Countries take part voluntarily and the most recent report from 2016 states a total of 30 countries (27 EU member states, 2 EEA countries as well as Switzerland) providing data (European Medicines Agency und European Surveillance of Veterinary Antimicrobial Consumption, 2018); Grave et al., 2014;. Results are reported in terms of population-corrected sales in mg of substance per population-corrected unit (PCU) (European Medicines Agency, 2017; Grave et al., 2014).
Although easily obtained, the presentation of antibiotics usage in terms of sales data is prone to being influenced by factors of several types. Both the numerators, that is, the amount of substances, as well as the denominator, that is, the population at risk, underlie various factors, which may strongly influence the outcome. Consequently, data from different populations, for example, countries, or time periods cannot be compared directly and may lead to misinterpretation of the results (Bondt et al., 2013). Collineau et al. (2017) recently published a detailed overview and discussion regarding the indicators for quantification of antimicrobial usage and their applications. They point out that nationallevel data are useful for certain study objectives such as monitoring trends over time, but only if comparability of the populations is given. Lekagul et al. (2018) as well as Werner, McEwen, and Kreienbrock (2018) also reviewed the diversity of methods for quantification.
The purpose of the present publication is to explain the pitfalls and consequences of using sales data for inter-country comparison, that is, the challenges regarding the composition of animal husbandry forms of the countries, the different dosing regimens of substances and the different impact of substances regarding public health. To demonstrate these restrictions, the concept of analysing sales data is shown briefly by exemplarily using the methods in the ESVAC report 2015 (European Medicines Agency, 2017).
Additionally, alternatives to sales data such as the quantification of defined or actual treatments are presented and discussed.

| MATERIAL S AND ME THODS
Methods are described in general, details refer to the ESVAC report 2015 as example (European Medicines Agency, 2017), which is known to be the most comprehensive system on antibiotic sales data documentation in the world and has been evaluated by a large group of experts from the European countries.

| Population at risk
The definition of the population at risk of being treated is a crucial variable for which several different calculation approaches are available (Collineau et al., 2017). One concept is to identify the biomass or live weight at risk of being treated by multiplying the number of produced or live animals with the respective expected body weight at typical treatment age. The expected body weight differs between animal species, age or production groups. Alternatively, the number of animals at risk of being treated can be used.

| Sales data
The amount of antibiotics sold is expressed as amount of antimicrobial substances in tonnes and is usually completed by information on the substance and its pharmaceutical form. The countries submit the number of sold packages per product and package size to ESVAC.
The countries also provide information on name and concentration TA B L E 1 Average animal weights at typical age of treatment (European Medicines Agency, 2009;Montforts, 2006)

| Quantification of defined treatments
The frequency of treatments is described by means of technical

| Quantification of actual treatments
UDDvet is the used daily dose that is applied to an animal on one day. The following formulas 3 and 4 are suitable for these means of calculation.
Inaccuracies occur when long-acting compounds are used that are effective for more than 24 hr. Correction may be performed by replacing treatment duration by duration of effectiveness typical of this substance, but currently no internationally accepted solutions for this problem exist.

| Examples
To illustrate differences in the results depending on the calculation method used, the treatment of respiratory diseases in broilers is given as an example. The mg/PCU of the specific treatments is calculated following formulas 5 and 6.

| Statistical analyses
Statistical analyses were carried out with IBM SPSS Statistics version 24. Published data of the ESVAC report 2015 were used to carry out the following analyses: the variable 'animal density in PCU/km 2' was calculated by dividing the overall number of PCU (European Medicines Agency, 2017) by the area of the respective country (www.wikip edia. com) in square kilometres. Additionally, the variable 'percentage of the overall PCU related to pigs and poultry (intensive animal husbandry species)' was calculated.
Variables were checked for normality by Shapiro-Wilk test as well as visually. 'Population-corrected sales in mg/PCU' was transformed to logarithmic values (to the base of 10) to achieve normal distribution.
Pearson's correlation coefficient was calculated to display the correlation between the percentage of overall PCU related to pigs and poultry and the population-corrected sales in mg/PCU (log). To investigate the influence of the percentage of PCU related to pigs and poultry as well as the animal density in PCU/km 2 on the population-corrected sales in mg/PCU (log), a multivariable linear regression model was adjusted. Values of the independent variables were centred. The animal density did not show a linear relationship to log mg/PCU but its cubic transformation x_cubic = x * x 2 *x 3 fitted well to the data. Thus, the variables 'animal density', '(animal density) 2' and '(animal density) 3' (1) nDDDvet = amount of substance Defined Daily Dose (2) nDDDvet∕PCU = nDDDvet number of animals × average animal weight

| RE SULTS
The percentage of PCU related to pigs and poultry (intensive animal husbandry species) as reported in the ESVAC report 2015 ranged from 9% to 79% (European Medicines Agency, 2017) and showed moderate correlation to the log population-corrected sales in mg/ PCU (Pearson's r = .586, Figure 1).
Since a high animal density (PCU/km 2 ) correlated with a high percentage of intensive animal husbandry species (Pearson's r = .479), we investigated whether the percentage of intensive animal husbandry species or the animal density had significant effects on the overall population-corrected sales in log mg/PCU. The multivariable regression analysis (p < .001) demonstrated that the percentage of PCU related to intensive animal husbandry species (p < .001) as well as the cubically transformed animal density (p = .002 for animal density, p = .003 for (animal density) 2 , p = .010 for (animal density) 3 ) had statistically significant effects on log mg/PCU ( Table 2, Table 3).
The regression coefficient of.020 is defined as the populationcorrected sales increased by 0.020 log mg/PCU when the percentage of PCU related to pigs and poultry increased by 1 percentage point. The adjusted R squared of .524 indicated that more than half of the variation of log mg/PCU could be explained by the variables in the model.

| Relationship between sales, nDDDvet and nUDDvet
The relationship between sales, nDDDvet and nUDDvet, is displayed by the following example: respiratory diseases in broilers can be treated with different substances. As example, 100 broilers of 800 g live weight each must be treated either with tylosin (macrolide) or with a combination of lincomycin (lincosamides) and spectinomycin (aminoglycosides).
Alternatively, the veterinarian decides for a treatment of the combination preparation with 17 mg/kg lincomycin and 34 mg/kg spectinomycin for 4 days (recommendation: 17-25 mg/kg lincomycin and 34-50 mg/kg spectinomycin for 4-7 days (Löscher et al., 2014)). The amount of substance per PCU is calculated following formula 6.
Treatment with tylosin results in.
while treatment with the combination preparation results in.
The nDDDvet is calculated as introduced above (formulas 1 and 2). It needs to be considered that the average animal weight = PCU =1 kg is used in these calculations instead of the actual animal weight = 0.8 kg. Based on the calculated usage data, treatment with tylosin results in 2.96 nDDDvet per PCU (Table 4) (Table 4).

| Sales data
Sales data are usually reported on national level and thus summarize the results over all animal species and substances included. Direct 100 mg∕kg ⋅ 3 days = 300 mg∕PCU, 17 + 34 mg∕kg ⋅ 4 days = 204 mg∕PCU Table 4 .  (2017) provide a complete and balanced review, the intention of this paper is to highlight reasons for the differences between countries that could be prevented using nDDDAvet or nUDDAvet instead.

| Substances
Sales data must be analysed separately for each substance or substance class due to two reasons. First, the doses differ significantly between substances, animal species as well as between different administration routes (examples given below). Second, the substances and substance classes differ concerning their impact on selection of resistant bacteria.

| Dosage
The main variables of substance selection are animal species, pathogen species, affected tissue (due to pharmaceutical characteristics and the respective availability in tissues) and route of administration.
The availability and the price of the pharmaceuticals also appear to be important as well as habit, ease of application and marketing strategies of products.
The dosage of colistin in terms of weight of active substance per kg bodyweight is 5.1 mg/kg for poultry, whereas tylosin has a dose of 81 mg/kg (European Medicines Agency, 2016. Consequently, treatment with tylosin uses much more mg/PCU than treatment with colistin. Doses also depend on the animal species. For example, tylosin has recommended dosages of 81 mg active substance/kg in poultry, 12-13 mg/kg in pigs, and 13-41 mg/kg in cattle, depending on the administration route (European Medicines Agency, 2016). This also influences the amount of substance used and can cause a bias in comparisons between countries.

Sum of squares
Bias can also occur as the recommendations in the Summary of Product Characteristics (SPCs) vary between different pharmaceutical products, although they contain the same substance and refer to the same species, administration route and indication. These circumstances not only affect the mg/PCU but also the nDDDvet. Solely the nUDDvet is not affected as it considers the used dose instead of a defined dose.

| Classification of substance groups
The second reason for substance-specific analyses is that the usage of antimicrobials also affects the selection and distribution of bacterial strains that acquired resistance against specific anti- As some critically important substance classes such as cephalosporins and fluoroquinolones have low doses, the usage of these instead of other substance classes can cause an overall sales reduction while increasing the risk for public health at the same time.

| Quantification of defined treatments nDDDvet
Although defined doses are determined following the recommenda- in a multi-country study on broiler farms revealed differences within and between countries. The same research consortium 'Ecology from Farm to Fork Of microbial drug Resistance and Transmission' EFFORT yielded similar results in pig farms .
The calculation of defined treatments allows for comparison of treatment frequencies between substances. It is also possible to look for trends and differences between countries. For assessment of antibiotics usage in animals concerning public health aspects, possible shifts from undesired substance classes such as fluoroquinolones or third-/fourth-generation cephalosporins to less critical substance classes can be observed.
In conclusion, the number of defined treatments is well suited for population or time period comparison. The main challenge is to point out that although this form of data presentation may seem to reflect the number of actual substance application to animals, it must be regarded as technical unit which is not identical to the actual number of treatments. Thus, it is recommended not to be used in communication with farmers. Nevertheless, in some countries such as Denmark and the Netherlands, this method is successfully employed for benchmarking of farms (Bager et al., 2015;SDa expert panel, 2016).

| Quantification of actual treatments nUDDvet
The other approach is to determine the actual number of used daily doses, nUDDvet. The relevant information can only be collected on farm level and thus requires much effort (Menéndez González et al., 2010;Persoons et al., 2012;van Rennings et al., 2015).
Furthermore, data often originate from different sources, for example, farmers or veterinarians, and thus, reliability of data might be violated due to reporting bias. The possible (negative) consequences further decrease the willingness of data owners to submit data.
However, the nUDDvet is well suited for analysis and communication of antibiotic usage on farm or veterinarian level, as its calculation is based on real treatments and can be easily understood.

| Comparison of the different approaches
The results of the three calculation methods differ. If the substance is unknown, it is not possible to infer treatment frequency from the amount of substance used. Differences between nDDDvet and nUD-Dvet rely on differences between defined and actual dose or on differences between average and actual animal weight. Discrepancies between used and defined doses do not imply that an inappropriate dose was chosen, as appropriate dosing depends on certain conditions within the animal group such as age, health status or resistance situation. The discussion of this paper focuses on the possible dif- In summary, decision-making remains difficult for the practitioner.
In conclusion, the assessment of antibiotics usage in animals is not straightforward. Some measures seem to be more precise concerning the frequency in which an animal is treated with antimicrobials on average. As these measures require much time and effort, they are applicable for studies with a limited (representative) sample size and may include reporting bias. Thus, the results underlie uncertainty and bias.
As sales data can be made available as complete and unbiased datasets by many countries, they are currently the best choice for comparisons between countries (Grave et al., 2014). This approach facilitates sampling of respective data and thus encourages countries to implement a monitoring system. It represents the necessary first step which is ideally followed by more detailed monitoring and surveillance systems focusing on the reduction of antimicrobial use in general or of critically important antimicrobials in the context of public health. EMA already initiated data collection activities on species level enabling the calculation of nDDDvet for different animal species/category separately (European Medicines Agency, 2018).
As explained above, this change in methods will contribute to data stratification for animal species/categories and will additionally control the effects of substances.
Recently, several reviews have been published that discuss the methods available and their applications seriously. Interested readers may refer to Collineau et al. (2017) and to Werner et al. (2018).
Regardless of the method of data collection, monitoring of the antibiotics usage must be regarded as a tool to observe changes in the pattern of antibiotics use. The main goal is to reduce the risk for public health which requires the distinction between substances.
Monitoring and surveillance should not only focus on a general reduction of antibiotics use. The risk of selection and distribution of resistant bacteria should also be minimized while protecting animal health and animal welfare at the same time. It is the task of veterinarians, animal owners, scientists as well as authorities to develop applicable approaches to enhance animal health and thus reduce the need of antimicrobials. The quantification of antibiotics usage in this context is only a tool and should not be misinterpreted as objective.
Actions should focus on prudent use to reduce the burden of resistant bacteria and thus enhance not only public health but also animal health and welfare.

ACK N OWLED G EM ENTS
The authors gratefully acknowledge Lothar Kreienbrock (University of Veterinary Medicine Hannover) for carefully editing the manuscript draft and Nancy Erickson for English editorial.

CO N FLI C T O F I NTE R E S T
All authors do not declare any competing interests and agree to the submission of the manuscript.

E TH I C A L S TATEM ENT
The authors confirm that the ethical policies of the journal, as noted on the journal's author guidelines page, have been adhered to. No ethical approval was required as this is an article with no original research data.