Pharmaceutical pollution: Prediction of environmental concentrations from national wholesales data

Abstract

The regulation and monitoring of pharmaceutical pollution in Europe lag behind that of more prominent groups. However, the repurposing of sales data to predict surface water environmental concentrations is a promising supplement to more commonly used market-based risk assessment and measurement approaches. The Norwegian Institute of Public Health (NIPH) has since the 1980s compiled the Drug Wholesale Statistics database - covering all sales of both human and veterinary pharmaceuticals to retailers, pharmacies, and healthcare providers.

To date, most similar works have focused either on a small subset of Active Pharmaceutical Ingredients (APIs) or used only prescription data, often more readily available than wholesale data, but necessarily more limited. By using the NIPH’s product wholesale records, with additional information on API concentrations per product from, we have been able to calculate sales weights per year for almost 900 human and veterinary APIs for the period 2016–2019.

In this paper, we present our methodology for converting the provided NIPH data from a public health to an ecotoxicological resource. From our derived dataset, we have used an equation to calculate Predicted Environmental Concentration per API for inland surface waters, a key component of environmental risk assessment. We further describe our filtering to remove ecotoxicological-exempt and data deficient APIs. Lastly, we provide a limited comparison between our dataset and similar publicly available datasets for a subset of APIs, as a validation of our approach and a demonstration of the added value of wholesale data.

This dataset will provide the best coverage yet of pharmaceutical sales weights for an entire nation. Moreover, our developed routines for processing 2016–2019 data can be expanded to older Norwegian wholesales data (1974–present). Consequently, our work with this dataset can contribute to narrowing the gap between desk-based predictions of exposure from consumption, and empirical but expensive environmental measurement.

Plain english summary

Pharmaceuticals, by design, affect human biology, target specific organs and biological systems to treat diseases. Pharmaceuticals and their metabolites—partly degraded or transformed ingredients—that reach the environment may have unwanted and long-lasting biological effects on plants, animals, and microbes. This comes in addition to environmental footprint of chemicals that are used during the production of pharmaceuticals. In Norway, a coastal nation of more than five million people, the primary route of pharmaceuticals in the environment is via human consumption. Although some pharmaceuticals can be metabolised in the body and degraded in sewage treatment plants, a proportion reaches rivers, lakes, fjords, and coastal zones.

A better overview of the types and amounts of pharmaceuticals in the environment is important for assessing and managing environmental risk, but doing so everywhere can be cumbersome, resource-intensive, and expensive. With limited funds available for environmental monitoring and management, a rapid and cost-efficient method for predicting concentrations of pharmaceuticals in the environment should be used to screen for the substances most likely to pose a problem.

In this paper we present such an exercise: we worked with the Norwegian Institute for Public Health’s wholesale drugs data, adapting, and translating it from a medical resource to a set of sales weights for each pharmaceutical ingredient. These sales weights were in turn used to predict concentrations of drug pollution in receiving freshwaters. In total, we predicted sales weights and environmental concentrations for almost 900 Active Pharmaceutical Ingredients, from abacavir to zuclopenthixol, sold between 2016 and 2019.

Introduction

Pharmaceutical consumption is widely recognised as an important source of anthropogenic chemicals in the environment ( European Commission, 2019; Richardson & Bowron, 1985). In much of the European Union (EU) and the European Economic Area, prospective (prior) environmental risk assessments of pharmaceutical products begin with an exposure assessment. Conservative, or worst-case Predicted Environmental Concentrations (PECs) of active pharmaceutical ingredients (APIs) are calculated by extrapolating from the highest average daily dose of a pharmaceutical, and the proportion of a nation’s population taking said pharmaceutical – by default, 1% ( EMA, 2006).

More recently, refined approaches have been suggested using pharmaceutical sales data collected by government agencies or market research agencies, to provide a more accurate and comprehensive prediction of environmental concentrations of APIs at the national ( Grung et al., 2008) and European ( Gunnarsson et al., 2019) level. In some cases, available data is limited to prescription sales, but where available wholesales data provides a far more complete picture of overall consumption.

In this paper, we present a dataset of predicted API consumption by weight based on reported sales weights of pharmaceuticals from a unique public sector source, the Drug Wholesale Statistics database of the Norwegian Institute for Public Health ( NIPH, 2019). This source covers all sales of pharmaceuticals and medicines to pharmacies, supermarkets, hospitals, and other healthcare providers, from the year 1974 onwards. We describe (1) the sales data and additional information on pharmaceutical API content for the years 2016–2019, (2) the procedures for converting the sales data from number of packets per product to amount (kg) of each API, and (3) a final dataset of total amount of API sold per year, which can be used for prediction of environmental concentration. Although these methods have only been applied to and evaluated for the years 2016–19, they may also be applicable to past data.

With this dataset, we aim to provide a highly accurate resource describing sales weights and predicting environmental concentrations of environmentally relevant pharmaceutical products sold across Norway, providing a useful snapshot of pharmaceutical pollution for our and others’ work. In particular, it will provide a useful resource for the characterisation of their environmental risk – on which our work is currently ongoing ( ECORISK 2050 Deliverable D6.2).

Methods

Classifications and grouping of pharmaceuticals

The classification of pharmaceutical substances for human and veterinary use is standardised by the World Health Organization (WHO) under the Anatomical Therapeutic Chemical/Defined Daily Dose (ATC/DDD) code system (RRID:SCR_000677). An ATC code ( Figure 1a) is a seven or eight character tiered alphanumeric code based on the target organ, therapeutic indication and/or pharmacology, and chemical structure of substances, while a DDD is defined as the average maintenance dose for a drug used in its main indication in adults. The ATC system’s widespread global use since the 1970s make it a useful tool for the broad classification of drugs within the Norwegian Drugs Wholesale Database.

Figure 1. Relationships between APIs and ATC codes.

( a) An example of the ATC code for paracetamol taken as an analgesic (N02BE01), ( b) one ATC code can represent multiple APIs – in this example, N02BE51 represents a combination of paracetamol and ibuprofen, ( c) one API can have more than one ATC code, paracetamol is represented here by three codes—N02BE01, N02BE51 and N02BE71—corresponding to the forms and indications it is sold under in Norway. API, Active Pharmaceutical Ingredient; ATC, Anatomical Therapeutic Classification.

ATC codes serve principally as a tool for drug utilization monitoring and research and are difficult to adapt to a substance-driven ecotoxicological approach. APIs are a more relevant entity for the characterisation of environmental risk, as ecotoxicological information is available for individual APIs rather than pharmaceutical products or ATCs. Under the ATC system, a product is characterised by a single ATC code that can contain multiple APIs, which are taken as a cocktail in the same pharmaceutical product ( Figure 1b). Conversely, one API can be used for treatment of diverse disorders of different organs and thereby be associated with different ATC codes ( Figure 1c). This complex set of many-to-many relationships between APIs and ATCs poses a distinct challenge for their interconversion, requiring a great deal of manual cross-referencing of products.

Publications of pharmaceutical sales from WHO Collaborating Centre for Drug Statistics Methodology and the NIPH are given in DDDs, limiting their utility for ecotoxicology work. DDDs aid comparison between pharmaceuticals consumption independent of price, package size and strength, but are impractical for ecotoxicological studies in which the weights of APIs sold are needed and are not always available for individual APIs or combinations of APIs.

Consequently, we elected within our dataset to calculate from scratch overall sales weights for each API, as a proxy of the emission of APIs. This required the assessment of each recorded sold product to determine the mass of each API in the product. The calculation of the total API emission per year is based on (1) the strength of the product ( i.e., the API concentration in units such as mg/L, mg/g, or mg/pill), (2) the amount of the product sold in one package (in units such as L, g, or no. of pills per package) and (3) the number of packages sold per year. See Table 1 for a summary of product and API vocabulary.

Table 1. Specific definitions of vocabulary used in this paper.

Vocabulary	Definition
ATC code	Anatomical Therapeutic Classification Code, a code classifying APIs or groups of APIs based on their medical use, target human organ, chemical structure, etc.
API	Active Pharmaceutical Ingredient, the therapeutic chemical(s) in a pharmaceutical product
Combination drug	A single product containing more than one API
Item	The components of a package, such as individual pills, dispensed sprays of an inhaler, etc.
Package	A single sold unit of product, such as a packet of multiple sheets of pills, a flask of liquid, etc.
(Pharmaceutical) Product	A specific manufacturer’s pharmaceutical, as sold, by unique product ID
Strength	The amount of a given API in an Item, Package or Product
Unit	The unit assigned to a given Strength, such as mg L -1, mg pill -1, International Units, etc.
DDD	Defined Daily Dose, “the average maintenance dose per day for a drug used in its main indication in adults” ( WHOCC, 2018), a standardised unit per ATC code and route of administration used to give a rough estimate of consumption.

Active Pharmaceutical Ingredients

Most—more than 50% in 2007—APIs are sold as pharmaceutical salts, with positive or negatively charged ions appended to their structures to increase stability and solubility in water ( Bastin et al., 2000; Paulekuhn et al., 2007). Where the given mass of API in a product in fact refers to the salt form, this can lead to over-estimation of the total volume of active substance sold, especially where the ion represents a substantial portion of the overall weight. Information on the salts used in each product was not listed in the source data. However, we aim to include an assessment of the effects of salts on PECs in future analyses of the data.

Data sources and management

Sales data for years 2016–2019 were extracted from the Norwegian Drugs Wholesale Database ( Figure 2, Figure 3a, Sales data), covering all sales to pharmacies, hospitals, nursing homes, and non-pharmacy outlets licensed to sell drugs within Norway, including prescriptions, over-the-counter sales, and procurement by medical establishments ( NIPH, 2019). In its raw form this dataset consisted of per-product sales, such as a packet containing multiple sheets of pills, or a suspension of liquid medicine.

Figure 2. Diagram of information sources to FHI Norwegian Drug Wholesale Statistics and Norwegian Prescription Database.

Figure was reproduced and adapted from Sommerschild et al. (2021a) with permission from the publisher. The Norwegian Prescription Database is, at time of writing, in the process of being renamed to the Norwegian Prescribed Drug Registry.

Figure 3. Simplified diagram of data extraction and management pipeline.

Data sourced from NIPH denoted by dashed blue box, data and code to be made publicly available denoted by the dashed orange box. NIPH, Norwegian Institute of Public Health.

In adherence with NIPH’s commercial confidentiality requirements, sales in currency values, and commercially sensitive information on the sales of individual manufacturers’ products were removed from the final published dataset.

Additional information on individual products that was required for calculating the sales weight per API ( Figure 3a, Product information), including number of items per package, strength (concentration of API per item), and associated unit were obtained separately from the centralised NIPH sales database and matched to sales data using internal product codes. In a sizable number of cases, no additional data were available for given products, automatic matching failed, or the data available were inappropriate for use in our workflow. Here records were checked manually against product contents records online, principally the Norwegian pharmaceuticals specialties site Felleskatalogen, the UK Electronic Medicines Compendium, and the US site Drugs.com. Cases where one product contained two or more APIs (combination drugs) were split into separate entries for each API to ensure substances were fully accounted for.

Although efforts were made to include the sales of as many products as possible, products with sales below 1000 packages over the four-year period, except for categories of special interest (antibiotics, sex hormones), were excluded as a time-saving measure. Additionally, gas APIs (such as anaesthetic gases) were likewise excluded.

The two primary data sources, and supplementary product information where gaps were present in the former, were imported into a Microsoft Access database and organised into a related set of tables. The main table types were data tables, conversion tables, and code lists. The main data tables are shown in Figure 4 and described below.

Figure 4. Simplified diagram of database structure: the main data tables.

API, Active Pharmaceutical Ingredient; ATC, Anatomical Therapeutic Classification; PNEC, Predicted No-Effect Concentration.

• 1)t_Product: the description of each pharmaceutical product (identified by product number), including information on the product type and the product amount per package ( Table 2)
• 2)t_Product_API: the concentration of each API per item and the total amount of API per package of the product ( Table 3)
• 3)t_Sales_Product: the number of packages sold per product per year ( Table 4)

Table 2. Field names, types, and descriptions for the Product Table t_Product.

Field name	Data type	Description
ProductCode	Number	Database internal unique product ID
ProductName	Short Text	Full product name from NIPH records
ProductName_short	Short Text	Product name with medium/dose removed
ATC_code	Short Text	Full ATC Code
ProductDetails	Short Text	Additional medium/dose data from ProductName
ProductType	Short Text	Standardised medium: pill, fluid, etc.
PackageQuantityValue	Number	Quantity of medium per package (number of pills, L of fluid, etc.)
PackageQuantityUnit	Short Text	Unit of medium per package
NoOfAPI_PerProduct	Number	Number of APIs in a product

NIPH, Norwegian Institute of Public Health; ATC, Anatomical Therapeutic Classification; API, Active Pharmaceutical Ingredient.

Table 3. Field names, types, and descriptions from the API per Product Table t_Product_API.

Field Name	Data Type	Description
ProductCode	Number	Database internal unique product ID
API_name	Short Text
StrengthValue	Number	Original strength information from NIPH (not standardised)
StrengthUnit	Short Text	Original strength information from NIPH (not standardised)
API_ConcentrationPerItemValue	Number	Converted API strength value (with standardised unit if possible)
API_ConcentrationPerItemUnit	Short Text	Standardised API strength unit (if possible)
API_AmountPerPackageValue	Number	Calculated API amount value (with standardised unit if possible)
API_AmountPerPackageUnit	Short Text	Standardised API amount unit (if possible)
Comment	Short Text
Exclude	Short Text	Yes (if record should be excluded from extraction)

NIPH, Norwegian Institute of Public Health; API, Active Pharmaceutical Ingredient.

Table 4. Field names, types, and descriptions from the Product Sales Table t_Sales_Product.

Field Name	Data Type	Description
sYear	Number	Sales year
ProductCode	Number	Database internal unique product ID
NoOfPackagesSold	Number	Number of packages of a unique product sold

Information on APIs in a given product was not available in the original data sources but had to be extracted from the ATC codes associated with the sales data. In some cases, extracted data corresponded directly to an API, but for combination products, and ATC codes where the included APIs were not immediately interpretable, API content was determined, stored, and converted at the individual product level. Ultimately, for each product ( Table 2), the associated API names associated were extracted from the full ATC name and entered in the table t_Product_API ( Table 3).

In most cases the information needed for calculating the amount of API per package (the concentration of API in the product and the amount of the product per package) was available in the original data source (the product information table). In some cases, where this information was not provided, it was still possible to extract the information manually from the product name.

For products where API information could not be found in the included data, it was instead sourced for each individual product from the Norwegian pharmaceutical specialties website Felleskatalogen or Summaries of Product Characteristics (SPCs) from the pharmaceutical specialties websites of other nations ( Electronic Medicines Compendium (UK), Pharmaceutical Specialties in Sweden, Medical Online Information Centre (Spain)). This was also the case for combination products containing two or more APIs, which typically required further work to determine and confirm the APIs present.

The resulting many-to-many relationship between ATC and API (see Figure 1) is represented by the code lists and junction tables shown in Figure 5.

Figure 5. Diagram of code lists and conversion tables.

Defines the many-to-many relationships between ATC and API in database. ATC, Anatomical Therapeutic Classification; API, Active Pharmaceutical Ingredient.

Finally, the information on yearly sales (number of packets) per product was stored in the table t_Sales_Product ( Table 4). During data extraction ( Figure 3d), This yearly sales information was combined with the calculated amount of API per product package, to obtain the total amount of API per year from the sales data.

Data processing in R

Data extracted from the Access database ( Figure 3d) were subsequently exported into flat files ( Figure 3e) for calculation of PECs and future analysis. For this purpose, the records were grouped by API and year and the calculated amount sold aggregated by sum. The exported dataset was prepared for analysis and publication in R version 4.1.2 “ Bird Hippie” ( R Core Team, 2021; RRID:SCR_001905). A full list of the R packages used is available as Underlying data ( Welch et al., 2022).

Sales weights per product per year were filtered to remove any zero values, and values for which no units were assigned, representing records for which the API amount could not be calculated. Sales weights were then summed by API, per year, and APIs were filtered according to a list of exemptions from risk assessment on the basis of non-toxicity (as applies to vitamins, vaccines, antibodies, etc. ( EMA, 2006)). Unique products excluded at each state are illustrated in Figure 6, and the total number of entries input (unique products) and APIs output are summarised in Table 5. The final dataset is published as a comma-separated values (.csv) file.

Figure 6. Records retained/removed at each stage of data processing.

Count of unique products sold in 2019 retained and removed at each step of data processing, categorised as human (upper) or veterinary (lower).

Table 5. Table of number of unique human and veterinary products input from starting dataset ( Figure 3e) and number of unique API output, by year.

	Starting dataset entries		Unique APIs
Year	Human	Veterinary
2016	5,713	660	804
2017	5,904	655	820
2018	5,991	611	820
2019	6,034	597	831

API, Active Pharmaceutical Ingredient.

Graphics. Graphs were rendered in R (see repository for code and packages used ( Welch et al., 2022)). Diagrams were drawn in Adobe Illustrator (RRID:SCR_010279), with the exception of Figure 6, which was rendered by the website SankeyMATIC.

Data evaluation

The predicted sales weights in this dataset were compared to similar datasets gathered by both co-authors in NIPH and other Norwegian agencies ( Table 6) in order to detect discrepancies and quality assure PECs. Although the primary output of this data paper is PECs, their limited availability made it more practical to carry out comparisons at the sales weights level, particularly as the choice of variables in the calculation of PECs is a question of judgement and conservatism as well as mathematics.

Table 6. Summary and labelling scheme for datasets used and referenced in this paper.

Label	Source	Type	Output format	Years used (Total coverage)	Reference
Welch	NIPH	Wholesale	g/API	2016–19	DOI: https://doi.org/10.17605/OSF.IO/GMX58
Felleskatalogen	FK	Wholesale	g/API	2018	Felleskatalogen, 2022
NorPD	NIPH	Prescription	DDDs	2016–19 (2004–20)	NIPH, 2021
Grung	NIPH	Wholesale	DDDs & g/API	2005	Grung et al., 2008
NIPH	NIPH	Wholesale	DDDs	2007–19	Sakshaug et al., 2013; Sakshaug et al., 2018; Sommerschild et al., 2021b

NIPH, Norwegian Institute of Public Health; NorPD, The Norwegian Prescription Database; API, Active Pharmaceutical Ingredient; DDD, Defined Daily Dose.

The Norwegian Pharmaceutical Specialties website Felleskatalogen maintains a rolling risk assessment on a yearly basis of pharmaceutical risk, using sales data from private market research. In order to benchmark the completeness and accuracy of our dataset, we compared our calculated sales weights to theirs.

Comparisons were performed using a Bland-Altman plot, also known as a Tukey mean-difference plot ( Bland & Altman, 1999), which allows for the visual comparison of two measurements of a single parameter.

Further comparisons were conducted between our dataset and prescription data for a high-use subset of APIs. In addition to the Drug Wholesale Statistics database, NIPH also maintains The Norwegian Prescription Database (NorPD), the Norwegian Prescription Database ( NIPH, 2021). NorPD is a publicly available resource, comparable to those available in other nations, that can produce reports of drug consumption by age, region, sex, and year across Norway. However, as a record of prescription this database is necessarily more limited than the Drug Wholesale Statistics database; additionally, all sales are recorded only in DDDs, introducing inaccuracy compared to actual quantities sold, and excluding drug formulations for which no DDD has been assigned. A further Bland-Altman plot was created to compare prescription and wholesales predicted sales weights.

Lastly, we compared our predicted sales weights to two further analyses based on the same dataset. An analysis of 2005 API sales weights for a panel of 11 APIs was conducted by Grung et al. (2008); we selected three high-use APIs with a wide range of constituent ATC codes—paracetamol, ethinylestradiol and ibuprofen—and compared these sales weights with our predictions for 2016–19.

To further benchmark trends in consumption, these sales weights were normalised by dividing the figures by the annual population of Norway. They were then compared to wholesale data published by NIPH – available as PDF reports ( Sakshaug et al., 2013; Sakshaug et al., 2018; Sommerschild et al., 2021b) of consumption in DDDs per thousand people per day for a limited range of substances. Although direct comparisons between normalised sales weights and DDD/1000 people/day were not possible, we were able to compare overall trends in consumption to look for disagreement.

Predicted Environmental Concentrations

PECs of individual APIs in the compartment Surface Water were calculated using a modified form ( Equation 1) of the standard refined PEC _SW equation, with default variables ( Table 7), outlined in the EMA’s guidelines for pharmaceutical environmental risk assessment (2006). As no specific bodies of water are specified in the guidelines, the model is assumed to apply to all relevant freshwater bodies, i.e., rivers and lakes.

Table 7. Table of PEC _SW equation default variables and parameters.

Component	Unit	Description
g of API sold	g year -1	The total weight (g) of an API sold in a year
WWTP removal	unitless	The proportion of the API removed at WWTP (default of 0)
365	days year -1	The number of days in a year
Wastewater consumption	L person -1 day -1	The average wastewater consumption (L) of the population of a given area per day
Population	persons	The population of a given area
Dilution factor	unitless	The ratio of dilution between WWTP effluent and receiving waters (default of 10)

PEC, Predicted Environmental Concentrations; API, Active Pharmaceutical Ingredient; WWTP, Wastewater Treatment Plant.

Equation 1.

$$PE{C}_{SW}=\frac{APIsold\times (1-WWTPRemoval)}{365\times Wastewaterconsumption\times Population\times Dilutionfactor}$$

PEC, Predicted Environmental Concentrations; API, Active Pharmaceutical Ingredient; WWTP, Wastewater Treatment Plant.

As mentioned, the standard equation estimates sales weights from the maximum dose of a given API and the proportion of people in a population taking that API. By contrast, by using our dataset of pharmaceutical wholesales we can input a more exact figure for consumption across the entire population of Norway. Default values for removal in wastewater treatment plants (0% removal) and dilution factors (dilution to 1 part in 10 upon entering receiving waters) were still retained as conservative assumptions, potentially contributing to overestimation of PECs.

PECs were individually calculated per API, per year, using information on yearly average wastewater generation and Norwegian population, obtained from Statistics Norway and included as Underlying data ( Welch et al., 2022).

Identification and grouping of APIs

To aid in the contextualisation and machine reading of the dataset, additional data were collected and appended to API sales data. Firstly, standard InChIKeys, a short, unique string based on molecular structure, were, where possible, found for all APIs ( Heller et al., 2015) using the R package webchem ( Szöcs et al., 2020) (RRID:SCR_017684) to look up API names via the Chemical Translation Service ( Wohlgemuth et al., 2010) (RRID:SCR_014681).

Additionally, APIs were sorted into single categories based on function and/or target organ (antidepressant, respiratory, antibacterial, etc.), adapted from ATC classifications and sourced from Felleskatalogen, Drugs.com, and WHOCC for Drug Statistics records. A short description of the type and application of APIs was also included, based principally but not exclusively on use in Norway.

Potential applications

Reported measurements of environmental concentrations of pharmaceuticals in the environment are generally highly specific to a few APIs, locations, time periods and abiotic conditions. By providing calculated wholesale weights and thereby a proxy of emission for an unusually broad range of APIs with an extensive geographic coverage (all of Norway), we are providing a resource for the easy approximation of upper and lower bounds of potential environmental exposure. Further processing and analysis of this dataset will permit a better understanding of the potential exposure and risks of pharmaceuticals to the environment ( Welch et al., 2021, ECORISK 2050 Deliverable D6.2). In particular, a ranking of which pharmaceuticals provide the greatest contribution to overall environmental burdens will facilitate targeted prioritisation and mitigation.

As a resource for wide-scale prediction of PECs for the years 2016–2019, this dataset also lays the groundwork for the retrospective prediction of PECs from extensive machine-readable wholesale drug records from NIPH back to 1999. This extended time series will make it possible to examine how temporal trends in pharmaceutical sales correlates with environmental and demographic shift, to the eventual end of predicting future environmental exposure and risk of pharmaceuticals and support pre-emptive mitigation decisions.

At the European level, various market research agencies collect pharmaceutical sales data for private consumption. Although repurposing commercial market research is an undoubtedly useful approach, dissemination and transparency in such cases is difficult, as such agencies have a strong financial interest in keeping their work private and limiting its access to paying customers.

By contrast, the use of publicly available wholesale data has largely so far been an targeted effort, limited to a subset of APIs, from a high of 300 in 2014 in Japan ( He et al., 2020), to 100 in China ( Li et al., 2019), or 165 APIs consumed by the elderly in Catalonia ( Gómez-Canela et al., 2019). To the authors' knowledge, our work is the first attempt at creating a full dataset of all potentially ecotoxicologically relevant APIs in a nation. As existing prioritisation efforts are often limited by inconsistent coverage of toxicity data, this approach allows for a more comprehensive coverage of groups characterised by high environmental risk—such as sex hormones—even when toxicity data are not available for all constituent substances.

One notable limitation of this work compared to the above studies is our conscious ignorance of drug behaviour in both the human bodies and Wastewater Treatment Plants (WWTPs). Due to the scope of the prediction effort, we elected not to focus on these processes and instead assume no removal or transformation. This will likely drive an overestimation of PECs, especially where APIs are extensively metabolised or effectively removed by treatment. However, for a high-priority subset of substances, we hope to explore the effects of these stages of the consumption-to-environment pathway more closely. The prediction of API Risk Quotient, combing publicly available toxicity data with our dataset, will allow for such high-priority APIs to be identified. We hope to publish this work in the near future.

Additionally, the authors are currently developing Bayesian network models for probabilistic risk calculation based on a subset of this data. Such models can integrate quantified spatial and temporal variability in exposure, allowing the uncertainty integral to the prediction of environmental concentrations under a variety of future scenarios to be retained and quantified.

The processing of this dataset has required a considerable investment of time and resources. This is, compared to repurposing market research data, a less efficient approach. Due to the relative transparency and comprehensiveness of our approach and the ability to make the output exposure data publicly available, we believe this dataset surpasses any previous resource available for Norway. On an international level, we hope that by providing a resource that can be compared with more readily available prescription data, we can provide a benchmark for the underestimation of exposure from such resources.

We hope that this dataset and information on data processing will prove a useful resource for other researchers, providing a means to calculate a proxy for realistic environmental concentrations for use as a benchmark in studies of pharmaceutical pollution in Norway, and countries with similar drug consumption, environmental characteristics, and/or national wholesales data available.

Data evaluation

Comparison with Felleskatalogen data

Figure 7 summarises agreement between the two datasets for the year 2018. A mean difference (blue line) extremely close to zero on the y-axis indicates little average difference between calculations. However, a number of substances below the lower red line (95% CI) indicate potential errors in either our or Felleskatalogen’s calculations ( Table 6). In total, Felleskatalogen sales weights are available for 203 APIs, of which 193 have available toxicity data in the form of Predicted No Effect Concentrations (PNECs), while our dataset contains sales weights for 821 APIs, 255 of which have available PNECs.

Figure 7. Comparison between NIPH-derived and Felleskatalogen datasets, for sales in 2018.

Bland-Altman or Tukey mean-difference plot ( a) of difference (y axis) and mean (x axis) of log10-transformed sales weight data from our and Felleskatalogen sources. Blue line marks mean difference, and red 95% Confidence Intervals. A substance with no difference between the two predicted weights would fall on the 0 line on the y axis. Also included is a dot plot of APIs only calculated in our data ( b) graphed across log10 sales weight. NIPH, Norwegian Institute of Public Health; API, Active Pharmaceutical Ingredient; PNEC, Predicted No-Effect Concentration.

Of these, discrepancies between figures for ethinylestradiol and levonorgestrel are due to the mistaken substitution of milligrams (mg) for micrograms (mcg or μg) for one combination product containing levonorgestrel and ethinylestradiol in Felleskatalogen data and have consequently been excluded from summary statistics. Differences in sales of salicylic acid may be due to its presence in a number of non-medical skin products not included in NIPH data, and/or from the combination of the weights of salicylic acid and 5-aminosalicylic acid, treated as separate APIs in our data. The discrepancy for levofloxacin between our data (54 g) and Felleskatalogen (3854 g) is likely due to the exclusion of eye drops containing the antibiotic from the NIPH source data, while no explanation was found for the difference in vildagliptin, 37008 g compared to 4376318 g.

Comparison with prescription data

To assess the value of our dataset compared to NorPD ( Table 6), we compared predicted sale weights for six substances ( Table 8) present in both datasets, a selection of common human, veterinary, over the counter (OTC) and prescription APIs, for the year 2019 ( Figure 8).

Figure 8. Bland-Altman or Tukey mean-difference plot of difference (y axis) and mean (x axis) of log10-transformed sales weight data from our and NorPD sources for six selected APIs in 2019.

Blue line marks mean difference, and red 95% Confidence Intervals. A substance with no difference between the two predicted weights would fall on the 0 line at the centre of the y axis. NorPD, The Norwegian Prescription Database; API, Active Pharmaceutical Ingredient; OTC, over the counter; PNEC, Predicted No-Effect Concentration.

Table 8. Panel of human and veterinary drugs selected for comparison between our dataset and NorPD.

Where multiple DDD values were possible for one ATC code, the highest value was used. Codes beginning with Q correspond to veterinary applications. Inj. refers to injected forms of drug, vag. to vaginal.

API	Description	Availability	ATC Codes	DDD	Notes
Paracetamol	Human painkiller	OTC & Prescription	N02AJ06 N02BE01 N02BE51	3.0 g (oral) 3.0 g (oral) 3.0 g (oral)	High consumption
Ibuprofen	Human painkiller	OTC & Prescription	M02AA13 C01EB16 M01AE01	N/A 0.03 g (oral) 1.2 g (oral)	High consumption
Xylometazoline	Human nasal decongestant	OTC & Prescription	R01AA07 R01AB06	0.8 mg (nasal) N/A	High consumption
Amoxicillin	Human & vet. antibacterial	Prescription	J01CA04 J01CR02 QJ01CA04	1.5 g (oral) 3 g (inj.) 1.5 g (oral) 3 g (inj.) N/A	Significant consumption
Progesterone	Human & vet. sex hormone	Prescription	G03DA04 QG03DA04	30 mg (oral) 5 mg (inj.) 90 mg (vag.) N/A	High consumption
Atorvastatin	Human statin	Prescription	C10AA05 C10BA05	20 mg (oral) N/A	2 nd most used prescription
Metoprolol	Human beta blocker	Prescription	C07AB02	0.15 g (oral)	9 th most used prescription

NorPD, The Norwegian Prescription Database; API, Active Pharmaceutical Ingredient; DDD, Defined Daily Dose; ATC, Anatomical Therapeutic Classification; OTC, over the counter; N/A, not applicable.

Comparing wholesale and prescription sales weights for these substances ( Table 8), it can be seen that on average, prescription data predicted lower sales weights for APIs, but this was driven by the decongestant xylometazoline, whose sales weight was predicted to be around 1000 times higher than prescription weight. The OTC and prescription painkillers paracetamol and ibuprofen had a sales weight of roughly 1.5 times and 2.3 times wholesale than prescription.

The prescription-only APIs metoprolol and atorvastatin showed strong agreement between wholesale and prescription weights (<10% difference), while amoxicillin and progesterone were predicted a 45% and 28% higher prescription weight than sales weight. In both cases, this is likely due to the difficulty of distinguishing the appropriate DDD to use with prescription data, as it does not distinguish between routes of admission at the ATC code level, and the highest DDDs for these substances are 2–3 times higher than the lowest.

Comparison with Grung et al., 2008 and NIPH Wholesale Report Data

Predicted sales weights, normalised by population, were also compared to earlier (2005) ( Table 6) predictions and (non-comprehensive) published trends in consumption by DDD. Comparing our predictions of paracetamol sales weights to those in 2005 ( Figure 9) shows a plausible growth in normalised consumption, the majority of which is driven by growing consumption in plain paracetamol over time.

Figure 9. Comparison of predicted sales data sources for paracetamol and paracetamol-containing products.

( a) Calculated sales weights, by ingredient, for products containing paracetamol in 2005 and from 2016–19, normalised by annual population of Norway. ( b) Consumption of paracetamol-containing products by ingredient from NIPH published reports, in DDD per 1000 people per day. The combination “paracetamol + non-psycholeptics” corresponds to combinations of paracetamol with caffeine, acetylsalicylic acid, or ibuprofen. NIPH, Norwegian Institute of Public Health; DDD, Defined Daily Dose.

Consumption of ibuprofen ( Figure 10) is also driven by the consumption of ibuprofen as a painkiller (variously classified as M01AE01 (oral/rectal/injected) and M02AA13 (topical)). Drawing direct comparisons between different combinations of the API is difficult due to changes in API encoding, patchy data availability in Wholesale Reports, and the disappearance of dexibuprofen, an enantiomer of ibuprofen. Nevertheless, in overall trends, a similar pattern of overall decline offset by a small bump in 2017 can be observed.

Figure 10. Comparison of predicted sales data sources for ibuprofen and ibuprofen-containing products.

( a) Calculated sales weights, by ingredient, for products containing ibuprofen in 2005 and from 2016–19, normalised by annual population of Norway. ( b) Consumption of ibuprofen-containing products by ingredient from NIPH published reports, in DDD per 1000 people per day. NIPH, Norwegian Institute of Public Health; DDD, Defined Daily Dose.

Interpreting individual sales patterns for ethinylestradiol, also known as EE, is harder than the above due to the wide range of combination contraceptives and hormone therapies. An overall trend of decline in consumption in Figure 11a can be seen, driven by small decreases in constituent consumption, but in Figure 11b it is less apparent whether the trends of different compositions balance each other out. Historical data on ethinylestradiol consumption was absent in more dated Wholesale Reports ( Sakshaug et al., 2013; Sakshaug et al., 2018), except in the case of vaginal rings, where consumption was given in units sold in one report and DDD in the next, making comparisons difficult. Nevertheless, trends for individual combinations that appear in both datasets – EE and levonorgestrel (in fixed static doses), vaginal rings containing EE and etonogestrel, and EE and cyproterone showing corresponding trends.

Figure 11. Comparison of predicted sales data sources for ethinylestradiol and ethinylestradiol-containing products.

( a) Calculated sales weights, by ingredient, for products containing EE in 2005 and from 2016–19, normalised by annual population of Norway. ( b) Consumption of EE-containing products by ingredient from NIPH published reports, in DDD per 1000 people per day. Fixed and sequential ingredients refer to a course of pills of either a fixed dose, or a changing (sequential) dose. NIPH, Norwegian Institute of Public Health; DDD, Defined Daily Dose; EE, ethinylestradiol.

Checking for extreme changes

In addition to the above comparisons of our data with similar datasets, we elected to compare sale weights by API internally to detect outliers. Sale weights per year were compared to a mean weight over the sales period, and APIs for which at least one year’s sales weight was more than 10 times greater than the mean were highlighted. The substances are graphed in Figure 12.

Figure 12. Calculated sales weights 2016–2019 for APIs where at least one year’s weight is 10x bigger or smaller than the mean API sales weight.

A total of 31 APIs were shortlisted under this criterion; see Table 9 for further details. Coloured by type. API, Active Pharmaceutical Ingredient.

For these 31 shortlisted APIs, registration and deregistration dates were checked, where available, to determine if changes in consumption could be explained by regulatory status. As products, and therefore product API content tend to remain consistent over the 2016–19 period, the above changes are expected to represent actual changes in consumption. However, it was considered prudent to check medical and pharmacy literature for possible explanations, nevertheless ( Table 9).

Table 9. Shortlist of APIs where at least one year’s weight is 10× bigger or smaller than the mean.

API name	Type	Description	Comments
altrenogest	sex hormone	veterinary birth control	New formulation (“Altresyn Ceva” authorised in Norway 2018 ( Statens legemiddelverk, 2022)
asenapine	antipsychotic	atypical antipsychotic for schizophrenia and bipolar disorder	Sole product (“Syncrest”) deregistered 2017 ( Felleskatalogen, 2022)
carglumic acid	metabolic	carbamoyl phosphate synthetase inhibitor for hyperammonaemia	Two products, one of which (“Ucedane”) was first authorised in June 2017 ( Felleskatalogen, 2022)
cefalotin	antibacterial	beta-lactam cephalosporin antibiotic	Shortage of cefalotin in Norway recorded 2019 ( Antibiotika.no, 2019)
cladribine	antineoplastic	antimetabolite and immunosuppressant for multiple sclerosis and leukaemia	Authorised August 2017 ( Felleskatalogen, 2022)
cobimetinib	antineoplastic	mitogen-activated protein kinase inhibitor for melanoma	Authorised November 2015 ( Felleskatalogen, 2022)
cyclizine	antiemetic	piperazine antihistamine for nausea relief from motion sickness, vertigo	Cause of change unknown
dacarbazine	antineoplastic	alkylating agent for skin cancer and lymphoma	Authorised March 2017 ( Felleskatalogen, 2022)
dasabuvir	antiviral	antiviral used in combination for treatment of hepatitis C	Manufacturer withdrew application for dasabuvir/ ombitasvir/paritaprevir/ritonavir in 2016 ( Nye Metoder, 2016); however, ritonavir is also available alone
edoxaban	antithrombotic	Factor Xa inhibitor for clotting reduction for strokes, atrial fibrillation, DVT	Authorised June 2015 ( Felleskatalogen, 2022)
eluxadoline	antidiarrheal	treatment for diarrhoea from IBS	Authorised as reimbursable prescription 2017, withdrawn from market 2019 ( Statens legemiddelverk, 2017; Felleskatalogen, 2022)
fomepizole	antidote	antidote to methanol and antifreeze poisoning	Cause of change unknown
gadobenic acid	diagnostic agent	gadolinium contrast agent used for magnetic resonance imaging	Cause of change unknown
gadodiamide	diagnostic agent	gadolinium contrast agent used for magnetic resonance imaging	Deregistered 2018 ( Felleskatalogen, 2022)
glecaprevir	antiviral	protease inhibitor used in combination with pibrentasvir for hepatitis C	Glecaprevir/pibrentasvir (“Maviret”) Authorised July 2017 ( Felleskatalogen, 2022)
ixazomib	antineoplastic	proteasome inhibitor for multiple myeloma	Authorised November 2016 ( Felleskatalogen, 2022)
nitrofurantoin	antibacterial	antibiotic for bladder infections	Shortage recorded from 2018–2021 ( VG, 2019)
nystatin	antifungal	topical antifungal	Cause of change unknown
ombitasvir	antiviral	antiviral taken with paritaprevir and ritonavir for hepatitis C	See dasabuvir
osimertinib	antineoplastic	tyrosine kinase inhibitor for non-small cell lung cancer	Authorised February 2016 ( Felleskatalogen, 2022)
palbociclib	antineoplastic	selective cyclin-dependent kinase inhibitor for breast cancer	Authorised November 2016 ( Felleskatalogen, 2022)
paritaprevir	antiviral	combination treatment for hepatitis C	See dasabuvir
pibrentasvir	antiviral	antiviral used in combination for hepatitis C	See glecaprevir
prednisone	steroid	corticosteroid and immunosuppressant for many immune and allergic disorders	Shortage recorded 2019 ( Statens legemiddelverk, 2019)
safinamide	dopaminergic	MAO inhibitor for Parkinson's	Authorised February 2015 ( Felleskatalogen, 2022)
toceranib	antibacterial	receptor tyrosine kinase inhibitor for canine cancers	Deregistered 2019 ( Felleskatalogen, 2022)
tofacitinib	immunosuppressant	treatment for arthritis, ulcerative colitis	Authorised March 2017 ( Felleskatalogen, 2022)
velpatasvir	antiviral	NS5A inhibitor for hepatitis C	Sofosbuvir/velpatasvir (“Epclusa”), Sofosbuvir/ velpatasvir/voxilaprevir (“Vosevi”) authorised July 2016 ( Felleskatalogen, 2022)
venetoclax	antineoplastic	treatment for leukaemia	Authorised December 2016 ( Felleskatalogen, 2022)
vinflunine	antineoplastic	alkaloid derivative for bladder cancer	Cause of change unknown
voxilaprevir	antiviral	protease inhibitor for hepatitis C	See velpatasvir

API, Active Pharmaceutical Ingredient; DVT, deep vein thrombosis; IBS, irritable bowel syndrome; MAO, monoamine oxidase; NS5A, nonstructural protein 5A.

Stark changes largely corresponded with recorded changes in marketing authorisation (23 substances, 74.1%). Use in some APIs appears to result from shortages in supply (three, 12.5%), while the remaining five (16.1%) were not immediately explicable. These latter substances were then re-checked in source data, no errors were found between years. In three cases, where 2018 sales weights were available from both our and Felleskatalogen data (osimertinib, gadobenic acid and edoxaban), both predictions were in close agreement (<10% difference between values).

Data availability

Underlying data

Open Science Framework: Pharmaceutical pollution: Prediction of environmental concentrations from national wholesales data. https://doi.org/10.17605/OSF.IO/GMX58 ( Welch et al., 2022).

This project contains the following underlying data:

• -miljo-uttrekk_10.05.2019.xlsx (A Felleskatalogen spreadsheet, in Norwegian, of drug toxicity, persistence and bioaccumulation from 2018)
• -NO_EN_API_names.csv (A supplement to the above (author's own) of Norwegian and English API names)
• -API_toxicity_2019.xlsx (A spreadsheet (author's own) of the status of all drugs sold 2016-19 in Norway)
• -ATC_colour_codes.csv (A set of thematic colour codes per ATC level 1 used for graphs)
• -FHI_2016_2020_ATC_codes_DDD_etc.xlsx (A spreadsheet compiled from NIPH's report on drug sales in Norway 2016–20)
• -InChI_Shortlist.csv (A list of InChIKeys corresponding to APIs studied, saved and imported to reduce database calls)
• -KOSVREGesthushfo0000.xlsx (Wastewater consumption per person per day in Norway 2015–2020 (Statistics Norway))
• -Folkemengde.xlsx (Mainland Norwegian population on 1 Jan per year 1951–2021 (Statistics Norway))
• -NorPD_API_Subset.xls (A report exported from the Norwegian Prescription Database on prescriptions of a sample of APIs, 2016–2019)
• -DDD_conversion_factors.xlsx (Corresponding DDDs taken from the WHOCC ATC/DDD Index)
• -API_desc_short.xlsx (A list of all APIs with calculated PECs, sorted to broad categories of use and appended with a short description)
• -sales_by_API_year_processed_2022-04-12_09.35.csv (Sales weights per API per year)
• -Pipeline_Data_Draft.Rmd (R code)

Input data are available with the exception of the processed output of the Norwegian Drug Wholesale Statistics database. This dataset is not available due to NIPH confidentiality obligations to pharmaceutical manufacturers. A published summary of wholesale data can be found at https://www.fhi.no/en/publ/2021/drug-consumption-in-norway-2016-2020/; in addition to the contact details of relevant NIPH personnel.

Data are available under the terms of the Creative Commons Attribution 4.0 International license (CC-BY 4.0).

Ethics and consent

Ethical approval and consent were not required.

Funding Statement

This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Skłodowska-Curie grant agreement No 813124 and from NIVA’s Computational Toxicology Program (NCTP).