RxTrends: An R-based Shiny Application for Visualising Open Data on Prescribed Medications in Ireland

Data source and preparation

The HSE PCRS pharmacy claims data for Ireland was accessed via the PCRS’s Reporting and Open Data platform (available at https://www.sspcrs.ie/portal/annual-reporting/). We used their reports of monthly figures for the top 100 prescribed medications, ordered by their prescribing frequency or ingredient cost, as well as equivalent figures for total prescribing by therapeutic group, for the GMS scheme. We also used their reports on monthly numbers of eligible people for the GMS scheme. The current RxTrends application uses monthly prescribing data from January 2016 to October 2024 for the subset of the population who were eligible for the GMS scheme. GMS scheme data captures dispensing information for all reimbursed items for all eligible persons. We used this drug scheme for the initial development of this application, as for the Drugs Payment Scheme, data for any month only includes dispensings where household prescription medication expenditure exceeds a certain threshold. Additionally, Long Term Illness scheme data includes only medications that are related to specified health conditions, and only overall scheme eligibility is reported rather than by individual condition. Based on feedback following application launch, we also incorporated data from other drug schemes, however this paper focuses only on GMS scheme data where eligibility figures are reliable.

Downloading of data from the PCRS open data platform and initial processing were performed using JupyterLab, a web-based interactive computing environment for Python (https://jupyter.org). All code and packages utilised are available in the Data Availability section.

The monthly data are structured in a tabular format, detailing the medication or therapeutic group name, total monthly frequency of their prescriptions, their percentage share of total GMS prescriptions for the selected month, the overall ingredient cost of these prescriptions in euros, and their percentage share of total GMS prescription cost. Data on monthly GMS scheme eligibility for 2017–2024 were also accessed through the PCRS’s Reporting and Open Data platform. Eligibility figures for 2016 were directly requested via email correspondence with the PCRS’s Reporting and Open Data team. The eligibility data detail the number of eligible persons in each Community Health Organisation and Local Health Office area, broken-down by gender and age groups. This application utilises the total eligible number in all areas for each month.

The prescriptions data categorised under therapeutic groups and physiological systems are coded according to the World Health Organization (WHO) Anatomical Therapeutic Chemical (ATC) Classification System¹³. Additionally, ATC codes for medications were added to enhance the utility of the individual medications data. Therefore, this coding facilitates comparisons of related ATC codes at different levels. Certain medication groups under the "Various" physiological system (V) were reported as pharmacological subgroups (i.e. 3rd level ATC) within both the top 100 medications data and the therapeutic groups data (i.e. 2nd level ATC). These groups, such as urinary requisites, diagnostic agents, and needles, were excluded from the list of individual medications, and were only included in the application as part of the therapeutic group data. This was the more appropriate categorisation of these groups (as they were not individual medications), and the therapeutic group figures appear to provide a more comprehensive representation of the dispensing of these items.

Prescribing rates and cost rates were derived by divided values of prescribing frequency or cost of each medication, therapeutic group or physiological system by the total number of GMS scheme-eligible persons for the same month, and presented in rates per 1,000 eligible persons. The ingredient costs and related rate figures were also adjusted for inflation using the monthly Consumer Price Index (CPI), normalised to the most recent month in the data (i.e. October 2024). The monthly CPI figures, which track the average price change of consumer goods and services purchased by private households, were sourced from the Central Statistics Office (CSO) of Ireland¹⁴. In May 2021, A cyberattack on the health service of the Republic of Ireland resulted in transmission issues for pharmacy claims for that month¹⁵. As a result, there was a notable fall in the apparent dispensing for that month and an increase for June 2021, as most data for dispensing in May were transmitted in June¹⁵. The weighted mean of the metrics value (e.g. prescribing, cost) for the two months was calculated for each month. If a record was only available for one month (i.e. a medication was not part of the top 100 for one of the months), its value was utilised for that specific month and was not modified and applied to the other.

Shiny application development

Implementation

An interactive web-based Shiny application was developed to enable dynamic visualisation and provide descriptive analysis of prescribing patterns using data sourced from the HSE-PCRS. The application leverages a variety of R packages, including Shiny for the web framework, ggplot2 and plotly for graphical representations, and dplyr for data manipulation. The application’s architecture is divided into two main components: the user interface (UI) and server-side logic. The UI provides various interactive elements such as dropdown menus with search function for selecting medications, therapeutic groups, and physiological systems, and a choice of slider or dropdown menu for selecting date ranges. The server-side logic processes the PCRS data which is initially fed from an Excel file using the readxl package.

Upon application launch, the UI dynamically populates selection choices based on the data. Users can select specific criteria to filter the data, which are then visualised through a series of tabbed panels displaying prescribing frequencies, costs, and rates (see Table 1) using plotly-enhanced ggplot2 charts. Each plot visualises trends over time, and users can interact with the data points for more detailed information. Additionally, the application includes custom JavaScript via shinyjs to enhance the UI elements like date range sliders, improving the user experience by formatting dates and updating labels dynamically.

The RxTrends application's capabilities were extended by allowing users to select an unlimited number of entities across multiple categories, including medications, therapeutic groups, and physiological systems. The "Compare Selected" button in the application enables users to compare frequencies, costs, or rates over time when two distinct entities are selected. This feature allows the application's server-side logic to compute for the selected metric (e.g. prescribing frequency, cost of the medication/therapeutic group) the proportion of higher-level entity in the WHO ATC classification hierarchy (i.e. therapeutic group or physiological system) made up of the lower-level entity (i.e. medication or therapeutic group). These comparisons are dynamically visualised in the interactive plots powered by ggplot2 and plotly.

The application also features a download button allowing users to download data containing the selected inputs currently displayed for further analysis or use. Downloadable data is provided in Comma-separated values (CSV) format, including columns for the metrics listed in Table 1 and one row per month per selected medication, therapeutic group, and/or physiological system. A contextual help and documentation feature has also been embedded within the application, which describes output data format in detail.

During the development of RxTrends, the application underwent two refinement cycles, with team members (from clinical, epidemiology, and data science backgrounds) evaluating a prototype application and providing feedback on features, functionality, and presentation.

Operation

This RxTrends Shiny application was developed using R programming, version 4.3.3 (www.r-project.org) for Microsoft Windows 10 x64, with the following attached base packages: stats, graphics, grDevices utils, datasets, methods, and base. Other attached packages were dplyr (1.1.4), lubridate (1.9.3), DT (0.33), plotly (4.10.4), ggplot2 (3.5.1), openxlsx (4.2.5.2), readxl (1.4.3), readr (2.1.5), shinyjs (2.1.0), htmltools (0.5.8.1), and shiny (1.8.1.1).

This Shiny application is available online on shinyapps.io (https://frankmoriarty.shinyapps.io/RxTrends-RShiny/) and on the webpage for the project this work relates to (https://cdrx-project.eu/?page_id=90) to enable users to run it embedded within the website. Additionally, users can run this application locally in RStudio either by downloading the code underlying application and associated data (see Data Availability section), or by running it directly from GitHub using the following R command: shiny::runGitHub(repo = 'RxTrends-RShiny', username = 'moriarty-pharmacoepi', ref = "master")

Prescribing frequency of preferred drugs in Ireland

The Preferred Drugs Initiative in Ireland aims to identify a certain drug within a therapeutic drug class as the preferred choice¹⁶. This initiative offers prescribers clear guidelines on how to select, prescribe, and monitor the preferred drug for specific medical conditions¹⁶. The selection of the drug is based on a comprehensive assessment of several aspects including clinical effectiveness, dosage, drug interactions, adverse effects, cost, national prescription patterns, and clinical recommendations¹⁶. The HSE Medicines Management Programme has evaluated ten therapeutic classes and chosen a preferred drug within each, for example pantoprazole being the preferred proton pump inhibitor (PPI)¹⁶.

This application can be used to visualise the different medications within a therapeutic class, and evaluate trends in use of the HSE preferred drugs over time relative to other agents within the class. Users have the ability to choose various medications from a therapeutic group in order to observe the prescribing trend for each medication and the overall therapeutic group. Examining PPIs as one of the most commonly prescribed therapeutic classes, Figure 1 shows that esomeprazole is the most frequently prescribed PPI, ranging from 55.38 to 103.61 prescription dispensings per 1,000 eligible persons per month, with an increasing trend over time. Pantoprazole, the HSE preferred drug, was the third most frequently dispensed PPI at the start of the period of data coverage (a prescription rate of 32.82 per 1,000 in January 2016), however the rate of prescribing has increased and surpassed lansoprazole since 2021 to reach 60. 26 per 1,000 in October 2024. Esomeprazole is increasing as a proportion of prescribing within drugs for acid-related disorders (from 0.298 in January 2016 to 0.397 in October 2024, Figure 2), while pantoprazole (Figure 3) increased from 0.177 to 0.231. This indicates improved alignment with the preferred drug initiative recommendation for PPIs, but significant potential for further improvement.

Figure 1. Prescribing rates of individual Proton Pump Inhibitors (PPIs) and drugs for acid related disorders.

Figure 2. The proportion of prescribing of esomeprazole with all drugs for acid related disorders.

Figure 3. The proportion of prescribing of pantoprazole with all drugs for acid related disorders.

Cost implications of patent expiry and policy change

The expiry of patent protection and the availability of generic versions of off-patent medications can substantially affect medication expenditure. In Ireland, the Health (Pricing and Supply of Medical Goods) Act 2013 allows pharmacists to substitute medications which have been deemed as interchangeable by the Health Products Regulatory Authority (HPRA)¹⁷. ‘Interchangeable products’ are those medications that are judged to be suitable for substitution with another medication in the same group if they have the same qualitative and quantitative composition in each of their active substances, are in the same pharmaceutical form, have the same route of administration, and have no more than two active substances¹⁷. Subsequent to the introduction of interchangeable products, the HSE introduced a policy of reference pricing, which involves establishing a common reimbursement price, or reference price, for a group of interchangeable medications¹⁷. This is the price the HSE will reimburse each time a product from this group is dispensed; when products are priced at or below this reference price, patients do not face any additional expenses, however dispensing a more expensive product may result in patients having to pay the difference in price¹⁷.

In December 2019, the HPRA published an updated list that included interchangeable pregabalin and gabapentin medications for the first time¹⁸. The HSE than set reference prices effective from 1^st April 2020 for the groups of interchangeable pregabalin products¹⁹, and from 1^st May 2020 for gabapentin products²⁰, in both cases representing a substantial price reduction compared to the previous prices. Figure 4 illustrates the impact of allowing interchangeability between brands and then reference pricing for gabapentinoids. Following these changes, the ingredient cost per dispensing of prescribed pregabalin and gabapentin experienced a considerable decrease. Specifically, the cost for pregabalin dropped from €32.49 per prescription in March 2020 to €10.14 per prescription in April 2020, and the cost for gabapentin dropped from €21.98 per prescription in April 2020 to €11.95 per prescription in May 2020 (Figure 4).

Figure 4. Cost per prescribed unit of gabapentin and pregabalin.

As another measure to manage pharmaceutical costs, the HSE introduced a change to the reimbursement of lidocaine plasters, due to a level of use that far exceeded approved indications, indicating off-label prescribing for conditions other than post-herpetic neuralgia (PHN)²¹. This was implemented in two stages, the initial stage was implemented in September 2017 for new patients, necessitating approval for each patient diagnosed with PHN to be reimbursed under the GMS scheme for lidocaine plaster dispensing. This requirement was subsequently extended to pre-existing patients who had been receiving the drug prior to September 2017²¹. As shown in Figure 5, the monthly rate of lidocaine plaster dispensings per 1,000 eligible persons reduced from a high of 15.6 in August 2017 to 10.2 and 8.7 in September and October 2017 following the introduction of the first stage of the policy change. Data are not reported for lidocaine plasters for the period November 2017–April 2018, as it was not among the top 100 medications by frequency or cost, though from May 2018 it is again included in the top 100 by cost. Previous research has shown prescribing frequency decreased further in December 2017 with the introduction of the second stage of the policy change and has remained low since²².

Figure 5. Prescribing rate of lidocaine products (including lidocaine plasters).

Seasonal variation in medication prescribing

The RxTrends application provides a graphical analysis of prescription trends, presenting monthly data over the course of years, which allows potential seasonality in medication use to be evaluated. Figure 6 and Figure 7 clearly show a considerable increase in prescription rates of antibacterials for systemic use and drugs for chronic obstructive lung disease over the winter months, reaching a peak between December and January of each year. This spike is likely associated with the greater occurrence of infections and respiratory diseases during the colder months (as a possible trigger), requiring higher usage of these medications. The sharp seasonal peaks are less pronounced or absent for the winters of 2020/2021 and 2021/2022, likely due to social distancing and other measures to reduce transmission of COVID-19 during the initial phases of the pandemic²³. These measures also reduced circulation of other respiratory infections²³.

Figure 6. Prescribing rate of the systemic antibacterials therapeutic group.

Figure 7. Prescribing rate of the drugs for obstructive airway diseases therapeutic group.

Similar to data from OpenPrescribing in England, the current application shows a spike in the prescription rate of drugs for chronic obstructive airway diseases (R03) in March-April 2020, coinciding with the onset of the COVID-19 pandemic in Ireland²⁴. This may be attributed to increased patient stockpiling, and improved adherence to inhalers during the initial months of the pandemic due to public health messaging that people with respiratory conditions may be at high risk of complications of COVID-19²⁴.