We conducted a retrospective observational study, accessing population-scale individual-level community dispensing data using the Secure Anonymised Information Linkage (SAIL) Databank [12, 13]. For the purposes of COVID-19 research and intelligence, the SAIL Databank was granted permission under the Control of Patient Information (COPI) [14] notice to acquire and anonymise the Welsh Dispensing Data Set (WDDS). The WDDS records information on all national health service (NHS) primary care general practitioner (GP) prescribed medications, and associated information, for medications dispensed by community dispensing contractors (community pharmacies, dispensing general practices, general practices that personally administer prescription medication, and dispensing appliance contractors). The data were available on a monthly basis from 2016 to end of July 2021 [15].

The WDDS data are provided to the SAIL Databank by NHS Wales Shared Services Partnership (NWSSP). The data are captured from prescriptions submitted to NWSSP, on a monthly basis, by all primary care dispensing contractors to claim remuneration for dispensing in accordance with the provisions of the National Health Service (Pharmaceutical and Local Pharmaceutical Services) Regulations 2013. The WDDS data are limited to NHS prescriptions containing a 2D matrix barcode; forms with a bar code are only produced by GP clinical systems. The GP system-produced forms account for 98% of total dispensing (remainder from, hand written, hospital outpatient, dental, etc.). Nearly all (97%) of GP system-produced forms are read by scanners. Therefore, the dataset accounts for 94% of all prescriptions dispensed and subsequently submitted for reimbursement. These proportions did not change over the period of the data supplied to this study. Medications for Welsh residents that are dispensed against non-barcoded prescriptions, over the counter or dispensed outside Wales are not included in this data. The WDDS data are made available to SAIL on a monthly basis with approximately a 6-week lag to allow for official reporting and quality assurance processes [13].

Dispensing data from 1^st January 2016 to 31^st December 2019 (counterfactual pre-COVID-19, “C16 cohort”) were compared to every year data from 1^st January 2020 and followed up until 27th July 2021 (COVID-19 period, “C20 cohort”). We used two purpose-built residency spine e-cohorts capturing all people who were alive and resident in Wales [16]. As part of the process in creating the residency spine, for each period cases were removed if they died or moved out of Wales at the start of the observation window [16]. Drug dispensing history was retrieved from individual-level linked dispensing records using anonymised linkage fields (ALFs), and the date of dispensing of the medications.

Dispensed items in WDDS are originally coded in Dictionary of Medicines and Devices (DM + D), which we mapped at the code level to their corresponding British National Formulary (BNF) code(s) using a complete extract of mappings from National Health Service Business Services Authority (NHS-BSA) services published in December 2020 [17]. Where the BSA mapping was not available, NHS Terminology Reference-Data Update Distribution (known as TRUD) tables were used to map DM+D codes to seventh character BNF codes [18] (see Supplementary 1 for more details on mapping). We created a semi-automated pipeline to generate, for each monthly extract of the data, an RRDA. The most recent version of documentation is provided in Supplementary 2.

Data were aggregated by month and year for items in each BNF chapter. We compared dispensing rates between the C16 and C20 cohorts for all of the BNF chapters and provide them as part of the interactive dashboard; we also highlight in text the three major BNF chapters: I) Cardiovascular Systems; II) Central Nervous System; and III) Immunological Products & Vaccines. We calculated age-standardised dispensing rates per 100,000 population. Five-year age bands were used with upper bound of all those aging 90 years or more using the 2013 European Standard Population (ESP) [19]. Dispensing trends were assessed based on total quantity of dispensed items for each BNF chapter per year [20] and the number of patients who have had at least one item dispensed, as documented in their WDDS records.

Per person dispensing rates were calculated based on the total number of dispensed items and the total number of patients with a dispensing record on each period. Dispensing rates were calculated for each year. The difference (D) and relative changes (RC) were calculated for comparison of monthly trends in dispensing between 2020 records and 2019. All statistical procedures are performed in R (v3.5) and the code is available at: https://github.com/SwanseaUniversityMedical/WDDS.

We developed an interactive dashboard using R shiny [21] allowing up-to-date dynamic monitoring of trends in the current year, an enables the opportunity for comparison between the most up to date data with previous years’ patterns. The dashboard is accessible at https://wdds.ml/.

3,228,062 patients in the WDDS were included in the two e-cohorts [22]. Of those 3,154,657 (97.7%) had at least one dispensed item (with a total of 393,235,198 dispensed items), were resident in Wales from 2016 onward, and were able to be linked. This comprised 86.2% of Welsh residents. A mapping rate of 99.9% was achieved for mapping all DM+D recorded dispensed items to BNF codes (Figure 1 & Supplementary Figures 1a & 1b).

We observed a -0.96% relative decrease in total number of dispensed items in 2020 compared to 2019 (73,410,543 vs 74,121,180; 95% CI [-3.72, 1.80]). At the same time, the overall rate of dispensed items per person increased from 32.43 items per person in 2019 to 34.00 items per person in 2020, representing a Difference (D) of +1.57 and a Relative Change (RC) of 4.8% (Table 1.a).

Comparison of monthly patterns of dispensed items between 2020 and 2019 showed a notable difference between the total number of dispensed drug items each month, with an average monthly difference of -59,220 and an average RC of -0.74%. In the first two months of 2020, total number of dispensed items were approximately the same as expected numbers in 2019 (average D = -11,801, average RC = -0.19%).

In March 2020, following lockdown restrictions the total number of dispensed items was 20.62% higher than 2019 (D = +1,224,909, RC = +20.62%). Following the March peak, number of dispensed items in April and May 2020 were lower compared to the same months in 2019 with a 5.34% relative decrease in April (D = -338, 715, RC = -5.34%) followed by the biggest observed fall in May 2020, where the total number of dispensed items was 11.34% lower than in May 2019 (D = -726, 065, RC = -11.34).

The second peak in the total number of monthly dispensed items was observed in June 2020 (D = +257,920, RC = +4.50%). Although the number of dispensed items increased in June, similar to the March peak this was also followed by a further reduction in total number of dispensed items in subsequent months with an average of -8.22% lower dispensing in July and August 2020 compared to the same periods in 2019 (average D = -523,554, average RC = -8.22).

A third peak representing a 4.35 % greater number of dispensed items was observed in September 2020 compared to September 2019 (D = +264,138, RC = +4.35) which similarly was followed by a fall in total number of dispensed items in the October 2020 compared to October 2019 (D = -412,452, RC = -6.35%).

Similar to trends at the start of 2020, we observed minimal difference in the total number of dispensed items in November 2020 compared to November 2019 (D = -44,710, RC = -0.74), this was followed by a 2.13% relative increase in December 2020 compared to 2019 (D = +135,048, RC = +2.13%). This December increase followed by a -9.51% relative decrease in January 2021 (D = -607,836, RC = -9.51%) and the trend was upward until a peak in March 2021 (D = +842,447, RC = +14.18%) which was similar but smaller than the 2020 peak. This was followed by a (again, similar) decreased total number of dispensed items in April and May 2021 and then an even greater peak in June 2021 than observed in 2020 (D = +717,496, RC = +12.52%) (Table 1.b & Figure 2).

The observed increase in March 2020 was consistent for items in the majority of BNF chapters, except for items in the Anaesthesia and the Immunological products & Vaccines products. The total number of dispensed Anaesthetic products in March 2020 was consistent with March 2019 and the Immunological products & Vaccines were significantly lower in March 2020 compared to March 2019 and stayed on the same low level during 2021. This differed from the overall patterns (Figure 3). For dynamic comparisons of all years, please see Supplementary Figures 2.1 to 2.3 or visit our online dashboard at http://wdds.ml/).

Using European age-standardised dispensing rates, we focused on the comparison between 2020 and 2021 versus 2019 for dispensing in Cardiovascular Systems (CVS), Central Nervous Systems (CNS) and Immunological Products and Vaccines (IPV) chapters. All three groups showed similar trends between the years for January and February.

For CVS, dispensing increased considerably during March 2020, followed by a reduction in April and May, recovery to 2019 levels in June, a drop in July and August, and recovery again in September 2020 (Figure 4 – Supplementary Figure 3.1).

For CNS dispensing, the March 2020 peak was observed, followed by a reduction in May followed by a notable reduction at the end of the 2020 in November and December compared to 2019 (Figure 5 – Supplementary Figure 3.2).

For IPV, although the 2020 dispensing patterns for January and February mirrored 2019 levels, for all other months (and across all age groups), there was a large decline in 2020 that continued in 2021 (Figure 6 – Supplementary Figure 3.3).

Community dispensed medication provides a direct insight into healthcare delivery and service utilisation. Monitoring of trends in dispensed drug items under classified drug categories can also provide a proxy into communities’ health and existing comorbidities. This assimilation of national dispensing data in Wales resulted in the construction, curation and ongoing maintenance of a linkable RRDA. This can be used for near real-time monitoring of dispensing trends in Wales as well as facilitating the current and future research and intelligence outputs around COVID-19 within the SAIL Databank, and longer-term, pending governance approvals for wider non-COVID-19 use. The RRDA is accompanied with sufficient metadata allowing researchers, analysts and other users comprehensive access to research resources when using the dispensing data. We have specifically initiated the development of a harmonised approach to establish a federated analysis process across multiple trusted research environments. Our dashboard was developed and updated each month for this comparative analysis using aggregated level data, to enable timely monitoring of dispensing trends over time as the care system adapts to the evolving challenges of the pandemic.

Our data show a small relative decrease of 1% in total number of dispensed drug items in 2020 compared to 2019 despite the huge impact of COVID-19 on attendances. We show in detail, across all BNF chapters, changes in patterns of drugs dispensed during the COVID-19 period. In general, the peak in the total number of dispensed items coincided with the first UK and Wales national lockdown starting from 23 March 2020 (See supplementary Figure 4 for timeline of lockdowns in Wales-UK), during the first peak of the pandemic in March 2020. We observed a dramatic reduction in the dispensing of Immunological Products & Vaccines at the start of lockdown in the COVID-19 period. While routinely dispensed vaccines in pharmacies are mainly related to travel - such as yellow fever and rabies, the demand for these naturally been reduced during the COVID-19 period. Similarly, obtaining vaccines from pharmacies through practice nurses prior to house visits may have been less frequent during COVID-19 lockdown periods (See supplementary Figure 4 for timeline of lockdowns in Wales-UK). The current COVID-19 pandemic has interrupted routine immunisation pathways and led to dramatic shifts in the dispensing of Immunological Products & Vaccines [19, 20].

While our observation matches what has been reported in other studies such as Kaye et al. [7]; the increase of dispensed items in March cannot be considered to reflect better adherence, but more likely reflects a planned change in service delivery procedures in the context of the COVID-19 contact mitigation measures taken to reduce transmission across communities.

These peaks could also be explained by stockpiling prescriptions or filling prescriptions claimed. Although there is very limited other UK evidence, this pheonomenon has been evident elsewhere [23]. Our data showed that this peak occurred in most drug categories and was followed by a dip in the following two months (lockdown period), suggestive of batch-dispensing in preparation for the national lockdowns in March 2020 (See supplementary Figure 4 for timeline of lockdowns in Wales-UK). It is also possible that patients had spare prescriptions to hand and felt a sense of urgency to get them dispensed.

Interestingly, we observed similar monthly trends during 2021 (up to the last data extract available to the study). Given that there was no additional lockdowns in Wales in 2021 (supplementary Figure 4), observing similar trends in dispensing peaks in March and June of 2020 and 2021indicates a persistent change in community dispensing practices beyond the first year of the pandemic.

The population-scale individual-level data used in our study are available as a monthly extract to SAIL Databank; this provides the opportunity to use aggregated level data for monitoring trends in near real-time, as well as exploring the effect of individual-level factors such as age, sex, socioeconomic status and ethnicity. We acknowledge the known difference between prescribing and dispensing and accounting for dosage and duration of dispensed medication. We nonetheless believe that providing an easy to analyse platform for monitoring monthly trends of dispensing data provides a unique opportunity not only to monitor and evaluate prescribing and dispensing trends in near real-time, but also to explore the effect of changing treatment patterns on outcomes considering a comprehensive selection of patient-level factors. Our study developed a linked dispensing dataset which mapped DM+D codes to the BNF codes and chapters. This provides many further opportunities to investigate trends within each drug category, as well as wider collaboration across harmonised data sources. While our study demonstrates changes in dispending behaviour, our ability to identify whether these observed changes were due to prescribers’ dispensing behaviour or other public health interventions such as programs aiming to increase adherence to medications or appropriate prescribing is limited.