Data Resource Profile Update

Routinely collected data in the UK

The National Health Service (NHS) has provided freely available healthcare for the UK population since it was launched in 1948 [1]. The digital collection of its primary care data started in the late 1980s with the purpose of managing general practitioner (GP) practices and patients, and over the years has generated extremely large data sources that are used for research purposes [2]. As, in the UK, GPs play the role of NHS gatekeepers, they are also responsible for patient hospital referrals and receive outcome documentation from secondary care, which allows this information to be coded into primary care records, further enhancing the capture of patient health events. When linkages to secondary care data, mortality, disease-based or treatment-based registries, and socioeconomic status (SES) information are available, the benefits of routinely collected data for evidence-based research may expand even further, exceeding the value of using individual datasets separately [3].

Clinical Practice Research Datalink

Clinical Practice Research Datalink (CPRD) is a not-for-profit, cost-recovery UK government research service delivered by the Medicines and Healthcare products Regulatory Agency (MHRA) with support from the National Institute for Health and Care Research (NIHR), as part of the Department of Health and Social Care [4]. CPRD collects data from three primary care electronic patient record (EPR) systems: Vision^®, EMIS Web^®, and TPP^®. Vision^® and EMIS Web^® contribute to the observational databases CPRD GOLD, which includes practices across the UK, and CPRD Aurum, which hosts practices from England, respectively. Vision^®, EMIS Web^®, and TPP^® underlie the CPRD Interventional Research services.

The CPRD engagement team enrols GP practices by conducting outreach and recruitment activity across the UK. Practices using any of the supported software systems can join CPRD on a voluntary basis and can leave at any time, e.g. if they disenroll, close down, merge with another practice, or transfer to another software system. In the latter case, practices are encouraged to continue contributing to CPRD via the new system.

CPRD GOLD

CPRD GOLD, previously known as the General Practice Research Database and Value Added Medical Products Research Databank, is a large primary care data source that was established in 1987. Its history has been widely documented [5–8] and Vision^®, originally produced by In Practice Systems, is now part of Cegedim Healthcare Solutions. CPRD GOLD has been recognized as one of the first and largest collections of anonymized electronic health records in the UK and worldwide [9]. CPRD GOLD provides longitudinal information about patients’ demographics, symptoms, conditions, referrals, vaccinations, prescriptions, measurements, and laboratory test results collected by participating GP practices, as part of their NHS care. While, in the raw data, clinical and prescribing information is coded by using Vision^® bespoke coding systems (i.e. medcode and prodcode, respectively), CPRD provides dictionaries for their translation into standard coding systems (e.g. READ 3 [10] and dm+d [11], respectively). In 2015, the first data resource profile of CPRD GOLD was published [12] and its reference was added to the official CPRD GOLD documents on the CPRD website.

Why a new CPRD GOLD profile?

In the last decade, there have been significant changes in CPRD GOLD due to the evolving market of primary care EPR systems in the UK. The Vision^® software system, which underlies CPRD GOLD, went from controlling 10%–20% of the UK EPR market share in 2018 and 2019 to holding only 5%–10% between 2020 and 2022. Moreover, while, in 2022, Vision^® held only 0–5% of the market share in England, it controlled 50%–60% in Wales, 40%–50% in Scotland, and 30%–40% in Northern Ireland [13]. This has resulted in many new GP practices joining CPRD GOLD from Scotland, Wales, and Northern Ireland, and fewer contributing from England. On the other hand, the use of EMIS Web^® software has grown in England, with many practices now contributing to the CPRD Aurum database [14] instead of CPRD GOLD. These variations affect the CPRD GOLD population size, regional coverage, and eligibility for data linkages, as currently CPRD can only provide linked data at the patient level for England. This paper addresses these changes and provides information about the current characteristics of CPRD GOLD. This renewed profile will help researchers to better understand the present dataset, and therefore to tailor research questions and study designs to a greater degree, avoiding possible analyses pitfalls, results misinterpretations, or inaccurate conclusions.

Main change in CPRD GOLD: regional distribution

The CPRD GOLD January 2024 release [15] (hereinafter CPRD GOLD Jan-2024) has 364 currently contributing GP practices (7 in England, 110 in Wales, 207 in Scotland, and 40 in Northern Ireland). Over the last decade, the practice distribution in CPRD GOLD has changed considerably for each UK constituent country: England, Wales, Scotland, and Northern Ireland (Figure 1). Regional distribution data of currently contributing practices have been provided by CPRD since April 2018 (Supplementary data).

Open multimedia modal

The number of currently contributing GP practices in CPRD GOLD from England has decreased sharply over time, with a total proportional drop of 94.0% (from 118 GP practices in April 2018 to 7 in January 2024). Conversely, the numbers of GP practices from Wales, Scotland, and Northern Ireland have risen, with Scotland having the highest numbers of currently contributing practices. From April 2018 to January 2024, the UK constituent country distribution expressed as a percentage of all practices contributing to CPRD GOLD has changed from 43.4% to 1.9% in England, from 7.4% to 11.0% in Northern Ireland, from 24.6% to 30.2% in Wales, and from 24.6% to 56.9% in Scotland.

Who is in CPRD GOLD?

CPRD GOLD Jan-2024 contains 984 historical and current GP practices from the four UK constituent countries [median of 19.1 (13.4–21.9) years of GP contribution], with coverage of >21.3 million patients. Of these, nearly 3 million are alive and currently registered in a contributing GP practice. Patients with indeterminate sex (N = 1137) or with year of birth prior to 1875 (N = 3) are excluded from the analyses. Key demographic information on current and total patients is presented in Table 1.

Open multimedia modal

These patients have data quality considered acceptable for clinical research and are registered in practices flagged as up-to-standard (UTS) for use in research because of their continuous high-quality data. The detailed definition of UTS practices is provided in the Supplementary data.

Patients in CPRD GOLD are labelled as ‘acceptable’ for use in research by a process that identifies and excludes patients with non-continuous follow-up or patients with poor data recording that raises suspicion as to the validity of that patient’s record. Patient data are checked, for the following issues:

• an empty or invalid first registration date;

• an empty or invalid current registration date;

• absence of a record for a year of birth;

• first registration date prior to their birth year;

• current registration date prior to their birth year;

• transferred-out reason with no transferred-out date;

• transferred-out date with no transferred-out reason;

• transferred-out date prior to their first registration date;

• transferred-out date prior to their current registration date;

• current registration date prior to their first registration date;

• gender other than female/male/indeterminate;

• age of >115 years at end of follow-up;

• recorded healthcare episodes in years prior to birth year;

• all recorded healthcare episodes have empty or invalid event dates;

• registration status of temporary patients.

If any of these conditions are true, then the patient is labelled as unacceptable and is not recommended for use in research.

CPRD GOLD Jan-2024 historical and current patients have a median follow-up of 5.9 years [interquartile range (IQR): 2.0–13.5] excluding any time before the registering practice was last flagged as UTS and 7.2 years (IQR: 2.3–18.5) without considering the UTS date. While checking the data-quality standards of GP practices at each data release supports robust clinical research, in CPRD GOLD, newly calculated UTS dates overwrite those previously recorded. This makes the identification of multiple UTS and not-UTS observation periods during a practice lifetime impossible. When using the UTS date in calculating the start of the patient observation period, 18.5% of CPRD GOLD historical and current patients are automatically excluded, as they have no observation time, while, without considering the UTS date, only 2.0% are excluded (Supplementary Tables S1 and S2). Therefore, considering or ignoring UTS dates might have an impact on studies investigating rare diseases, disease incidence/prevalence, or the trajectories of chronic conditions.

Supplementary Tables S1 and S2 give an overview of the demographic characterization of CPRD GOLD Jan-2024 historical and current patients, overall and stratified by UK constituent country, considering and ignoring UTS dates, respectively. Not considering UTS dates, CPRD GOLD Jan-2024 comprises 51.8% female patients. Regarding ethnicity (using linked data from the CPRD GOLD Ethnicity Record [16^, 17]), 53.5% of patients have their ethnicity recorded. Among those, 87.8% of patients are classified as ‘White’, 6.3% as ‘Asian’, 3.2% as ‘Black’, with the remaining 2.7% from a ‘Mixed’ and ‘Other’ ethnic minority background. A similar distribution is found when UTS dates are considered.

Representativeness—currently registered patients

Demographic characteristics

In mid-2023, the estimated UK population was ∼68.3 million (57.7 in England, 3.2 in Wales, 5.5 in Scotland, and 1.9 in Northern Ireland) [18]. CPRD GOLD Jan-2024 has 364 actively contributing practices, with ∼3 million currently registered patients (i.e. alive, registered at actively contributing GP practices, and not transferred out of the practice). These patients represent ∼4.3% of the UK population and cover 0.1% of the estimated population in England, 32.3% in Wales, 28.6% in Scotland, and 16.2% in Northern Ireland (Figure 2 and Supplementary Table S3). CPRD GOLD Jan-2024 currently registered patients have a median follow-up of 14.4 (IQR: 5.3–27.6) years without using the UTS date and 13.3 (IQR: 4.8–20.3) years when considering it.

Open multimedia modal

Patients’ sex and 5-year age distributions are generally representative of the general population in the UK despite some differences. For example, data from CPRD GOLD Jan-2024 show a consistent greater representation of males in the middle age groups (30–59 years) and a lower proportion of both males and females in some of the youngest age groups (0–19 years) compared with Office for National Statistics (ONS) data (Figure 2).

When CPRD GOLD Jan-2024 is linked to CPRD GOLD Ethnicity Record [16^, 17], 49.1% of all currently registered patients have ethnicity records. The proportion of patients with a recorded ethnicity is lower in Wales (38.6%), Northern Ireland (38.7%), and Scotland (56.6%) compared with England (89.1%) (Supplementary Table S4).

The ethnicity distribution of CPRD GOLD Jan-2024 is compared with ethnicity distribution from the 2021 Census in England, Wales, and Northern Ireland, and the 2022 Census in Scotland [19]. With the limitation of 50.9% missing ethnicity data (10.9% in England, 61.4% in Wales, 43.4% in Scotland, 61.3% in Northern Ireland), CPRD GOLD Jan-2024 linked to Ethnicity Record has a higher proportion of ‘White’ current patients compared with the UK population (90.7% vs 83.1%). This might be due to the populations of Wales, Scotland, and Northern Ireland being less ethnically diverse than the population of England.

Practice area-level socioeconomic characteristics

GP practice area-level SES [20] is measured by using the Index of Multiple Deprivation (IMD) and the Rural and Urban Classification (RUC).

In this paper, IMD and RUC measures at the practice level are used as a proxy for registered patients’ SES due to the availability of linkages for small areas at the patient level being limited to England. The distribution of both measures is shown in Table 2.

Open multimedia modal

Due to the small number of practices from England (N = 7) and Northern Ireland (N = 40), it is recommended not to stratify by IMD or RUC individually for these countries, and therefore they have been combined in Table 2.

The representativeness of CPRD GOLD-linked IMD and RUC at the practice level in January 2024 is compared with the general population measures for each UK constituent country. Further details on how and where IMD and RUC measures of national data were obtained are available elsewhere [21] and in the Supplementary data.

The IMD in each UK constituent country population is assessed based on the distribution of national IMD quintiles, ranging from 1 (Least deprived) to 5 (Most deprived) areas. Quintiles ensure that each category includes 20% of the national population, making the distribution across the five groups even.

Overall, in the UK, there is an under-representation of GP practices in the least-deprived areas compared with the general population (IMD quintile 1: 14.3% in CPRD GOLD vs. 20% expected in the general population). Of the 364 currently contributing GP practices, 69 (19.0%) are classified as most deprived (IMD quintile 5) and 52 (14.3%) as least deprived (IMD quintile 1).

The distribution of RUC in the national population for each UK constituent country is shown in Supplementary Table S5. Table 2 shows a higher percentage of urban CPRD GOLD Jan-2024 practices in Wales compared with the total percentage of urban practices in this country (CPRD GOLD-RUC compared with national figure: 80.0% vs. 68.3%). CPRD GOLD Jan-2024 practices in Scotland are more rural compared with the total percentage of the Scottish population (25.1% vs. 17.2%).

Data linkages

Patients from practices in England participating in the CPRD GOLD linkage scheme who did not opt out can be linked to other data sources to improve the evaluation and monitoring of patients’ health and care services. No data linkages are provided by CPRD for patients from Scotland, Wales, and Northern Ireland. For England, the actual linkage is undertaken by NHS England as a trusted third party so that no patient personal information is held by CPRD or by researchers. At present, each data source is linked by using a patient-level eight-step deterministic algorithm designed by NHS England and comes with a linkage coverage period [3^, 22^, 23].

These linkable data sources for England include, but are not limited to, secondary care data (e.g. Hospital Episode Statistics Admitted Patient Care, Outpatient, and Accident & Emergency), disease-based or treatment-based registries (e.g. Cancer diagnoses and Cancer treatments, COVID-19 virology tests, and hospitalizations), ONS death registration data, and socioeconomic datasets (e.g. IMD, RUC, Townsend Deprivation Index). More information about linkages available for CPRD GOLD can be found at https://www.cprd.com/cprd-linked-data.

Historical patients from England present in CPRD GOLD can be linked to the available datasets, but the small number of currently contributing English practices (N = 7) limits the usefulness of linked data for most studies.

Algorithm-derived data

CPRD has developed several derived datasets based on primary care data and, if relevant/available, linked secondary care data. These datasets use algorithms to bring together and consolidate related data from throughout the patient care records [24]. Currently, three algorithm-derived datasets are available to support research using CPRD GOLD. The CPRD GOLD Pregnancy Register contains a list of all pregnancy episodes recorded in CPRD GOLD [25]. The CPRD GOLD Mother-Baby Link links likely mother–baby pairs within CPRD GOLD [26^, 27]. The CPRD GOLD Ethnicity Record reports a single derived ethnicity category for each patient in CPRD GOLD [17].

Acknowledgements

We would like to thank Sonia Coton (CPRD-MHRA), Stuart Fordyce (CPRD-MHRA), Susan Hodgson (CPRD-MHRA), and Puja Myles (CPRD-MHRA) for supporting this study with their valuable contribution and expertise.