How an electronic health record became a real-world research resource: comparison between London’s Whole Systems Integrated Care database and the Clinical Practice Research Datalink

The healthcare sector and regulatory bodies increasingly need to understand more about the real-world implications of diseases and healthcare interventions, requiring access to good-quality fully integrated healthcare datasets. England’s National Health Service (NHS) is well placed to deliver this for several reasons. With a single healthcare system, it is possible to follow patients from birth to death. With a low proportion of healthcare services being provided outside of the NHS (£9bn compared with £126bn in the NHS in 2017), [1] it is possible to obtain a near-complete view of both existing and new services and treatments that patients access. The computerisation of UK general practice records and the fact that 98% of the population is registered with a GP leads to almost whole-population coverage. Unlike with clinical trials and biobanks, the denominator resulting from such databases is relatively free from selection bias and represents the entire population.

While primary care sees most contacts with patients, linkage with secondary care and other sectors is needed for a full picture of the patient’s journey through the health and social care system and their outcomes. In England, as in the other UK countries, hospital inpatient data are combined to give a single database, but primary care uses several different systems, so far preventing the creation of an equivalent database for primary care. Instead, large samples of vendor-specific primary care data for research are available from several sources. The Clinical Practice Research Datalink (CPRD), for example, now includes records of over 11 million currently registered patients (16% of UK population), with linkage to hospital records, the national cancer registry, area-level social deprivation information and national mortality data, though some of these sources are for England only; the Health Improvement Network (THIN) [2] and QResearch [3] databases are similar but smaller. CPRD has generated over 1000 research papers [4]. Various initiatives have created local data warehouses such as the KID in Kent [5]. It uses pseudonymisation-at-source to link patient-level records from services including general practices, hospitals, community health services, hospices, and adult social care for its nearly two million population in SE England. It was established to track service use by patients with any of a set of long-term conditions but has since expanded to cover all patients. It is overseen by a steering group, one of whose subgroups considers requests for access to the data. These are not epidemiological cohorts or resources like UK Biobank [6] but were developed primarily for direct patient care and commissioning. Technical issues such as interoperability of data systems and ethics have complicated their construction.

Over the past 5 years, the team behind one such local data warehouse in North West London has overcome such issues to make the dataset available for research. We cover the origin, funding, contents and structure of this data warehouse, derived from the Whole System Integrated Care (WSIC) programme, its anonymised research version Discover, and its consent-to-contact feature. We then compare it with CPRD and discuss access to Discover and its current and future uses and developments for research.

Using descriptive statistics, we compared the Discover patient mix with that of London as a whole and the UK. Mid-year population estimates for London were taken from the London Trust [10] and for the UK were taken from the Office for National Statistics [11]. As Discover lacks the date of patient registration with the GP, its populations are currently only known on the day of data extraction, not historically. To obtain denominators for the proportion of patients with key risk factors recorded, the Discover population (denominator) over time was estimated based on a current comparison with London (see Appendix). Year-specific recording rates were calculated using the disease status as at Dec 31.

We estimated the prevalence of long-term conditions covered by the Quality and Outcomes Framework (QOF) programme for general practice; UK prevalence figures were taken from QOF for 2017/18 [12].

WSIC/Discover is one of a new breed of local but large linked databases, derived from health and social care records, used for service evaluation and commissioning and increasingly also for research. While smaller than CPRD in terms of the numbers of practices and patients, it offers advantages in the inclusion of social care and mental health data and identifiable general practices and providers, allowing easy incorporation of institution-level data for service evaluation and research.

To enable researchers to use data collected for non-research purposes, ethical issues around consent for secondary use of patient data, robust de-identification and information governance procedures have been established. While the relevant legislation will differ by country, even with the advent of Europe’s General Data Protection Regulation, these issues are relevant internationally. The development of a consent-to-contact register alongside the Discover dataset promises to make this powerful linked dataset a tool to run real-world studies retrospectively or prospectively.

Although the data have been de-identified, a secure platform and access controls are still needed due to the potential for re-identification, which is possible by linking the unencrypted parts of the record with known information about the individual. Public engagement has been crucial for the project to explain the risks and benefits, something that was done badly in England’s care.data initiative, designed to extract data from primary care medical records for commissioning and other purposes, including research [13].

There have been few published descriptions of local data warehouses like WSIC. One is the Kent Integrated Dataset [5]. As with CPRD and Discover, the primary care data in KID are the richest, but with all such data there are recording differences between general practices and over time. Among the many data elements that are captured, data quality for some remains variable. Symptom severity for COPD is captured quite well in primary care EHRs [14] but for other conditions this is not the case, and hospital records that use ICD10 for diagnoses will be of limited help. The 2004 introduction of the Quality and Outcomes Framework, in which some of the payments that GPs receive depends on their management of chronic conditions, helped drive improvements in recording of key risk factors and intermediate outcomes such as BP, smoking and HbA1c [15]. Similar future initiatives are likely to have a similar impact.

Like CPRD and other similar databases in the UK, WSIC has been several years in the making and is still growing with further linkages (Table 4).

As well as the options considered in Table 4, future directions for databases like WSIC include the incorporation of data on patient-reported outcomes and telemonitoring data, for example entered by patients themselves via apps (several companies are doing this but generally only for patients with a given condition at enrolled practices) or collected from them via wearable sensors and analysed by machine learning. Other uses include long-term surveillance of medications, process mining and service redesign scenario modelling. There is government investment too: up to five Digital Innovation Hubs will be led by Health Data Research UK (HDR UK), the national institute for data science in health. Discover will be one of these, called “Discover-NOW”.

WSIC/Discover has been several years in the making, with both similarities to and differences from CPRD. With the groundwork done, it is ready to expand yet further and, providing that users understand how the data were generated and the limitations, it is a valuable research tool and a model for others developing similar data warehouses.