Background
The Dutch health system is based upon a gatekeeper system, in which patients only have access to hospital care after consulting the general practitioner (GP). All Dutch residents are registered with a GP, and all have mandatory health care insurance. GPs provide basic care for all health problems for all patient categories; emergency care, chronic disease, and mental health problems. GP care is completely insured and is not subject to initial payment; therefore the threshold for consultation is low. GP registries therefore adequately reflect morbidity patterns of the Dutch population, as far as it results in a contact with a healthcare provider.
General Practice has a long-standing history of registering routine healthcare data. In the early days of general practice, some GPs in the UK presented joint patients records as practice overviews to study morbidity patterns in primary care [
1,
2]. In the Netherlands Huygen kept a very detailed registration of patient contacts and demonstrated that such registration of routine care data could be used to answer clinically relevant research questions [
3]. Over the last three decades electronic patient records have increasingly replaced traditional paper files. With the introduction of electronic General Practice Information Systems (GPIS), patient contacts could be studied more easily [
4]. Initially, registration was primarily done for practice archiving, and data quality was variable. Later, professional organizations such as the Royal College of GPs in the UK, and the Dutch College of GPs (NHG) stimulated consistency and reproducibility of coding by the introduction of the structured diagnostic coding systems: READ codes in the UK, and International Classification of Disease Codes (ICPC) in the Netherlands [
5,
6]. Large-scale use of these coding systems improved registration uniformity and made routine care data accessible and attractive for research [
7]. The READ diagnostic coding system provided more detailed diagnostic information than the ICPC-coding system, enabling more diagnostic details for research. In more recent years, routine care databases also proved suitable for the monitoring of quality of care in general practice. In the UK, GPs within the National Health Service (NHS) monitor relevant quality parameters with standardized templates as part of the routine clinical process. In the Netherlands, GP network organisations initiated coding and monitoring of chronic disease management, which further improved the quality of the records in terms of completeness and accuracy [
8].
Routine care databases are widely used for research. Primary care research networks originated from local initiatives, but in recent years they merged to large-scale registration databases in many countries [
4,
9‐
11]. In the UK, the General Practice Research Database (GPRD) has been shown to be an effective environment for large scale observational research in primary care. Catalonia (SIDIAP), Scotland (ESCRO), and Canada (CPSSN) also provide comparable databases enabling research.
Academic research in routine primary care databases in the Netherlands has a longstanding tradition. In the 1960s a number of dedicated general practitioners in the Nijmegen area started the Continuous Morbidity Registration (CMR), which at present has more than 40 years of follow-up registration data of 15,000 patients [
12,
13]. In 1970 in the Amsterdam region, Lamberts introduced the ICPC coding system in the Netherlands, initiated in ‘the Transition project’; reasons for encounter and (working) diagnoses were uniformly coded [
14‐
16]. In 1998, The Netherlands Institute for Primary Care (Nivel) conducted the first national survey of general practices, resulting in a countrywide registry of presented morbidity, and diagnostic and therapeutic interventions in primary care. In follow-up, the NIVEL set up a permanent representative nationwide network of sentinel practices (LINH), which produces regular monitoring reports on epidemiology, management and organization of general practice in the Netherlands [
17].
In the Utrecht region, six general practice groups collaborating with the University Medical Centre started in 1996 a registration network, with the mission to make their routine care data accessible for research [
18]. With developing registration systems and training in systematic data recording, the data quality improved over the years, and this academic primary care network became a high-quality cohort for research. In 2016 the network had expanded to 64 practices, and presently consists of routine primary care data of 370,000 enlisted individuals (see Table
1), with 1.38 million consultations annually, and 15–20 years follow-up.
Table 1
General characteristics of the Julius General Practitioners Network (JGPN, 2015)
Number of general (group) practices | Total number of patients enlisted | Average number of contacts per day per practice | Total number of patient contacts per year | Mean number of patient contacts per year (range) | |
64 | 371,028 | 79.9 | 1384.129 | 3.7 (0–193) | |
Age groups in years | Number of men (%) | Number of women (%) | | | |
0–19 | 41,405 (11) | 39,700 (11) | | | |
20–39 | 50,722 (14) | 60,717 (16) | | | |
40–59 | 51,256 (14) | 51,238 (14) | | | |
> 60 | 33,965 (10) | 40,845 (11) | | | |
Total | 177,348 (48) | 192,500 (52) | | | |
Top five prescribed drugs according to the Anatomic-Therapeutic Codes (ATC) for medication | Number of prescriptions (% of total) | Number of patients (% of total) | Mean number of prescriptions per patient | Number of patients with just one prescription (%) | Percentage of patients in the Netherlands with the prescription [ 48] |
Antacids (PPI’s, H2-antagonists, and antacids) (A02B) | 164,293 (7) | 39,613 (11) | 4,1 | 12,619 (32) | 16 |
Statins and other cholesterol lowering drugs (C10A) | 134,352 (5) | 26,24 (7) | 5,1 | 3183 (12) | 11 |
Antithrombotics (Vitamin K antagonists, NOACs, and anti-platelets) (B01A) | 123,209 (5) | 22,195 (6) | 5,6 | 3207 (15) | 10 |
Beta-blockers (C07A) | 105,737 (4) | 21,413 (6) | 4,9 | 3689 (17) | 9 |
Antidepressants (N06A) | 100,883 (4) | 18,046 (5) | 5,6 | 3048 (17) | Below top 10 |
Top five diagnoses according to the International Code Primary Care (ICPC) for diagnoses | Number of contacts with such an ICPC-labelled diagnosis (number of new/relapsed episodes) | Percentage of total number of contacts with an ICPC-code (%) | Number of patients (% of total) | Mean number of contacts per patient (minimum-maximum) | Percentage of patients in the Netherlands with the ICPC-code [ 49] |
Hypertension (K86, K87) | 52,508 (2708) | 3.9 | 18,445 (5) | 2.8 (1–29) | 4 |
Type 1 and 2 Diabetes Mellitus (T90) | 51,194 (11,547) | 3.8 | 11,283 (3) | 4.5 (1–79) | 2 |
Cystitis/urinary tract infection (U71) | 28,834 (19,495) | 2.1 | 12,473 (3) | 2.3 (1–35) | 3 |
Cough (R05) | 25,487 (22,794) | 1.9 | 16,966 (5) | 1.5 (1–15) | 2 |
Upper respiratory tract infection (R74) | 23,579 (22,330) | 1.7 | 17,580 (5) | 1.3 (1–13) | 2 |
In this paper we describe the organisation and data content of the Julius General Practitioners’ Network (JGPN), the different types of research that have been performed with the data and the potential of the networks for future innovation and how the data can be supportive in education in general practice.
Discussion
Routine-care primary care databases such as the JGPN offer excellent potential for different types of clinical research. In addition, the database can be used for feedback and monitoring purposes in support of quality of care programs in daily care. In future the dataset will also be used for educational purposes, to monitor the clinical performance of GP trainees and medical students in primary care practice. Important assets of the JGPN are the size of the data set, the length of the follow-up, the representativeness of its population, and the variety of clinical information that is registered. Much effort has been put in uniformity of the registration of the routine care, resulting in a high accuracy of the data over the years.
Comparison with other data networks
Many primary care routine datasets exist; most are used only for research purposes. In the Netherlands, for example, all academic universities have such a (research) network. In the UK the General Practice Research Database is focused on facilitating larger scale observational research in primary care. Other networks have surveillance or monitoring purpose, such as the LINH sentinel network of NIVEL in the Netherlands that provides representative (monitoring) data on morbidity and clinical management in primary care. The content and the quality of the data vary between different databases. Some networks only store the coded information, and do not extract the qualitative text data, laboratory results, or the referral letters. The quality mainly depends upon the uniformity in coding, the training that participating GPs received, and the frequency of extractions. Some networks provide extensive training sessions, while others largely depend on routine data entry in the participating practices. The JGPN offers several advantages over other networks: it stores information on all aspects of clinical care; qualitative textual information on reason for encounter, diagnostic data, ICPC coded diagnoses and ATC coded prescription data, and referrals and return letters from specialists. Participating GPs are used to working with a structured coding program. The JGPN is based on a long-standing collaboration between regional participants and the academic department, and the practices actively participate in the management and exploitation of the database. Finally, the JGPN dataset is used for various objectives: not only for academic research, but also for regional surveillance, managed care programs and quality benchmarking purposes as well for educational purposes. This supports the feeling of joint ownership and creates a ‘win-win’ with participating practices.
Limitations of the database
The JGPN database contains observational data, of the consultations that were registered by the GP, depending on the interpretation of the consultations by the GP and the way it was registered by the GP. This is also the limitation of the dataset. Because of the anonymous character, it is not possible to go back to the patient in order to collect additional information. Additional information becomes available only if the patient reappears on consultation, or if data are linked to other sources. Missing data for specific research questions are not retrievable.
A second limitation of the data is that follow-up may be interrupted by both moving or by death. With the removal of patients from the database, the distinction between those who moved or those who passed away is no longer traceable. Through proxy indicators on morbidity or through linkage to the official death registry (CBS) this limitation can often be overcome.
A third limitation is that the detailed data of the electronic chronic disease management programs, that many practice have introduced to monitor their patients with type 2 diabetes, chronic obstructive pulmonary disease, and cardiovascular risk management, were initially not automatically transferred to the routine care registration. Thus, these data were not available in JGPN. In recent years however, key indicators of chronic disease management programs are automatically copied to the routine care registration.
Finally, data extraction takes place only four times a year, so the data are not always up to date. Given the progress in data capturing techniques, however, a system of real-time monitoring should be realistic in the future.
Suitability for research; pros and cons
The longitudinal character of the database provides the opportunity to perform observational analyses in different designs. Etiological studies can be achieved by splitting up the cohort and compare the positive and negative study-arm adjusting for the potential confounders. In addition, the database is large enough to analyse matched groups who are matched on important characteristics. The presence of additional determinants such as disease history, signs and test results make the data also attractive for prognostic studies. Diagnostic studies are more difficult to perform as test results are incompletely entered in the database. This can be overcome by linking individual data from laboratory databases. Interventions can be evaluated, but only in observational comparisons, with quasi experimental designs. This is particularly effective in the evaluation of healthcare innovations or guidelines after introduction on population level.
One could argue that JGPN contains insufficient performance indicators for monitoring quality of care, since GP practices vary in the level of detail of information that is entered in the routine care registration. As a result, indicators of quality of care cannot always adequately be generated. These limitations, however, can be overcome in two ways. First, many clinical outcomes can be estimated by the use of proxy indicators, such as referrals (for treatment), or medication (for diagnosis). These proxy indicators have intrinsic limitations because of, for example, inter-physician variation in referrals or in individual health policies. Alternatively, clinical information can be enriched by linkage to other databases, such as those from hospitals, insurance companies or from disease-specific registries. Even when a deterministic or probabilistic linkage would result in a number of unidentified individuals, the number of patients in the JGPN is large enough to have adequate statistical power.
Future challenges
The main future challenges for JGPN are the development of a regional data warehouse, the need to adequately address the medical ethical legislation for large-scale routine care data collection and the need to safeguard the commitment of partners such as GPs and patients. JGPN was initially setup as a mono disciplinary registration network. However, for optimal use of the data in future, structural linkage to other sources is essential to meet the ever- increasing demands of researchers, to continuously upgrade the quality of the data and to broaden the scope of the database. Recently, linkage to laboratory data, disease registries, and hospital data proved successful, but time consuming. In future, real-time linkage through a virtual regional data warehouse, connecting all relevant data sources in the region, should overcome these limitations.
Legislation around medical databases is becoming more strict, to protect the privacy of patients and medical professionals. Many suggest that individual Informed consent is required for optimal protection, but experience has learnt that only a minority of patients do respond to requests for research participation. This would threaten both the size of the database as well as its representativeness. The present opt out procedure, with optimal information provision in the practices, a register of patients that refuse to participate, and shared governance over the data, adequately safeguards ethical use of the data. Ultimately, it is the patient, and not the professional or the researcher, that decides on the use of the individual health data, even though they are anonymised. Therefore not only researchers and GPs, but also patients should be actively involved in the JGPN steering committee. Patient participation is therefore one of the challenges for the near future.
Conclusions
Routine-care primary care databases such as the JGPN offer excellent potential for different types of clinical research. Moreover, such databases can be used to support quality management in participating practices, thus optimizing individual patient care. This secures the balance between academic interest, and value for the participating GPs and patients, stimulating the concept of joint ownership, and turning a database into a solid instrument in regional health care developments.