Background
Large computerised patient databases provide a useful source of real life observational data, and the General Practice Research Database (GPRD) has been successfully used to generate descriptive epidemiology data in chronic conditions such as Chronic Obstructive Pulmonary Disease (COPD) [
1-
3] and asthma [
1,
4,
5] from a large group of UK primary care practices. Historically the limitations of the GPRD for clinical research were a time gap between GP data capture and availability for the researcher and limited links to other healthcare databases, although these are currently being addressed with the development of the Clinical Practice Research Datalink (CPRD) and in ongoing pilot work for Phase 4 pragmatic clinical trials [
6,
7]. The use of electronic medical record (EMR) data in health research is a key objective in the Department of Health’s national research strategy [
8]. EMR is increasingly adopted to support both efficiency and quality of patient care and to facilitate clinical research. Several studies have described the design and implementation of EMR, electronic data capture (EDC), data extraction and EMR retrieval systems to enable accurate and efficient data entry for clinical research to be performed on-site in real time [
9-
11].
Asthma and COPD are both treatable diseases that can be well managed [
12,
13], but despite this, a large proportion of asthmatics are poorly controlled [
12,
14,
15], and COPD remains under-diagnosed and under-treated [
2,
16]. In everyday clinical practice, variations in asthma control and prescribing patterns across countries have been reported, as well as differences in disease perceptions amongst physicians and patients [
17-
19]. Similarly for COPD, variations in treatments, standards of care and adherence to guidelines have been reported across different geographical regions [
20-
23]. In asthma and COPD, the application of EMR retrieval systems would enable the monitoring of large patient populations to support evaluation of comparative effectiveness, safety, and health care resource utilisation (HRU) of treatments in a real life setting.
This paper describes the methodology of development of the NorthWest EHealth linked database (NWEH-LDB) and alert systems. The main outcome of the study was an assessment of the reliability of NWEH-LDB as a platform that can be used to support the delivery of pragmatic clinical trials. A retrospective analysis of asthma and COPD cohorts, identified within the database, was conducted as part of a feasibility assessment to evaluate whether this system could provide a feasible and valid platform for conducting pragmatic clinical trials.
Discussion
Results from these analyses suggest that the NWEH-LDB is able to provide the necessary elements of a platform for real-world benefit/risk assessment of clinical studies. NWEH-LDB enabled the identification of asthma and COPD cohorts and data extraction on prescriptions and healthcare events for all patients. We can speculate that these data include sufficient detail to produce alerts for patient safety and HRU monitoring during clinical trials. The dual verification exercise confirmed high concordance between the two independent data sources, SIR and Apollo. Only 37 data records from 3504 registered patients were discordant between the systems. Most of the discrepancies appear to be due to patients who have either transferred to another practice or died. Good concordance was also demonstrated between SIR and Apollo for prescription and events records. By collating data from multiple sources, NWEH-LDB is sufficiently robust to compensate for the small percentage of diagnosis records (0.9%) missing from the SIR database (Figure
1).
Previous studies demonstrate various methods to deploy EMR in clinical research. For instance, Goodman and colleagues explored solutions to integrate the EMR and EDC systems to enable accurate and efficient collection of cancer clinical trial data [
9]. Murphy et al. [
11] described the implementation of an EMR and data extraction techniques to facilitate real time and on-site data entry for clinical research. They used special ‘study management’ screens to capture additional data needed for clinical trials such as adverse events, enrolment, termination and missed visits. Similarly, Yamamoto et al. developed an EMR retrieval system to identify patients who met clinical research eligibility criteria [
10]. Clinical trial alert systems are also implemented in order to increase physician participation in clinical trials [
32] and improve recruitment [
33,
34]. In line with these studies, we introduce a template to collate electronic patient data from readily available systems in hospital and general practice with the capability to extract information as required for research. Similar to Murphy et al. [
11], the NWEH-LDB systems use real time and on-site data. Our systems provide a generic platform that can extract data previously defined according to the researchers’ interest. Here we describe an example of retrospective data extraction and analysis in asthma and COPD to demonstrate the reliability of this new template in health utilisation and outcome research.
Our systems are not intended to support recruitment. However, the retrospective cohort study identified patients with asthma and/or COPD, based on primary or secondary care diagnoses, together with their prescribing data, and provided longitudinal data on HRU and clinical endpoints of interest. These data were utilized in assessing feasibility of a future trial when compared to specific protocol elements, including endpoint selection and sample size calculation [
35]. Data extraction process via NWEH-LDB can run on a daily basis with a potential to create alerts each time patients have contact with healthcare providers, prescriptions or laboratory results. These functions could be applied in pragmatic clinical trials to support patient safety monitoring remotely and obtain real life evidence in the future.
Comparisons of our data for the asthma and COPD cohorts with other population based studies [
2,
4,
36] put our findings in context and provide additional evidence that the NWEH-LDB provides accurate and robust data to characterize disease severity, treatment, and outcomes. With respect to the findings on history of exacerbations, the results for the asthma cohort concur with the recent review by Dougherty and Fahy, which stated that a history of one or more exacerbation is an important risk factor for recurrent exacerbations suggesting an “exacerbation-prone” subset of asthmatics [
37]. There is also increasing evidence that different phenotypes exist in COPD [
38,
39] and the evidence supports the existence of a COPD frequent-exacerbator phenotype [
40,
41]. A recent retrospective analysis of the ECLIPSE cohort data showed that, although the frequency and severity of exacerbations were more severe as COPD progressed, the single best predictor of exacerbations was having a history of exacerbations [
40]. The NWEH-LDB data also provide evidence of a higher rate of COPD exacerbations in 2009 among those patients who had two or more moderate/severe exacerbations recorded in 2008.
Some of the limitations of electronic medical record data reflect the environment and purpose of its collection, namely they are not recorded systematically under strictly controlled conditions, as in traditional bespoke clinical research, and rely on the GP and other contributors to accurately complete the records for all patients equally. In addition, not all GP’s routinely participate in research and those participating in a future EMR-based clinical study may be more diligent about recording information well compared with GPs not participating in research. Another limitation of this dual verification process pilot study is that the data were generated from one GP practice, which was not randomly selected, and therefore may not be representative of the whole group of participating practices included in the retrospective cohort study.
Data from this study have several noteworthy implications. Firstly, retrospective EMR data give a real-life picture of the management of COPD and asthma patients in an unperturbed clinical setting, without the constraints of strict randomised controlled trial criteria. Secondly, multi-source databases like NWEH-LDB can be used to monitor changes in a cohort of interest (in this case Asthma and COPD) over a predefined time period if data are extracted from sources into NWEH-LDB at regular intervals. For example, changes in the trends of disease prevalence and comorbidities would require data extraction from sources over long time periods (i.e. years); however, the monitoring of prescribing trends, HRU, or admissions would require narrow extraction intervals depending upon the research question. Therefore, NWEH-LDB model has the potential to provide data necessary to inform the study design, planning and power calculations of real-life safety and effectiveness studies. Since the feasibility study was conducted, NWEH-LDB has been further developed with additional resources and links incorporated (Figure
1), increasing the capability for conducting successful future EMR-enabled trials.
While linked EMR platforms continue to improve in the quantity and quality of data linkages, there may be a need for flexibility in the data collection process to augment EMR for pre-defined efficacy, effectiveness, or safety events of interest to meet requirements for certain RCTs. For example, exacerbations of COPD may require an electronic case report form (eCRF) or link to a patient portal for recording a validated Patient Reported Outcome (PRO) to be added to the primary care software to capture the indication for OCS/antibiotic prescription or measure asthma control, as relying on the e-prescription data alone may not be specific enough and could lead to misclassification. Furthermore, an ideal system must be sensitive enough to capture all major safety events during a trial, with the potential for blinded adjudication and rapid reporting/follow-up.
Conclusions
Data collected via NWEH-LDB using sources from SIR and Apollo showed high levels of concordance and provided detailed information on the COPD and asthma population including details on HRU in addition to prescribing data and clinical end-points of interest.
These features have the potential to enable real-world data collection by a research network of general practitioners using EMR with flexible eCRF modifications and link to pharmacy dispensing to provide a platform for assessing novel versus standard treatments in phase 3 or 4 clinical trials. Data from these types of studies could provide complementary information in a more representative and heterogeneous group of patients than traditional RCTs, during the regulatory approval process, with the added value of demonstrating the benefit/risk profile of a drug compared to standard care treatments in the usual care environment. This type of study may be able to enrol a wider range of target patients more efficiently than traditional RCTs by recruiting at their usual source of care, ideally speeding the time to meaningful evaluations as part of a learning healthcare system with public health impact of improved decision making.
Competing interests
NorthWest EHealth received funding from GlaxoSmithKline (GSK) to conduct this study. NDB received consultancy fees from GSK relating to the study. JV and AAW have received consultancy fees from pharmaceutical companies including GSK. KJD is a full-time employee of GSK R&D and owns GSK stock.
Authors’ contributions
HE, JPN, MRD, AAW and NDB were involved in the design and conduct of the study, data analysis/interpretation, and manuscript writing and editing. KJD contributed to the study design, data interpretation, and manuscript writing and editing; NDS assisted with the data analysis/interpretation, and manuscript writing and editing; JMG contributed to the design and conduct of the study, and manuscript writing and editing; JV was involved with the design of the study, data analysis/interpretation, and manuscript writing and editing. All authors read and approved the final manuscript.