Development of a real-world database for asthma and COPD: The SingHealth-Duke-NUS-GSK COPD and Asthma Real-World Evidence (SDG-CARE) collaboration

Real-world data (RWD) in healthcare refers to data that are routinely collected as part of the care delivery process, rather than through clinical trial settings. RWD can be used to generate real-world evidence (RWE) [1]. The potential uses of RWE are broad, ranging from clinical guidelines development to enabling precision medicine in clinical practice [2‐4]. With the adoption of electronic health records (EHR) and recent legislations such as the 21st Century Cures Act [5], there has been an increasing interest in using real-world evidence (RWE) to satisfy the needs of the evolving healthcare industry [5, 6]. Various initiatives have been organized around the use of RWE, such as the Duke-Margolis Centre for Health Policy RWE Collaborative, to advance policy development related to regulatory acceptability of RWE [7]. RWE has successfully been used by the US Food and Drug Administration in its approval of a cancer therapy drug label expansion in April 2019 [8].

Obtaining RWD from information systems can be done manually or automatically. Manual extraction entails visual inspection of patient records and manual transcription. Such methods are laborious and vulnerable to transcription errors [9]. Given these issues, researchers have increasingly relied on the automated methods for data collection [10‐12]. This allows for efficient, near real-time research on clinical practice, while minimizing the risk of data entry errors.

There have been a number of well-reported large-scale RWD for various clinical care domains, for example, the Clinical Practice Research Datalink (CPRD) which is a primary care database of anonymized medical records [13], European Severe Heterogeneous Asthma Registry, Patient-centred (SHARP) Clinical Research Collaboration [14], UK Severe Asthma Registry (UKSAR) [15], US Advancing the Patient EXperience (APEX) in Chronic Obstructive Pulmonary Disease (COPD) [16] registry amongst others.

In Singapore, a public–private sector collaboration—the SingHealth-Duke-GlaxoSmithKline COPD and Asthma Real-World Evidence (SDG-CARE) collaboration—was formed in 2017 to accelerate the use of RWD. With the above in mind, the collaboration aimed to develop a near real-time integrated RWD database—the SingHealth COPD and Asthma Data Mart (SCDM). The RWD is updated every 24 h, thereby providing a near real-time basis for effectively querying updated clinical and operational data. This is the first large-scale registry in Singapore to fully realize the potential of RWD to improve the care of patients with COPD and asthma. The SCDM is intended to be sufficiently robust to support the conduct of most clinical and health services research trials surrounding asthma and COPD, while ensuring minimal intrusion via the electronic medical record (EMR) systems. This study describes the development of the SCDM and provides an overview of its contents.

We described the development of a near real-time integrated RWD database that includes demographic, clinical, laboratory and radiology data of 36,407 patients (as of December 31, 2019) with asthma and COPD across the spectrum from primary to tertiary care in our healthcare system. Data verification was performed and RWD database demonstrated near perfect agreement with the clinical EMR system. Having developed this data mart within an analytics platform simplifies the access to data via a drag-and-drop interface, rather than having to write SQL codes.

While several asthma and COPD databases already exist, the strength of the SCDM is that it links RWD from primary care to tertiary care and has a rich data capture for asthma and COPD that is near real-time. Data in the RWD are refreshed with a maximum of 24 h delay as the data refresh takes place overnight when the system utilization level is low. With an intentionally broad inclusion criteria and wide range of data elements, from demographics, clinical data, laboratory results to vaccinations and unscheduled visits, we are confident that it is sufficiently robust to meet most asthma and COPD research data needs. Table 6 shows a comparison of the SCDM (asthma only) with two other asthma databases, the International Severe Asthma Registry (ISAR) and Danish National Database for Asthma (DNDA) [4, 31, 32].

As our health system is based on geographical regions, it allows us to serve a captive population of patients who tend to seek care within the same health system. This provides researchers with the opportunity to use relatively more complete longitudinal data to study the disease and care trajectories of asthma and COPD patients as they move across the care chain, from primary care to specialist and acute care. A previous study on this health system showed that among the patients with stable chronic diseases, there were on average approximately 1.6 times more primary care visits as compared to specialist outpatient clinics visits [33]. The registry can further serve as a basis for determining computable phenotypes [34] such as frequent exacerbators, high risk (of poor outcome) patients, fixed obstruction and type 2 high inflammatory phenotype in an Asian population.

With the heavy investments in developing the ETL pipelines, we also designed the SCDM with flexibility and sustainability in mind. For this, we deliberately chose to perform minimal transformation to preserve the raw data and minimize information loss. Unlike specific disease or national registries that combine and transform raw data to derive composite variables, our database consists of almost completely raw data in their original format. The registry adopted the same classification as the raw data, and followed the International Classification of Diseases, ICD-9 and ICD-10 [35], and the Systematized Nomenclature of Medicine-Clinical Terms (SNOMED-CT) [36] coding standards. At the time of the study, Singapore adopted the Australian-refined Diagnosis Related Groups (AR-DRG) version 6 coding system [37]. Although not using a common data model (CDM), such as Sentinel, Observational Medical Outcomes Partnership (OMOP) and Patient Centred Outcomes Research Network (PCORNet), may make our data less linkable with data from other databases, we felt that the trade-off was in favour of generalizability of the data to meet a wide variety of definitions [38‐41]. Amongst the various classification systems used, mappings exist between them to ensure the interpretability of results across multiple systems and globally across time. Furthermore, as the healthcare CDM space is still actively developing, we will have the option of migrating our database to a CDM [42].

Minimal filtering of the data was done as we attempted to capture the complete dataset that is available throughout the clinical processes. For example, we chose to import all medications prescribed for a patient, including non-asthma related medications, instead of filtering them based on a pre-selected list of asthma-related medications. This endowed the SCDM with the following advantages: (1) the flexibility to select medications of interest to their own study; (2) the capability to study effects and associations with non-asthma medications, and (3) the adaptability to include any new asthma and non-asthma related medications that may be prescribed in future without the need to update the underlying ETL pipelines.

Although agile methodologies are gaining in popularity in IT development space, we elected a hybrid methodology where the waterfall project plan is required to secure the resources for milestone delivery and to ensure governance requirements are duly complied. Some of the requirements to determine the cohort and data elements were well-defined and amenable to a waterfall methodology whilst within the design and development process, we have adopted the agile methodology for the refinement and implementation of the requirements [43, 44]. The uncertain requirements inherent in the design of a registry which leverages on clinical and operational data requires flexibility in efficient requirement changes [25]. The hybrid framework also allowed us to perform robust data verification that adheres to national and organizational data security policies at the final phases of the SCDM development process. Limited by organizational and data governance constraints, whilst requiring the need for flexibility through close stakeholders’ engagement to refine the data requirements, we have adopted a hybrid waterfall-agile approach towards the development of the SCDM [27].

Our RWD database is not without its limitations. Although it currently includes patients with asthma and COPD follow-up at SGH RCCM specialist clinics in the tertiary hospital, it does not include those who are only followed-up with other departments such as Internal Medicine, Occupational Medicine, or those who only visited the Accident and Emergency Department (A&E) within the same hospital, and were not referred to the SGH RCCM. Also, although the data mart contains rich clinical details, a significant proportion of this is in free-text format which requires additional data mining tasks before the data can be analysed. One example is the smoking status data where almost half was not available from structured data input fields. With the continual effort to encourage the adoption of standardized clinical templates for asthma and COPD, we hope to improve the quality of data capture. Furthermore, the standardization of semi-structured text formats will further enable us to make use of natural language processing (NLP) algorithms to derive relevant information from the textual data. It is envisioned that we could augment the registry with NLP capabilities to improve data completeness.

Moving ahead, as the next phase in the SDG-CARE collaboration, we will leverage the SCDM in several areas. One immediate area is to develop interactive dashboards that will be able to provide a real-time overview of the key statistics in SCDM, monitor routine practice and for clinical decision support. In terms of clinical research, the team has embarked on a project using SCDM data to develop a model that uses routinely available data in primary care to predict asthma exacerbations. This will support identification of at-risk patients such that earlier and more resource-intensive interventions may be applied for this group. By working with SCDM data which is already routinely captured in the EMR, the team will be able to more easily deploy the model for use. The team also intends for the SCDM to influence public health policies, and is using the real-world data to investigate the impact of guideline non-conformance, such as yearly influenza vaccinations, on clinical outcomes, such as visits to emergency or hospitalizations for pneumonia. Findings from this may potentially result in guideline changes or lend support to tighter compliance. Further down, we also envision that the SCDM will provide the foundation for RWD collection for impactful, large-scale pragmatic clinical trials, akin to the applications from the Salford Lung Study [45].

In parallel, we will also work towards iteratively enhancing the SCDM. In the next phase, we will look toward including data from the only public paediatric and maternity tertiary hospital in Singapore—KK Women’s and Children Hospital (KKWCH). This will open up the potential to observe long-term trajectory of asthma from paediatric to adulthood and to perform more in-depth studies on determinants of poor outcomes.

We described the development of a RWD database for asthma and COPD in the largest public health care system in Singapore, spanning primary care to specialist and acute hospital care. By adopting a systematic process, we were able to ensure that it was robust, valid and applicable. This RWD database provides a unique opportunity for clinical and health services research in asthma and COPD, which can ultimately improve the care delivered to our patients.