Embedding routine health care data in clinical trials: with great power comes great responsibility

In the hierarchy of clinical evidence, well-executed randomised clinical trials (RCTs) are considered to be at the top of the evidence pyramid; however, large phase-3 RCTs are very complex endeavours. They are expensive and inherently slow. Also, the generalisability of RCTs has been questioned, with the underrepresentation of women, the frail elderly and the underprivileged.

Observational analyses of routine health care data (RHCD), derived from very large electronic patient databases, have been proposed as an attractive alternative. Unfortunately, these non-randomised analyses often do not suffice, or may even be misleading in evaluating the efficacy and safety of interventions [1]. Since the beneficial effects of treatments are usually modest, randomisation is a prerequisite for a reliable assessment. Nonetheless, there is considerable overlap in RHCD and trial-specific data. Integration of RHCD in an RCT has great potential, as it would reduce the effort and costs required to collect data, thereby overcoming some of the major downsides of a ‘traditional’ RCT.

This viewpoint first explains what RHCD encompasses, and provides a current overview of cardiovascular registries in the Netherlands (Netherlands Heart Registration (NHR), Heart4Data). Subsequently, the benefits and challenges of using RHCD in an RCT are discussed from a cardiovascular perspective, illustrated by a case study in heart failure.

RHCD refers to large datasets that contain health-related information primarily collected for clinical reasons, as opposed to trial-specific data [2]. These may comprise source data (medical notes, laboratory results, images) and/or secondary data (diagnostic codes, insurance claims, quality registries). Thus, every practicing physician and nurse generates large amounts of valuable data every day that could be repurposed for goals beyond direct patient care, such as quality control, implementation of new guidelines and clinical research. National registries that collect data on hospitalisation or mortality are other examples of RHCD. In the Netherlands, we already have a wide array of cardiovascular registries, with the BHN (Begeleidingscommisie Hartinterventies Nederland) registry for cardiac surgery being one of the first (since 1993). In October 2017, in collaboration with the professional societies (Dutch Society of Cardiology (NVVC); Dutch Society of Cardiothoracic Surgeons (NVT)), three separate quality registries merged into one national registry: the Netherlands Heart Registration (NHR) (Fig. 1; [3]), with registries on atrial fibrillation and heart failure being some of its most recent additions [4]. RHCD are certainly not new, but became more widely available—and therefore more relevant—for secondary analyses following the widespread introduction of electronic health records. The potential of RHCD for clinical research is immense. To illustrate this fact, the electronic health records of the National Health Service (NHS) contain a longitudinal medical history of 98% of the UK population (67 million people), and include over 900,000 heart failure patients [2]. The NHR already covers over 1.5 million cardiac procedures, and over 80,000 procedures are added yearly [3]. To unleash the scientific potential of current Dutch cardiovascular registries, the Heart4Data consortium was established under the wing of the Dutch CardioVascular Alliance (DCVA) [5].

Currently, the NHR is setting up a heart failure registry [3, 25], which is partially based on the previous experience of the CHECK-HF (Chronisch Hartfalen ESC-richtlijn Cardiologische praktijk Kwaliteitsproject HartFalen) registry, which was conducted between 2013 and -2016 in 34 outpatient clinics in the Netherlands [26]. With the NHR Heart Failure registry, heart failure outpatient clinics can, on a voluntary basis, anonymously and periodically share a set of clinical parameters that reflect their clinical care. This set includes patient characteristics, the heart failure therapy received (drug classes and current dose, device therapy) and outcome measures (mortality, heart failure hospitalisations, quality of life). For the purposes of standardisation, the NHR provides a handbook with definitions; moreover, certified data quality control systems are in place to ensure the completeness and quality of data [27]. The NHR then provides feedback on the quality of care at an institutional level (comparison among peers). This may trigger further improvements, with the NHR as an integral part of a plan-do-check-act (PCDA) cycle. TITRATE-HF and associated trials (RELEASE-HF, ENGAGE-HF) are acting as accelerators in making it easier for institutes to join the NHR Heart Failure registry [28]. The 2021 NHR report mentions 13 contributing institutes, and since its initiation 1350 heart failure patients have been included in the registry [25]. For 2023, it is anticipated that > 30 institutes will participate.

The first goal of the Dutch Heart Failure registry is to provide institutes with individual feedback on their quality of care in relation to other institutes, and to initiate a PDCA cycle. Secondly, national temporal trends can be described, e.g. the implementation of new heart failure guidelines. Ultimately, the registry must evolve into a nationwide data platform that is suitable for large, pragmatic, registry-based RCTs, with SELEQT-HF (selenium/CoQ10 supplementation in heart failure) as its first endeavour (Figs. 1 and 2).

To make this initiative a success, certain barriers need to be overcome (Fig. 3; [29]). In order to be able to access the full potential of the data, it is important to embrace the opt-out approach for informed consent for observational studies. In order to achieve nationwide coverage, it is necessary to install multiple incentives for institutes to participate, possibly with central enforcement mechanisms, for example by making it a mandatory requirement by the professional associations (NVVC, NVT), as is already the case for percutaneous coronary interventions and cardiac surgery, or by payment per performance. In addition, registration should be made effortless and of high quality by the use of smart data extraction and standardisation of data acquisition in routine clinical practice, with a coordinating role for the NHR. Under these conditions, the ultimate goal of performing large pragmatic RCTs at low cost will come within reach.

Large, well-executed RCTs remain the gold standard for clinical evidence, but they are complex, costly and slow. The use of RHCD in clinical trials may overcome these limitations. New European laws on privacy must prevent misuse of RCHD: obtaining informed consent by the opt-out method is a balanced way to perform large-scale observational studies that could contribute to the design of large pragmatic RCTs and improve recruitment. Linkage of data is challenging, requires extensive validation and the continuous efforts of many different stakeholders. Nevertheless, there are examples where RCHD sufficed for trial follow-up: the large size of the datasets compensated for underreporting and, by means of randomisation, observations were kept unbiased. For pharmacovigilance, the inherent delay when using RCHD should be acknowledged, and alternative methods for rapid safety reporting are essential. All things considered, the use of RHCD has immense potential, and while challenges exist, examples from the past have shown that these challenges can be overcome.