Digital health has opened up the possibilities for new data collection methods and trial designs. These are crucial for sustainable drug and device development. The limitations of current study designs are clear. Observational studies have inherent biases, rendering them inappropriate to evaluate novel therapies. Randomised controlled trials (RCTs) provide high quality data but are inefficient and costly: the design and start-up processes are lengthy, delivery is challenging, and more than 50% fail to recruit the target number of patients. While RCTs are the gold standard, participants often do not reflect clinical practice in terms of age, sex, and comorbidities.
There is a move towards using real-world data – such as electronic health records (EHRs) and registries – to collect RCT outcomes. Evidence is accumulating that with real-world data, large RCTs can be conducted quickly and at low cost. One example with EHR data is the REDUCE cluster RCT. It enrolled more than 600,000 patients from 79 general practices and found that inappropriate antibiotic use could be lowered with a simple education and support tool. In the cardiology context, the TASTE trial illustrates how registry data can be used in an RCT. It found a neutral effect of thrombus aspiration compared to standard percutaneous coronary intervention in patients with ST-segment elevation myocardial infarction.
Across the world, there are a growing number of real-world sources of cardiovascular data that can be used in research. These data are used by the BigData@Heart Innovative Medicines Initiative (IMI) project. More data collection systems are being established: through EuroHeart, the ESC plans to support countries to develop their own registries by providing a common IT and dataset infrastructure. In future these registries will be used to deliver registry-based RCTs.
The use of observational data in research is being transformed with machine learning. When applied to registry and EHR data, machine learning can identify factors associated with an outcome and predict incident events. To take one example, in the SwedeHF registry of 45,000 patients, four clusters of factors associated with one-year mortality stratified patients more effectively than left ventricular ejection fraction.
Wearables and smartphones enable continuous collection of real-world evidence for research and post-market surveillance. The VENTASTEP study illustrates how to integrate traditional and digital data sources, including a smartwatch for activity and heart rate assessment; a nebuliser connected to a smartphone app for inhalation frequency and completeness; the six-minute walk test; and health related quality of life by questionnaire. A pre-study feasibility questionnaire revealed that patients were unwilling to give GPS data or activity overnight, so these were excluded from the protocol.
- What can be done about missing data from EHR datasets? Some investigators impute; others do not. Machine learning can explore whether missing data could affect the analysis. Some investigators report that missing EHR data are often not vital for outcome analysis.
- There is a need to harmonise data definitions and ontologies. This is one aim of the BigData@Heart programme, of which the ESC is a partner, starting with heart failure, atrial fibrillation, and coronary syndromes.
- Will regulators and reimbursement authorities accept real-world evidence? Does it help establish value? Given its limitations, there is caution regarding the use of real-world evidence, as opposed to RCT trial evidence, for regulators. Reimbursement authorities could give a provisional opinion on a new development then wait for real-world evidence to show the population impact.
- Regulators need to clarify their requirements for safety reporting in registry-based RCTs and EHR studies – there is currently inconsistency. Studies of established drugs should not have the same rules for adverse event collection as traditional RCTs since new safety signals are not expected. Some investigators argue that adverse events will be picked up by EHRs while others say that additional data collection is needed for accurate and reliable monitoring.
- Patient fatigue may set in when frequently asked to provide information. They do not want to answer quality of life questionnaires on an app every day. Emojis are an easy way to notate their state each day, but they lack depth and cannot be compared with other established endpoints. What does an improvement in emojis mean?
- Some citizens are reluctant to share data from wearables due to fears that insurance companies might raise their fees. Do insurance companies have access to these data? Could participation in a registry lead to a fee hike due to failure to achieve 10,000 steps a day?
- What should be done when wearable technologies disagree? A smartphone may give the opposite result to a smartwatch – what are the implications? One solution is for the study protocol to specify: use this data from this device, and that data from that device.
RCTs will remain the foundation of evidence-based clinical practice, particularly for treatment outcomes. Registry-based RCTs are ideal for simple clinical questions and could become the gold standard for real-world testing and for monitoring the effect of implementing specific therapies. Using EHRs as data sources will be the next step forward – innovative methods are needed so that they provide the same level of evidence. Key benefits will be generalisable results and repurposing of data already collected as part of standard care. Clarity is needed on how regulators, reimbursement authorities, and guidelines task forces view the quality of real-world evidence and how to integrate it into the clinical armament.
Slides and videos
Review the slides and presentations from the CRT meeting.