Most clinicians would agree that the gold standard of evidence to support clinical decision-making is the randomized controlled trial (RCT), either comparing the treatment of interest with placebo, or with an active comparator. There is no doubt that RCTs provide the necessary evidence that a treatment is effective and sufficiently safe for clinical use, but trials do have their limitations when trying to extrapolate results to the people that may be treated outside of their necessarily narrow constraints (ie, in the "real world"). RCTs can only include volunteer patients who meet particular inclusion criteria; therefore, it is rare in diabetes trials that participants reflect the true spectrum of people treated in clinical practice. Furthermore, people who agree to take part in trials may differ from those who do not (eg, by demographic and clinical characteristics), which may have unknown effects on outcomes. People taking part in trials tend to have better outcomes than those who do not and this could be because the process of taking part improves health-related behaviors (such as concordance with lifestyle advice and medication adherence) or because they have a greater interest in their own health in the first place.
What is real-world evidence?
There are many types of real-world evidence and they are not all the same in terms of the quality of evidence produced. They may range from the anecdotal, such as case reports, to carefully conducted analyses of data from clinical databases or prospectively collected data, or even pragmatic trials that collect outcomes from routinely collected clinical data.
Real-world (pragmatic) clinical trials are an important development in clinical research; in their purest form the only interventions are consent and randomization, after that outcomes and adverse events are collected from routine data."
Clinical case reports and case series are potentially prone to biased reporting and patient selection, but can provide helpful insights. Spontaneous event reporting (such as the UK Yellow Card Scheme [1]) may detect adverse events that are too rare to be detected in clinical trials; however, after events have been reported, subsequent reports are prone to reporting bias that can appear to magnify the concern, eg, once it is known that an adverse effect is associated with a new drug, it is much more likely to be reported in people taking it than those who are not. There is also a potential problem of confounding by indication. For example, reports of an increased incidence of cancer in relation to the use of some insulins created a lot of concern [2, 3], but it can be hard to correct for confounders (such as body mass index, which is associated with cancer and likelihood of insulin use). This question was only resolved when robust RCT data with long-term follow up became available [4].
Dedicated databases combined with data collection tools can provide useful evidence that complements RCTs for clinical effectiveness. They may also provide limited data on adverse effects (examples of this in diabetes are the UK Association of British Clinical Diabetologists audits of various treatments such as glucagon-like peptide-1 receptor agonists and sodium-glucose co-transporter-2 [SGLT2] inhibitors [5, 6]), but data tend to be provided only by interested clinicians, largely based in secondary care, which may introduce bias.
Combining data from insurance claim and other clinical databases known to contain high-quality data, with matching of patients using non-parsimonious propensity scores can provide high-quality real-world evidence. This approach was used in the CVD-REAL study, which recently provided evidence that SGLT2 inhibitor use is associated with a lower risk of heart failure and all-cause mortality compared with other glucose-lowering drugs. This was achieved by combining data from over 300,000 patients from six countries [7]. Further analysis has also shown a lower risk compared with dipeptidyl peptidase-4 inhibitors [8].
In the future, we will see more pragmatic trials in diabetes that ask simple research questions, and use routine data to help provide the answers.
Real-world (pragmatic) clinical trials are an important development in clinical research; in their purest form the only interventions are consent and randomization, after that outcomes and adverse events are collected from routine data. An excellent example of this was the Salford Lung Study [9], which reported a pragmatic trial conducted in primary care in patients with chronic obstructive pulmonary disease. There is increasing interest in such an approach for trials in diabetes, and some trials are underway, for example, the DECIDE study (ClinicalTrials.gov identifier, NCT02616666), being conducted in UK primary care, is looking at the effectiveness of dapagliflozin compared with other treatment options as second line to metformin in type 2 diabetes, and is collecting electronic data from routine clinical practice to evaluate the outcomes.
The future
RCTs will always remain the gold standard for providing evidence of the clinical efficacy of treatments. However, there are many unanswered clinical questions in diabetes care that would never be possible to answer using the traditional RCT models of research. In the future, we will see more pragmatic trials in diabetes that ask simple research questions, and use routine data to help provide the answers. Such trials have potential advantages in terms of data collection, but require careful training of investigators who may not usually be involved in clinical research to be able to consent and randomize at the point of care. There are concerns about access to data and confidentiality when such information is used on a large scale, so appropriate safeguards need to be in place. Data quality is only as good as what is recorded in the electronic medical record, so such an approach may not detect adverse events that do not lead to contact with a healthcare professional, so would not be appropriate for early phase or phase III (registration) trials. With more use of pragmatic trials, more people with diabetes and clinicians advising them can be involved in research. This all depends on high-quality data being available to researchers. It is our responsibility to encourage patients to consent to their (anonymized) data to be collected and used in this way, as long as the appropriate checks and balances are in place, and to encourage uptake and use of clinical systems that can support this.