Large Simple Trials and Knowledge Generation in a Learning Health System: Workshop Summary (2013)

8

The Randomized Evaluations of Accepted Choices in Treatment Trials¹

INTRODUCTION

The keynote address was delivered during lunch on the second day of the workshop by Tjeerd-Pieter van Staa. van Staa is the head of research for Clinical Practice Research Datalink (CPRD), which is an observational data

________________

¹ The views expressed during the keynote address are those of Tjeerd-Pieter van Staa and do not reflect the official policy or position of the Medicines and Healthcare Products Regulatory Agency.

and interventional research service jointly funded by the United Kingdom’s National Institute for Health Research and the Medicines and Healthcare Products Regulatory Agency.

RANDOMIZED EVALUATIONS OF ACCEPTED
CHOICES IN TREATMENT TRIALS

Randomized Evaluations of Accepted Choices in Treatment (REACT) is the title of an effort launched by CPRD to conduct pragmatic, large simple trials (van Staa, 2011). The goal of REACT is to test alternative clinical interventions commonly prescribed by physicians that have not been evaluated for their comparative effectiveness through the use of data routinely collected in the single electronic health record (EHR) system used by the United Kingdom’s National Health Service. van Staa began by noting that the motivation for the program lay with the National Health Service’s finding that participants in randomized controlled trials (RCTs) are not necessarily representative of the general population and that the actions of patients and clinicians in RCTs are not necessarily those of patients and clinicians in everyday clinical practice. In other words, RCTs can lack external validity. van Staa illustrated this be mentioning as examples the drugs Celebrex and Vioxx, which were approved on the basis of high-quality RCTs. However, when they were routinely prescribed to members of the general population—most of whom would not have qualified to participate in the RCTs, he noted—serious side effects emerged. Another problem that van Staa identified is the lack of evidence for many common treatments for prevalent diseases. If a given condition has more than one commonly prescribed treatment, physicians and patients do not have a basis for knowing which one is safer or more effective.

One of the missions of CPRD is to use data that are routinely collected in the EHR during care to conduct RCTs of common treatments while imposing a minimum burden on clinicians or patients. CPRD can do this because it has arrangements with a large number of general clinical practices to report health care data regularly. Currently, CPRD has records on about 5 million patients, about 8 percent of the patient population served by the National Health Service, and this database is updated monthly. van Staa noted that CPRD uses pseudoanonymized data and can link its data sets to other data sets, such as hospital data, disease registries, and death certificates.

At this time, CPRD is conducting two small REACT trials to test the feasibility of this approach. One study, RETRO-PRO, involving about 300 patients, is comparing two popular statins, simvastatin and atorvastatin, which have each been previously tested in RCTs against placebo. e-LUNG, the other study, involving 150 patients, looks at antibiotic use in patients

with chronic obstructive pulmonary disease (COPD). Some clinicians prescribe antibiotics for patients who experience a COPD exacerbation, and others do not. However, little evidence on the clinical effect of antibiotics on COPD exacerbations currently exists.

van Staa described the trial processes in more detail. He noted that participants in eLUNG must be recruited in real time, when they are having an exacerbation, which is done through the use of flagging software in the EHR that alerts the clinician that the patient may be eligible for the trial. In contrast, for RETRO-PRO, participating clinicians are sent a list of potentially eligible participants derived from information in the EHR database. Despite the very inclusive nature of the trials, van Staa noted that the trials had few eligibility criteria that had to be evaluated. In RETRO-PRO, they included whether the patients had used a statin before and whether they were at high risk for cardiovascular disease. If neither of these applied, patients were randomized into the trial. In the case of both trials, he explained, participating patients must consent to be included in the study. After this point, clinicians and patients are generally followed unobtrusively. Follow-up information is generally collected from the routinely reported EHR data, unless some anomaly or other question arises, in which case CPRD researchers can contact the clinician for information.

Although REACT trials have the advantage of working in a single very large health care system with a single EHR, in which uniform patient information is continuously reported to a central data bank, they have faced challenges, which van Staa enumerated. One major obstacle is the burdensome requirements for informed consent and regulatory reporting oversight imposed by research governance. van Staa detailed how CPRD proposed a one-page informed consent form, which the ethics committee would not approve and insisted on lengthening. Additionally, regulators required reports every 7 or 15 days, depending on the item. The researchers argued that reports every 7 or 15 days would be burdensome on the clinicians and that monthly reports would be adequate, given the low degree of risk of the interventions. Clinicians were also required to undergo training in the protocol, despite CPRD’s argument that training in prescribing statins was probably not necessary and would be an additional burden on clinicians.

van Staa noted that another problem that the REACT trials face is the lack of incentives for clinicians to participate, given their already very busy care schedules. Possibly as a result of this, CPRD found that relatively few potentially eligible patients were being recruited. A few weeks earlier, for example, van Staa reported that 10 patients had been recruited in a certain practice but that another 260 who could have been recruited were not.

Variable recording and coding of health care data, especially across the linked data sets, is another challenge to the REACT trials, as are variations in practice across National Health Service sites. For example, one of the

issues that RETRO-PRO faced was a price difference between simvastatin and atorvastatin. Many local NHS organizations require clinicians to use the cheaper one, simvastatin, and will not make an exception for research, even though the research might have implications for cost-effectiveness. Because of this, van Staa shared, in some locations, clinicians prescribe atorvastatin for the 3 months of the study and then switch the patient back to simvastatin, preventing long-term follow-up.

Despite the many challenges that the REACT trials have faced, van Staa concluded with the observation that ways must be found to make LSTs work because not doing them has a cost to patients. For example, a patient who currently has a COPD exacerbation receives an antibiotic or does not receive an antibiotic, depending on which clinician he or she happens to see. Instead, he said, clinicians should know whether antibiotics work or not. Regarding the sharp ethical distinction between research and clinical practice discussed earlier in the workshop, van Staa posited that research on chronic conditions, such as COPD, would provide a benefit to the individual participant as well as a generalizable benefit, because the participant would continue to have the condition after the trial ends.

REFERENCE

van Staa, T., B. Goldacre, M. Gulliford, J. Cassell, M. Pirmohamed, A. Taweel, B. Delaney, and L. Smeeth. 2011. Randomised Evaluations of Accepted Choices in Treatment (REACT) trials: Large-scale pragmatic trials within databases of routinely collected electronic healthcare records, abstr. A104. Presented at the 12th Clinical Trials Methodology Conference.