By Deborah Borfitz 

September 17, 2024 | Researchers in the UK recently emulated a large, randomized controlled trial (RCT) testing the effectiveness of two blood thinners using data from real-world patients to help understand whether the real-world data can be used to explore treatment effects in people underrepresented in the study. While results of the reference trial and its simulation didn’t entirely align, the approach could help fill the evidence gap for population groups largely missing in RCTs, according to Emma Maud Powell, Ph.D., an alumna of the London School of Hygiene and Tropical Medicine who led the project.   

“Routinely generated healthcare data is a valuable resource that patients contribute to every time they go to the doctor... [potentially] helping to advance medical research,” she says. For the latest emulation study, published in PLOS Medicine (DOI: 10.1371/journal.pmed.1004377), the value was derived from data regularly collected by general practitioners and deposited in the Clinical Practice Research Datalink (CPRD) Aurum, a database containing routinely collected data from primary care practices in England, capturing diagnoses, symptoms, prescriptions, referrals and tests—as well as demographics and lifestyle risk factors such as smoking. 

CPRD Aurum is in turn linked to the Hospital Episodes Statistics, a data warehouse containing records of all patients admitted to NHS hospitals in England, providing an additional layer of information on hospital admissions and visits to accident and emergency units. This is how the research team caught their outcomes of interest, including stroke and major bleeding, says Powell. Records on any deaths came from a linkage to mortality data compiled by the Office of National Statistics.  

The reference trial, in this case, was the pivotal ARISTOTLE trial of apixaban versus warfarin. Although patient eligibility, selection, and analysis approaches of the original trial were mirrored for the simulation exercise with the real-world cohort, the RCT detected apixaban as being superior whereas the emulation study found the treatments to be roughly the same, she reports.  

An assortment of factors likely contributed to the outcome difference, including the fact that fewer patients in the CPRD Aurum group are of Asian ethnicity—a risk factor for hemorrhagic stroke among those taking warfarin—and lower use of certain concomitant medications among real-world patients in the UK, says Powell. Due to prescribing habits in the UK, use of both aspirin and amiodarone were lower in the CPRD Aurum cohort compared to those participating in the ARISTOTLE study. In addition, the average time in therapeutic range of the patients on warfarin in the emulation study was higher than that reported in ARISTOTLE. 

A key limitation of the study is the bias introduced when patients start on one drug but switch to the other, Powell says. When apixaban became available, it might be expected that patients doing less well on warfarin might give the new drug a try. 

Even before the study started, researchers expected to see slightly less benefit for people on warfarin, since this was what was seen with the EU subgroup in the RCT, she adds. During the National Institute for Health and Care Excellence (NICE) review of ARISTOTLE, it was hypothesized that the “time in therapeutic range” of the warfarin users in the study may be lower than what is typical in UK clinical practice. 

Design Guide

The overarching motivation for the stroke study was to better inform clinical decision-making about how best to prevent stroke in patient groups with non-valvular atrial fibrillation who were under-represented in or excluded from the pivotal studies of the treatments, says Powell, who now works for Compass Pathways, a biotech company focused on treatment-resistant depression. “We did this reference trial emulation to understand some of the limitations of real-word data and how comparable the results are with the clinical trials.” If successful, she adds, “then we could feel more confident in results that we might get in underrepresented groups.” 

The study established that using an existing RCT as a guide for the design of observational analysis of routinely collected patient data is an effective way to understand whether we can reliably estimate the treatment effects and risks of blood thinners given to patients with atrial fibrillation using this data source, Powell continues. Analysis of the noninterventional data generated results demonstrating noninferiority of apixaban versus warfarin, which was consistent with the prespecified benchmarking criteria. 

Since trial emulation was the goal, the primary endpoint was the same as in the reference trial—namely, the effectiveness of the two drugs in preventing stroke (ischemic or hemorrhagic) or systemic embolism. Researchers also looked at the same key safety outcomes, including major bleeding and all-cause mortality, she adds.  

Although the matching process has its limits, the reference trial emulation approach provides a framework that can be adapted to investigate treatment effects for other conditions, Powell says. But the results suggest that differences in standards of care between geographies can influence country-level treatment estimates, depending on the representativeness of the study population.    

Both the U.S. Food and Drug Administration (FDA) and the European Medicines Agency are interested in the extent to which real-world data can be tapped to “bridge” the evidence gap left after clinical trials. “Despite the best efforts of the people conducting the trials, you can still end up missing certain patient groups... for safety or feasibility reasons.” 

Data Needs 

Similar reference trial methodology has also been applied by a colleague who did a trial emulation for antihypertensive drugs using CPRD Aurum as the data source and was able to study under-represented ethnic groups. In the U.S., meanwhile, the RCT-DUPLICATE initiative of Brigham & Women’s Hospital, the FDA, and Aetion, Inc. has used real-world evidence from U.S. data to replicate numerous published clinical trials. 

For Powell and her team, the focus on oral anticoagulants for stroke prevention in atrial fibrillation was a matter of practicality. Other clinical trial therapeutic areas either didn’t have enough people or there wasn’t ready access to the needed real-world data for the emulation exercise, she says. 

In the UK, Powell points out, only treatments prescribed by primary care physicians show up in the CPRD Aurum. Emulation also requires that the drugs being studied be widely prescribed and the outcomes reliably captured in the data, which is why the preponderance of RCT emulation studies have been for commonplace cardiovascular disorders and chronic diseases such as diabetes.