In 1967 Daniel Schwartz and Joseph Lellouch1 argued that there are 2 kinds of randomised controlled trials (RCTs) embodying radically different attitudes to evaluation of treatment. They named these trials “pragmatic” and “explanatory” and stated that these 2 attitudes require different approaches to the design of an RCT. The pragmatic attitude seeks to directly inform real-world decisions among alternative treatments, and Schwartz and Lellouch show that this purpose is satisfied in trials that test feasible interventions on typical patients in common settings, with usual care as the comparator, to widen real-world applicability. The explanatory attitude, in contrast, is directed to understanding a biological process by testing the hypothesis that the specified biological response is explained by exposure to a particular treatment. Tight restrictions on eligible participants, intense and closely monitored treatment, inactive control interventions (such as placebo), and an idealised healthcare setting maximise the comparison between intervention and control groups and increase the ability to test this kind of hypothesis.

What attitude to RCT design is most useful for patients and clinicians? Clearly, the trial has to ask an important question that is relevant to some aspect of the care clinicians provide to their patients. The clinicians and patients in the trial should resemble the clinicians who are reading the trial report and the patients they typically treat. The intervention being evaluated in the trial should be deliverable by the clinician, and the outcome being used to judge whether the intervention is effective has to be something that the clinician and his or her patients recognise as being worth influencing. In short, the trial has to be applicable, or have what is often called external validity.2

Consider the NASCET trial.3 It asked the following question: among patients with symptomatic 70–99% stenosis of a carotid artery (and therefore at high risk of stroke), can the addition of carotid endarterectomy (performed by an expert vascular or neurosurgeon with an excellent track record) to fully specified best medical therapy, compared with best medical therapy alone, reduce the outcomes of major stroke or death over the next 2 years? The outcomes are certainly important but what if the skill level of local surgeons is not clearly known, and most patients in a particular setting are at lower risk of stroke or have more comorbidity? Could you apply these trial results directly to your patients in your context? Contrast this with an Australian trial comparing outcomes of elderly patients taking angiotensin-converting enzyme (ACE)-inhibitors with those taking diuretics.4 This trial asked the question: among patients between 65 and 84 years of age having systolic blood pressure of ⩾160 mm Hg or an average diastolic blood pressure of ⩾90 mm Hg (with systolic blood pressure ⩾140 mm Hg), does an ACE-inhibitor-based regimen (delivered by the patient’s family doctor), compared with a diuretic-based regimen (also delivered by the family doctor), reduce the rate of cardiovascular events or death by any cause? Here the trial had few exclusion criteria; management of antihypertensive therapy and achieving blood pressure goals were left to family doctors; the addition of other drugs was allowed in both groups; and end points were assessed from routinely available records. This is much closer to usual care so the results are more likely to be applicable there. In both trials, the RCT design minimises bias (threats to internal validity), but the randomisation does not, itself, guarantee that the findings of any particular RCT are readily applicable. The design decisions (exclusions, setting, outcomes) made by the 2 sets of trialists are different, and these decisions affect the ease with which clinicians can apply the trial findings.

Schwartz and Lellouch1 wrote that “most therapeutic trials are inadequately formulated, and this from the earliest stages of their conception,” referring to the mismatch between the use we make of most trials (which is to inform patient and clinical decisions among alternative treatments in usual-care settings) and the design of these trials (which generally takes the opposite form, best suited to testing causal hypotheses). The only review of this subject has identified <100 pragmatic RCTs out of the quarter million or so RCTs listed by the US National Library of Medicine,5 which suggests that existing RCTs are mostly at the explanatory end of the explanatory–pragmatic continuum. While there is no doubt that explanatory trials are important in understanding the mechanism of action of interventions, trialists too often choose this design approach unthinkingly. Because information from an explanatory trial is unlikely to directly inform a pragmatic, clinical question, the mismatch between the context in which most clinicians work and the clinical context of the RCTs available means that clinicians and their patients are left without directly applicable evidence upon which to base most treatment choices.6

So, why are there so few pragmatic trials? The requirements (and influence) of drug licensing is likely an important factor. The US Food and Drug Administration (FDA) does not provide much guidance on the design of trials, but what guidance it does provide argues against the pragmatic attitude: “One problem is that there are numerous ways of conducting a study that can obscure differences between treatments, such as poor diagnostic criteria, poor methods of measurement, poor compliance, medication errors, or poor training of observers. As a general statement, carelessness of all kinds will tend to obscure differences between treatments. Where the objective of a study is to show a difference, investigators have powerful stimuli toward assuring study excellence.”7 Although usual care is often imperfect, it is by definition also the care that most patients get. Yet the FDA equates pragmatic attitudes and design choices, which are rooted in approximating usual care, with poor or careless study conduct. Such studies produce findings that are not directly applicable to patients or subgroups of patients for whom the treatment will often be needed.6

There is disquiet about the remoteness from real world decision-making of regulatory RCTs8 among clinicians and third-party funders like Medicare, because the potentially lower absolute benefits of a new treatment in usual care might be outweighed by higher rates of side effects and poor outcomes among their typical patients, who might well have more complex comorbidity than those in the explanatory trials. As a consequence, interest is rising in trials whose designs produce information that directly supports clinical decision making. We believe that appropriately designed and reported RCTs with pragmatic attributes (and systematic reviews and cost-effectiveness analyses based upon them) will help meet these needs. Of course, if a benefit is extremely large, it may not need an RCT at all, and if a side effect is very rare or occurs after many years, an RCT may not be large enough or of sufficient duration to detect it. But in the middle range, where the benefits of most interventions lie, a pragmatic attitude to RCT design could provide the foundation for such efforts as those in the USA on comparative effectiveness research.9

How can clinicians spot the differences between an explanatory trial and a pragmatic one? RCTs are not one or the other—they sit on an explanatory–pragmatic continuum, and trialists are forced to make compromises when designing their trials. There are, however, pointers to a trial’s place on this continuum. The table lists broad differences in intent and design features between explanatory and pragmatic trials. Clinicians could consider these when reading through a trial report to understand how the designers have oriented and conducted their trial. Looking at how trialists have handled these design features should make these judgments easier.

For designers of trials, we have contributed to a tool to help differentiate between explanatory and pragmatic attributes.10 The usefulness of a trial depends not only on design but on the similarity between the user’s context and that of the trial. Although it is unreasonable to expect the results of a trial to apply in all contexts, trials should be reported in such a way that users of the results can make meaningful judgments about applicability to their own context.2 For clinicians planning to publish a trial, we have published an extension to the CONSORT statement with some suggestions for the reporting of RCTs designed using a pragmatic attitude.11

Schwartz and Lellouch end their paper with this conclusion: “Most trials done hitherto have adopted the explanatory approach without question; the pragmatic approach would often have been more justifiable.” More than forty years later, this remains true. Although RCTs with explanatory attributes have their place, they should not be the default. It is now time to shift our design choices so that they match our usual purpose in conducting a trial, most often to directly inform the decisions of real-world patients, clinicians, and those who commission healthcare provision.