Comment on: The DISCO Trial – GLA:D vs saline injections
Protocol:
- Exercise therapy and patient education versus intra-articular saline injections in the treatment of knee osteoarthritis: an evidence-based protocol for an open-label randomised controlled trial (the DISCO trial). Bandak E, Overgaard AF, Kristensen LE, Ellegaard K, Guldberg-Moller J, Bartholdy C, Hunter DJ, Altman RD, Christensen R, Bliddal H, Henriksen M. Trials. 2021 Jan 6;22(1):18. https://trialsjournal.biomedcentral.com/articles/10.1186/s13063-020-04952-5
Primary report:
- Exercise and education versus saline injections for knee osteoarthritis: a randomised controlled equivalence trial. Bandak E, Christensen R, Overgaard A, et al. Annals of the Rheumatic Diseases 2022;81:537-543. https://ard.bmj.com/content/81/4/537
1-year outcomes report:
- Exercise and education versus intra-articular saline for knee osteoarthritis: A 1-year follow-up of a randomized trial. Henriksen M, Christensen R, Kristensen LE, Bliddal H, Bartholdy C, Boesen M, Ellegaard K, Guldberg-Møller J, Hunter DJ, Altman R, Bandak E. Osteoarthritis and Cartilage, 2023. https://doi.org/10.1016/j.joca.2022.12.011
As a clinical epidemiologist, what I look for first in a trial is a good match between the rationale and the key design elements of a clinical trial. The key elements are the participants and setting, intervention characteristics, comparison/control condition characteristics, outcome of interest, and timing/timeframe (PICOT). Then, after that, I zoom the focus in more closely to look at the usual ‘checklist’ quality and risk of bias details.
The characteristics of these key elements are fundamentally different between a mechanistic study (how does a treatment work?), an efficacy trial (can this treatment cause a specified outcome?), an effectiveness trial (does this treatment provide specified benefits in the intended setting, compared with a plausible alternative of no/background/’standard’ care or – in the case of comparative effectiveness – another competing treatment?), or a pragmatic trial (once implemented, are the expected effects apparent in routine delivery?).
The authors stated in the protocol paper that the DISCO Trial was designed to be an efficacy trial.
In an efficacy trial, the only difference between the ‘control’ and the ‘intervention’ groups should be whether the active intervention is delivered or not, with all other conditions including (especially) the non-specific contextual effects held constant (i.e. the same in both arms). Only if this is the case, can causation be attributed to the active intervention.
That’s not the case in this design. The contextual effects of getting an ultrasound-guided injection from a doctor in a highly procedural medicalised context has very powerful non-specific placebo effects (and saline may even have an active effect in the joint), and these known to be very different from the contextual non-specific placebo effects of an education+exercise intervention in an outpatient physiotherapy department context.
Ironically, the authors had stated this trial was needed as “none of the numerous randomised controlled trials of exercise and education for knee OA has used adequate sham/placebo comparison groups”. But therein lies my main concern with this trial: it cannot be interpreted as an efficacy study because the ‘control group’ intervention has very different contextual effects than does the ‘active’ intervention, therefore offering no ‘control’, and thus causation cannot be attributed to either intervention.
Also, efficacy trials (also known as explanatory trials) should be conducted under ideal conditions. In this study, the education & exercise sessions took place in the exercise facilities in the department of physiotherapy at Frederiksberg Hospital, by GLAD trained providers. This is more of a generalisable, real-life setting, and appropriately the authors state “we expect similar benefits of the GLAD programme”, as treatment effects are usually greater in ideal settings and attenuated in more routine, real-world settings. That’s fine, and is not necessarily a problem provided a priori expectations of treatment effect are not unreasonable. The COVID-19 pandemic struck halfway through recruitment, with lockdown forcing the suspension of the trial, but that only affected a small proportion of participants (around 15%), so should not have had undue influence.
However the within-group change in the primary outcome in the education + exercise group in this trial (14.9% pain reduction) was much lower than that seen in usual GLAD programme outcomes: a recent study of >3,700 GLAD patients in the same country and similar timeframe (from May 2020) showed almost double that (27.6% pain reduction at 12 month follow-up) [DOI: 10.1002/msc.1765], and a report of >13,000 GLAD patients with knee OA also showed 27.6% pain reduction at 12 months [https://doi.org/10.1016/j.joca.2022.02.001]. As the DISCO trial limited inclusion to people with an average knee pain of at least 4 out of 10, we would expect to see an enhanced pain reduction due to the ‘regression to the mean’ phenomenon.
Furthermore, an efficacy trial is typically run to ensure low loss to follow-up, often 10-20% at one year: the authors of the DISCO Trial quite appropriately expected of up to 23%. However they ended up with a relatively large attrition of 30-40%, which is similar to the real-world GLAD registries at their 1-year follow-up. It is not clear why unusually small pain reduction and large loss to follow-up has occurred within this trial, compared with the authors’ own reasonable, evidence-based a priori expectations.
So the design doesn’t meet the standard for an efficacy trial: it cannot test causation. So is it an effectiveness trial? The answer to that lies (among other considerations) in its generalisibility to the intended population and setting: in what real-world healthcare setting is it plausible that a series of ultrasound guided injections of saline, by a doctor? I don’t think this is offered in any real-world context. So it has no sensible interpretation as an effectiveness or comparative effectiveness study.
Is it a mechanistic study? The authors correctly state that “the study is not designed to expand the knowledge about the underlying mechanisms.”
What does it tell us? All it really answers is “Is the magnitude of effect from the contextual non-specific placebo effects of ultrasound guided injection from a doctor in a highly procedural medicalised context DIFFERENT from the magnitude of effect from an education + exercise intervention plus the contextual non-specific placebo effects of an outpatient physiotherapy department context”. Previous research has already separately shown the magnitude of within-group effect of injections is similar to the within-group effect of education + exercise. The authors show this in great detail in Fig.1 and Fig.2 in the protocol paper. On this basis, they specified it was an equivalence design study, and powered the trial appropriately. So all this trial does is compare those two very different interventions head-to-head, and conclude that, yes, the magnitude is indeed similar. What now? What does that mean?
This trial doesn’t answer any sensible question, in my opinion, because it is useless as a efficacy trial, is explicitly not a mechanistic study, and has no generalisable implication as an effectiveness or comparative effectiveness trial. As a result, whatever might be the implications of the findings of this trial cannot be sensibly interpreted.
The authors acknowledged from the outset that “The interpretation of the results of this trial will likely be difficult and controversial” – in this they are correct – but they then stated that it “… will contribute to a better understanding of the bias introduced in the effect estimation of classically unblindable exercise and education interventions for knee OA.” I haven’t heard any explanation of this trial that has made much sense to me, or moved me to agree with this premise.
Indeed, the entire rationale for the study seems to be because they took a vote among the 13 people on the study authorship team about whether exercise + education, or saline injection, or neither would be superior, and from the split result they made a claim of equipoise. Was that really was the basis of spending this much time, money, trainee effort, volunteer effort, clinical opportunity cost, participant burden, participant risk, and social license? If so, it’s research waste – I would recommend the authors consider reading some Chalmers & Glasziou articles for their next journal club: https://www.sciencedirect.com/science/article/pii/S0140673609603299?via%3Dihub ; https://www.bmj.com/content/363/bmj.k4645 😊.
The authors say that “because the ‘active’ ingredients are unknown… designing and executing an adequate and ‘blindable placebo’ version of an exercise and education intervention is impossible.” I have some sympathy for the first half of this statement, because studies of the ‘components’ of exercise interventions (including, but not limited to: strengthening, muscular endurance, cardiovascular endurance, neuromuscular coordination, neurophysiological, psychological, etc) have shown inconsistent results in regard to the classic criteria for causation, for example, effect size, consistency, dose-response relationship, and it has proven difficult to pull apart their inter-relationships. However I disagree that it is impossible to design a blindable placebo or sham that can serve as a feasible comparison for the specified ‘components’ and hold the others constant between the two arms of a trial. These questions deserve rigorous investigation, which will require a well-thought-out series of tightly-designed trials.
Prof. J. Haxby Abbott, PhD, DPT, FNZCP
Director, Centre for Musculoskeletal Outcomes Research
University of Otago Medical School
Dunedin, New Zealand
