Adaptive approaches to randomised trials of nonpharmaceutical interventions Chris Metcalfe University of Bristol, Bristol Randomised Trial Collaboration, & MRC ConDuCT Hub for Trials Methodology
Adaptive trial designs A clinical study design that uses accumulating data to decide how to modify aspects of the study as it continues, without undermining the validity and integrity of the trial. Kairalla et al 2012. Trials 13:145 Pre-specified flexibility in sample size, allocation ratio, dropping & adding arms
Mainly for pharma trials? True to say that adaptive methodology is more commonly applied to trials of pharmaceuticals? Funded by pharma? Grant funding tends to be for a tightly prespecified trial design Hence separate feasibility studies more common for non-pharmaceutical trials Internal pilots are becoming more common
Feasibility studies Exact implementation of intervention? Can the intervention be delivered? Can enough people be recruited? Do participants / physicians follow the allocated treatment, and can this be measured? Can outcome be measured? Blind to treatment allocations?
The ProtecT study Surgical / radiological procedures evolve within lifetime of project Technical procedures need to be learned Patients may not want to be randomly allocated to one of two very different interventions surgery versus conservative management Primary assessment point is at 10 years average follow-up
A more formal approach Traditionally, the findings of feasibility studies have been interpreted in an informal way. (May include qualitative and quantitative information) However, funders are imposing greater formality as stop / go / adapt rules. Are there adaptive methodologies that can be adopted in feasibility studies & other flexibilities in trials of non-pharma?
Sample size Sample size is based around a minimum clinically important difference Interim analyses will pick up when a bigger effect is supported, and the trial can stop early So long as outcome can be ascertained early enough, this applies to trials of all medical interventions
Recruitment Often more of an issue in trials of nonpharmaceuticals: Public health trials often recruit healthy people Patients can be reluctant to be randomised between very different treatments Trials staff often less familiar recruiting to a non-pharmaceutical trials with very different treatments
Is recruitment rapid enough? Common in feasibility studies to attempt increases in recruitment by: Training of recruiters Widening eligibility criteria Formal evidence of recruitment rate later on in feasibility study and of improvements through eligibility criteria?
Recruitment later on Could present the recruitment rate for the latter part of the feasibility study, but may be seen as cherry picking, plus will be a imprecise estimate Possible to borrow strength from earlier on?
Widening eligibility criteria Eligible, wider criteria Not eligible, wider criteria Eligible, original criteria Always recruited Not possible Not eligible, original criteria Additional recruits Never recruited Sample size for test that wider criteria improve recruitment?
Can one arm be dropped early? ProtecT compares the following treatments for localised prostate cancer: Surgery Radiotherapy Conservative management, radical if disease predicted to worsen Could we see a broader comparison here between radical and conservative management?
Can one arm be dropped early? The degree of overlap between different arms of a drug trial is clear where each arm is a different dose of the same drug, or different combinations of two or three drugs Some parallels in dosage of nonpharmaceuticals e.g. sessions of therapy Tends to be more reluctance to combine two non-pharmaceuticals on the basis of an abstract commonality
Can evolution be accommodated? In trials needing long recruitment and/or follow-up periods, one or more of the treatments may be refined E.g. introduction of robotically assisted surgery for localised prostate cancer Have allowed inclusion of these cases unlikely to affect prostate cancer mortality outcome, but may reduce side effects
Can evolution be accommodated? Not aiming for definitive evidence of benefits from robotic surgery want to preserve the integrity of the surgery arm If restricted to certain centres, could look at variation in effect over centres (metaanalysis approach?) If 100% coverage once introduced, then could have separate centre effects pre- and post- introduction
Can learning be accommodated? New highly technical procedures may show increases operator skill during the lifetime of the project Will be claims that including the data collected first will dilute the full treatment difference
Conclusion Some initial ideas have been presented on how some feasibility study questions can be seen in the more formal adaptive design framework