- Posted By: Michael Schlussel
The 3+3 design remains the mainstay for dose-escalation studies despite the introduction of model-based designs almost 30 years ago. The Early Phase Clinical Trials Research Section of the NIHR Statistics Group organised a one-day '3+3 consensus meeting' on 3rd December 2014 in Cambridge to review the prevalent use of the 3+3 design and explore model-based designs in finding maximum tolerated doses (MTD) in phase I clinical trials with a view to identifying what new designs should be adopted.
The day began with three short presentations providing overviews and discussions on the practical applications of rule-based and model-based designs. After highlighting the limitations of the commonly used rule-based methods, Adrian Mander and Thomas Jaki presented alternative adaptive methods (Bayesian and frequentist), which not only use cumulated data in deciding the next dose assignment but generally shorten the trial duration. Alun Bedding (Roche Pharmaceuticals) and James Matcham (AstraZeneca) also presented successful model-based dose escalation case studies from the pharmaceutical industry.
So why does 3+3 design remain so popular despite the better model-based alternatives proposed almost 30 years ago? In an attempt to address this paradox, participants were divided in groups to discuss, suggest ways of improving and promoting 'better' designs for dose-escalation studies. A common theme was the need to harmonise potential conflicting attitudes between statisticians and clinicians. While all parties involved in clinical research aim to improve treatment for future patients, clinical considerations must always overrule any competing opinions, particularly for patients now.
Although no silver bullet exists to solve this problem, the meeting reached a consensus: to form a working group with key mandates to (i) support documentation guidance templates; (ii) help collect success stories to promote 'modern' designs; (iii) apply for funding for database repository to collect information on data throughout design development; and (iv) promote continued learning by supporting statisticians to acquire competence in 'new' methods (e.g. MAST).
In addition to proposing prominent names as ambassadors, the working group will work collaboratively with relevant stakeholders (funders, regulators, industry, clinical investigators, CTUs, journal editors, ethics committees and the academia). As minimum requirements, these stakeholders will be made aware of the issues prior to reaching a consensus, asked about the best way to influence them and/or to engage them. Working group leads will help achieve these together with frequent/structured meetings and communications.
While rule-based methods may have been practical and useful 30 years ago, the changing nature of drug development requires better designs. However, guidelines are not enough and often rarely followed. There is a need to actively engage and support statisticians involved in design of early phase clinical trials. What is more, it is incumbent upon statisticians to be more persuasive, inclusive and clear in communicating the merits and limitations of both rule-based and model-based designs to their collaborating partners and other stakeholders.