Abstract 1559P
Background
Conducting a traditional, adequately powered multi-arm exercise trial is often challenging in terms of time and available resources. Novel, efficient trial designs are needed for rigorous comparison of multiple treatment arms. This study illustrates a Bayesian adaptive decision-theoretic design using a multi-arm exercise oncology trial.
Methods
In the PACES trial, 230 breast cancer patients receiving adjuvant chemotherapy were randomized to a supervised resistance and aerobic exercise intervention (OnTrack), a homebased physical activity program (OncoMove) or usual care. Data was re-analyzed using a Bayesian adaptive decision-theoretic approach incorporating interim analyses for trial continuation after every stage of 36 patients. Endpoint for the re-analyses was dose modifications (any versus none). We considered a symmetric setting aiming to identify the arm with the lowest proportion of patients requiring dose modifications, and an asymmetric setting aiming to identify exercise arms with an absolute risk reduction in dose modifications of ≥10% compared to usual care. Interim decisions for continuation were based on expected absolute increases in the probability of a correct decision in the next stage, with continuation thresholds of 1% and 0.1%. Settings with and without early dropping of arms were considered.
Results
Dose modifications occurred in 34% of patients in the usual care and OncoMove arms versus 12% in OnTrack. Frequentist analysis showed these proportions to be statistically significant (p=0.002). Using the 0.1% continuation threshold, the Bayesian adaptive decision-theoretic analyses required 72 patients to identify OnTrack as the most effective arm in the symmetric setting. In the asymmetric setting, 144 patients were required to identify OnTrack as the only exercise arm superior to usual care.
Conclusions
A Bayesian adaptive MAMS design for multi-arm trials that makes decisions on efficacy or futility based on the expected change in the posterior probabilities instead of the posterior probabilities itself substantially reduced the sample size of a three-arm exercise oncology trial. This design is worth considering for multi-arm trials.
Clinical trial identification
NTR2159.
Editorial acknowledgement
Legal entity responsible for the study
The authors.
Funding
The PACES trial was funded by the Dutch Cancer Society (grant ALPE 2009-4299). The development of the statistical methods was funded by the Dutch Cancer Society (grant KWF 2012-5711).
Disclosure
G.S. Sonke: Non-Financial Interests, Institutional, Advisory Board: Novartis; Financial Interests, Institutional, Research Grant: AstraZeneca, Merck Sharp & Dohme, Novartis, Roche, Agendia, Seagen. All other authors have declared no conflicts of interest.