Abstract 332: Simulation to Derive the Operating Characteristics of a Bayesian Adaptive Clinical Trial Design to Determine Optimal Duration of Hypothermia Following Out-of-Hosptial Cardiac Arrest
Introduction: Therapeutic hypothermia (TH) improves outcomes in comatose survivors of cardiac arrest, but optimal duration of cooling is unknown. An innovative adaptive trial may address this question efficiently and with fewer prior assumptions, but its complexity requires simulation to characterize the trial’s power and accuracy.
Objective: To evaluate the operating characteristics (OC) of an adaptive clinical trial designed to characterize the TH duration response relationships for patients with shockable and non-shockable rhythms.
Methods: The proposed ICECAP trial design uses a flexible, Bayesian normal dynamic linear model to estimate the relationship between duration of TH and 90 day modified Rankin Scale. The trial initially allocates 150 patients equally to 12, 24, or 48 hrs of cooling and models duration-response. The model then iteratively refines itself using response adaptive randomization ratios updated every 100 subsequent enrollments to time windows from 6 to 72 hours as needed to identify transitions in duration-response. Shockable and non-shockable rhythms are fitted separately. Sample sizes from 1600-2300 were simulated. The design was tested under hypothetical scenarios of no treatment effect, small, medium and large effects, and harm (outcomes worsen with duration). Each combination of sample size and scenario was simulated 1000 times.
Results: The proportion of trials correctly identifying the target duration (for both rhythm types) ranged from 72% (N=1600), to 81% (N=2300). A sample size of 1800 had 85% power to select the target (the true or larger adjacent) duration. OC remained robust in specific challenging scenarios including a very early plateau with similar efficacy at 18-72 hours, and an inverted U with efficacy upsloping to 18 hours and diminishing from 30-72 hours. In the null scenario, identical outcomes for all durations from 6-72 hours, the design only incorrectly identifies a winning dose in 4-7% of trials. It was also efficient in this scenario, terminating 35% of trials early for futility. In harm scenarios, trials terminate early 86% of the time.
Conclusions: The ICECAP trial design performs well under a variety of scenarios. Flexible, adaptive Bayesian trial designs are promising and ready to be implemented.
- © 2013 by American Heart Association, Inc.