Background: There is increased interest in mechanical circulatory support devices (MCSDs), such as implantable left ventricular assist devices (LVADs), as "destination" therapy for patients with advanced heart failure. Because patient availability to evaluate these devices is limited and randomized trials have been slow in enrolling patients, a workshop was convened to consider designs for MCSD development including alternatives to randomized trials.
Methods and results: A workshop was jointly planned by the Heart Failure Society of America and the US Food and Drug Administration and was convened in March 2006. One of the panels was asked to review different designs for evaluating new MCSDs. Randomized trials have many advantages over studies with no controls or with nonrandomized concurrent or historical controls. These advantages include the elimination of bias in the assignment of treatments and the balancing, on average, of known and unknown baseline covariates that influence response. These advantages of randomization are particularly important for studies in which the treatments may not differ from one another by a large amount (eg, a head-to-head study of an approved LVAD with a new LVAD). However, researchers have found it difficult to recruit patients to randomized studies because the number of clinical sites that can carry out the studies is not large. Also, there is a reluctance to randomize patients when the control device is considered technologically inferior. Thus ways of improving the design of randomized trials were discussed, and the advantages and disadvantages of alternative designs were considered.
Conclusions: The panel concluded that designs should include a randomized component. Randomized designs might be improved by allowing the control device to be chosen before randomization, by first conducting smaller vanguard studies, and by allowing crossovers in trials with optimal medical management controls. With use of data from completed trials, other databases, and registries, alternative designs that include both a randomized component (eg, 2:1 allocation for new device versus control) and a nonrandomized component (eg, concurrent nonrandomized control, historical control, or a comprehensive cohort design) should be evaluated. This will require partnerships among academic, government, and industry scientists.