| Bruce Stouch at Premier Research assesses the value of monitoring for futility in fixed and adaptive designs using conditional and predictive power
Statistical monitoring procedures are used in a broad range of clinical trials to ensure the early availability of efficacious treatments while at the same time preventing spurious early termination of trials for apparent benefit that may later diminish. When properly designed, these procedures can also provide support for stopping a trial early when results do not appear promising, conserving resources and affording patients the opportunity to pursue other treatment options and avoid regimens which may have known and unknown risks while offering little benefit. By weighing these considerations against each other in the specific study situation at hand, a satisfactory monitoring procedure can be chosen.
Formal statistical monitoring rules serve as guidelines for data monitoring committees to provide for possible early trial termination. The pre-specified monitoring rules allow for the possibility of early stopping in response to positive trends that are sufficiently strong to substantiate the treatment differences the clinical trial was designed to detect. At the same time, the monitoring boundaries must guard against prematurely terminating a trial on the basis of initial positive results that may not be maintained with additional follow-up. Premature termination may also be recommended if the current trends in the data indicate that eventual positive findings are highly unlikely. Early termination for negative results may be called for if the data collected to date are sufficient to rule out the possibility of improvements in efficacy that are large enough to be clinically relevant.
Alternatively, it may have become clear that study accrual, drug compliance, or follow-up compliance, among other factors have rendered the study incapable of discovering a difference, whether or not one exists. The focus of this article is to explore the construct of asymmetrical boundaries for interim monitoring of a fixed design and the design requirements for monitoring using a Bayesian approach. Additionally, the complement to monitoring using fixed boundaries by establishing optimistic and sceptical priors is also discussed relative to the complexity associated with modelling clinical assumptions and expectation. DESIGNING AN INTERIM MONITORING STRATEGY
In designing an interim monitoring strategy, it is preferable to include a plan for stopping in the face of negative results at the time the study protocol is developed. In particular, it is important to know what effect the plan will have on the power and significance level of the overall design. In calculating the asymmetric monitoring boundaries for a trial, many factors need to be estimated, and the expected results need to be modelled to ensure the boundary is clinically and ethically acceptable. Specifically, attrition, mortality (if appropriate), severity of the disease, the expected accrual rate, and the potential impact of the results from other concurrent studies, all need to be considered. |
