|
Abstract:
|
In formulating the monitoring plan for a long term randomized trial, one is faced with a wide array of possible monitoring boundaries. While many boundaries are constructed to satisfy a mathematical optimality criterion such as small average sample number, most boundaries are chosen because they subjectively balance a set of competing interests; ethical, practical and economic. Many potential features are of interest. Boundaries maybe one or two sided. If two sided, they may be symmetric or asymmetric. One may take a flexible, alpha spending approach or a use a more traditional approach such as Haybittle-Peto. Boundaries based on simultaneous alpha/beta spending rates are also available, automatically taking futility into account. Finally, standard computational techniques are available which allow commonly used boundaries to be employed for a wide variety of statistical tests.
|
