Abstract:
|
Prospectively controlling selection bias with randomization is limited because of concern about accidental bias. The success of the compromise, the randomized block design, is unknown because, for its detection, the Berger-Exner test of 18 years ago has not yet yielded any published data. No consideration appears to have been given to application of that test prospectively. Theoretically, it could be used prospectively by either randomized block or minimization designs but both present problems and are too limited. These limitations can be overcome by combining elements of the two designs, changing minimization from a preset to a flexible design, and changing randomization from directly assigning patients about half the time to indirectly using it for every assignment. Combined in this way, as Flexible Minimization, it has several advantages because prevention of selection and accidental bias become complimentary rather than antagonistic. Many supporting actions in the four phases of a clinical trial can be taken to improve their performance of finding and confirming predictors of outcome responses. Hopefully, it will also make statistical advice more consistent.
|