DATA-DERIVED ASSOCIATIONS ARE NOT TESTS OF THESE ASSOCIATIONS

[摘要] ...It is important to realize that comparisons of any subgroups within a randomized trial that are defined after randomization are no longer randomized comparisons. At best, such comparisons would have the characteristics of an observational study rather than a randomized trial, with all the attendant concerns about the control of confounding. At worst, analyses of subgroups formed after randomization may represent observation of data-derived relations, which are useful only to formulate, not test, research questions.

[学科分类] 儿科学