Clinical Trials




1. b

2. e, although strictly speaking even intention-to-treat analysis could contribute to this phenomenon, if some patients in the placebo group were accidentally treated with an effective experimental therapy.

3. b

4. A discussion might touch on control of several of the following features:


5. When one looks at many subgroups, the possibility of finding some apparent effects that are actually just chance fluctuations is substantially greater than when one looks at the group as a whole.  When one selects out a comparison that looks interesting and then performs a statistical test, the usual p-values and interpretation of statistical significance are no
longer valid, due to the "multiple comparison issue."  Further, the tendency of randomization to produce generally comparable treatment groups for the study as a whole depends on sufficient sample size and may not have been successful within the subgroup being studied.  Thus, the subgroup comparison may be based on non-comparable groups of patients.
The problem is even worse when, as sometimes is done, investigators simply isolate a group of subjects who did especially well on the experimental therapy and compare them to other subjects on the experimental therapy, without consideration of what happened to comparable control subjects.