RE: rjc - trial error or statistical anomoley
I agree things do not make sense - for many reasons. As you stated, if the results were erroneous, then it could be a computation error, an interpretation error or a scoring error (at either the beginning or the end of the 90 day period.) It is hard for me to say because it would take a serious error to come out with a placebo effect of that magnitude.
We do not normally see this degree of recovery from stroke with no treatment. The reason an improvement in the NIHSS score of 4 was chosen is for that reason - a score of 4 is statistically meaningful. The relevance of the improvement score of 4 points is based on the previous science which has supported the licensing of marketed drugs.
If it was typical for stroke patients to spontaneously recover to this degree without any medical intervention, there would be no need for treatment drugs