I just came across this talk and I think it is very important. It deals with the tendency for journals and researchers to publish only positive results while either ignoring or rejecting negative results. The problem exists not just in clinical medicine, but in every area of science. Science is a very competitive business and positive results often herald a "new discovery." By contrast, we tend to think that a negative result says very little, thus it's in scientists' own best interests to publish only positive results, assuming that the journal will even accept the negative ones.
From my own experience, I have seen scientists not only picking the best exemplar from a series of experiments and publishing only that one, but also essentially cherry-picking the data by using some creative averaging. I also know that I am not immune from this myself.