Allow me to explain further: two set of values compares two different populations, one is way sicker than the other and should not be compare. However, if we only compare the worst patients in one group, and the better of the other, we create a third group of patient with similar disease severity (but not the same). So in order for the comparison to be as fair as possible, I wonder if there is a statistical method to compensate for the handicap that the worst group has. This is relevant because it is well known that the characteristics at baseline are the most relevant factors that predicts the result.