>>>In the case of Interferon and Peg-Interferon, the INterferon was getting about a 50% success rate. Sizing their trial to prove a 10% difference meant that they wanted to be sure that if their sample showed 55% or greater (10 percent greater than 50%) then that difference is significant statistically. They achieved 56-58%. <<<
blandis,
I think in the case you cite , the interpretation of the 10% difference could be just as you describe , and that what we're talking about here is a difference in common language , not in statistics.
>>> You need much larger samples for the case of 50/60 than you do for 2/12. I don't think you need a sample anywhere near 250 to show 12 is different from 2(so long as you aren't assuming the statistical variance of the data to be very large) <<<
I don't disagree that a 10% difference is easier to demonstrate as the comparator value gets closer to zero but as you change what you say to 15 vs. 5 or 20 vs. 10 , which may be numbers that SCLN management had to consider as possible , then I think the trial was probably accurately powered. I do think it's way off base to say that this trial could demonstrate with statistical significance the difference between 5 and 5.5% or for that matter 1% and 1.1% (both 10% differences , by your preferred definition ).The reason is that the variances are significant . This is not like fipping a coin and measuring heads and tails. There is variance in the measurement of viral load , there is probably a very considerable variance in the histology readings as they are done by humans , not machines , and there are between-site variances in the data that may be considered in the analysis.
Just take a look at the patient numbers you get when you multiply those numbers , say 1% vs. 1.1% , by a trial arm size of 250 , and you'll realize that somebody will have to be dissected to provide the extra .25 of a patient that would be expected to get an SVR. At some point common sense tells you what a statistical calculator does not.