On Mon, Aug 16, 2010 at 11:42 AM, G Money <gm0n3...@gmail.com> wrote: > > I don't necessarily have a problem with that...let the companies spend the > time and money to do the research, and then let the FDA ensure that > everything is on the up and up.
Personally, I think that all data related to a clinical trial should be submitted to the FDA. Selective disclosure of data is an invitation to gloss over the ugly bits but the ugly bits are what ends up killing people. I think that good researchers with honest intentions can still feel the push to say, "well, I think this one experiment was just done wrong and we didn't see it pop up in any of the other studies, so we'll toss this one as an outlier". Companies should be still be able to make their case to the FDA about which studies are more important, useful, etc and why. But I think the FDA should have all the data. > I'd be curious how, if...and this might be a big IF...if all 10 trials were > on the up and up, why 9 would fail, and then suddenly 1 would succeed???? > Shouldn't they either all corroborate, or all refute....if all proper > variables are accounted for??? Several reasons. One is that study design (for both good and bad reasons) often changes from one study to another. These changes can be very small and largely irrelevant or they can be much more substantial. One of the most frequent things that happens in a study is inconclusive results. You think you see a trend one way or another but it isn't statistically significant one way or another by the standard definitions. Yet, as a researcher, you have a feeling that there is something going on there and you just need to tweak things to try and bring it out. It might be that your level of detail was not matched to what you needed to see an effect or maybe you think that the protocol wasn't followed 100% and you need to tighten up the protocol. It could be lots of things. So you tweak the study and do it over again. That's the right thing to do but when those studies never see the light of day, no one knows that there were a whole lot of "didn't show any statistical difference" studies. The other fundamental reason is basic statistics. When you run a study there is a computed margin of error. That margin of error (known as P value) tests against the null hypothesis. The null hypothesis is that there is nothing other than randomness going on in a system, that what you did has no directional effect. So if you run a test and get back a P-value of 0.05, that means that there is a 5% chance that the results you are seeing were based off of random variation instead of some inherent effect that the thing you are testing for had. As a result, if you run the same experiment again and again and again, you are likely to see the results you want on 1 of the tests even if you don't see them on most of the tests just because of inherent statistical noise. Hope that helps, Judah ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~| Order the Adobe Coldfusion Anthology now! http://www.amazon.com/Adobe-Coldfusion-Anthology-Michael-Dinowitz/dp/1430272155/?tag=houseoffusion Archive: http://www.houseoffusion.com/groups/cf-community/message.cfm/messageid:325121 Subscription: http://www.houseoffusion.com/groups/cf-community/subscribe.cfm Unsubscribe: http://www.houseoffusion.com/groups/cf-community/unsubscribe.cfm