iwfal........ Coming back to this:
<<<<"Significant of each individual prognostic variable considered in a model along with treatment arm.">>>>
When we discussed this before, you described (correctly) that the filtering process was not one of univariate analysis, and sure enough thats how the Dr Small slide describes how the variable is filtered.
Recall we were discussing how prognosticators might or might not have emerged from 9902a had the process started solely from its data set:
a) Take for example the clearly imbalanced factors PSA and ALK (PSA was one among the 9901 "five", and did have a HR favoring survival in the final 9902a analysis, but that is not really of concern to my point);
b) Assume they are prognosticators in this data set, favoring susvival at least to some extent.
c) Consider them by stratifying for treatment arm alone.
I expect it would be difficult for these two (PDA and ALK in turn) to advance to the multivariate model because:
1. The first reason is intuitive only - but given that Provenge has a muffled HR of 1.3 it seems unlikely that this "stratification" could redress the variable so profoundly as to yield significance. Sorry no proof, and perhaps intertwined with my next point.
2. The second reason is far simpler - the strength of the bone mets variable, with its huge HR of 3.0 or 4.0, remains as confounding noise in any analysis that does not filter for it.
Reason to believe that DNDN can argue, and the FDA might agree at least to some extent, that this CR methodology, based solely on 9902a data, would fail to address all of the known imbalances ........ I think so.
"....on the biotech battle-field, you need some élan...."
