I should clarify, PGS, because I partially misread your post, that the first step that Higano did in the Cox analysis of 9902A she presented at ECCO 2005 was to take the exact same five stat sig final covariates from 9901's Cox regression and apply them to 9902A. There is no question about this. It is the unequivocal truth. The resulting p value was improved from 0.33 to 0.023. In addition to your references in the paragraph below, Higano also did what io_io said and what I said in my quotes in the last paragraph below.
-----------------------------------------------------------------------------------------------------------
PGS: ...The backwards and forwards / any which way reference by Gold was referring to the fact that if they took the 5 distilled covariates from 01 and applied them to the 01 dataset, then the p value was stat sig; or if they took the 5 distilled covariates from 01 and applied them to the combined 01+02A dataset, then it was stat sig again.>>
Walldiver: Correct, they ran the equations both of the ways you cite above.
PGS: Although DNDN may have started with the 20 prognostic factors and used the data from 02A to distill them, they did not present that data to the public (that I'm aware of). They only presented this outcome for 01, not 02A.>>
Walldiver: Correct, they did not present the 20 prog factor data to the public. However, I am taking DNDN mgmt at their word when they said that they also performed the Cox regression for 9902A by beginning with the same 20 prognostic factors as 9901, then running through the same steps in the exercise as they did in 9901, and then arriving finally with a stat sig p value. I think our differences over this issue are rooted in how much each of us believes in the veracity of DNDN mgmt.
News 
Market Data 
Discover 
AI
