[Lani]

NO, you don't. Most series just have tumors divided by their ER, PR and her2 by Immunohistochemistry staining (or by ER and PR by IHC and her2 by FISH)

Here they divided the tumors by their genetic components by analyzing their "genetic fingerprint"

ER by IHC can be inaccurate due to residual dye from Sentinel lymph node locating, ER and PR and her2 by IHC can be inaccurate because the sample was not placed in formalin and fixed until later (see a recent one of my posts)

Going back to the gene rather than testing for the protein (the final product of gene DNA to messenger RNA to constructed protein, which may be modified) is going back to the most exact identification of what makes the tumor the way it is by the most exact testing.

It is as if they sent out partially blind people to do the census and they came back with their best bet of the characteristics of the citizens they spoke with, rather than a swab from the
citizen's cheeks which would have proven who was male vs female, carbon 14 testing of their age (rather than their stated age-- many "fib a bit'), hair samples to test roots (similarly true hair color vs that which is perceived or reported)...

So it is not surprising that the results on prognosis are different when one is looking at apples and oranges vs smoothie components.

I am a bit doubtful that these 10 groups are the end of the story... but none can blame them from trying to find groups of patients who can be treated similarly as treating each tumor individually will be very expensive for society.