It would be beneficial to find panels of markers that can predict the likelihood of cancer recurrence in various populations. By testing the gene expression markers of a patient, oncologists can identify those patients unlikely to benefit from adjuvant chemotherapy from those that would. Whether a patient would benefit from adjuvant therapy depends of two things: whether the tumor if "destined" to come back in the first place and whether the tumor is sensitive to drugs which might be used to keep it from coming back.
The gene expression markers (assays) actually can be calibrated to provide information both about the possibility of recurrence and also chemosensitivity. The problem is dissecting one from the other. Studies to date have just looked at whether people had a recurrence. This study seems to look at chemosensitivity.
You can identify gene expression patterns which correlate with this. But it can be hard and even impossible to tell what exactly you are measuring: is it intrinsic aggressiveness of the tumor? Sensitivity to Taxol? Sensitivity to Avastin? You find a gene expression panel which correlates with something, but picking apart the pieces is hard.
Sure, we know that ER predicts for hormone response, Her2/neu gene amplification predicts for response to Herceptin, and certain mutations predict for response to Iressa and Tarceva, but that is far from proving that profiling gene mutations will work in a more general setting.
You could begin to do this if you combine gene expression studies with cell culture studies. Use the cell culture to define the difference between sensitivity and resistance. Then see which pattern correlates with which for individual tumors and individual drugs. It can theoretically be done (and certainly will be done, over time), but it's not easy.
Then you come to the 1,000 pound gorilla of a question: What effect will the different individual drugs have in combination in different, individual tumors? This is where cell culture assays will be able to provide uniquely valuable information. But it's not one versus the other. The best thing is to combine these different tests in ways which make the most sense. We cannot afford too much trial and error treatment.