I’ve been covering this subject for the Journal of the National Cancer Institute (subscription required). As I noted in a recent article:
Though hundreds of studies have appeared in the medical literature making claims for the predictive power of multi-gene signatures of different cancers, few have been deemed ready for prime time. A review last year of 16 studies of multiple gene-signatures in non-small cell lung cancer found “little evidence that any of the reported gene expression signatures are ready for clinical application.”
Reduced to its essence, here’s the problem. These tests are based on the measurement of the under- or over-expression of dozens of genes. Those results are then subjected to a complex algorithm that is based on a series of “weights” given the expression levels of each of the genes in the targeted panel. Those weights are added up for a composite score that determines whether you have of a particular form of a disease, and whether it is more or less likely to be affected by a particular drug. In other words, this extraordinarily complex test based on dozens of unvalidated, epidemiological observations leads to a clinical decision that is binary. Either you have or don’t have that kind of cancer; either you will or will not benefit from that drug.