The dominant focus of precision or personalised cancer medicine has been on the discovery of predictive, companion diagnostics for the new era of targeted therapeutics. The central idea being that the pool of patients more likely to benefit could be enriched by identifying a tumoural biomarker that predicts response (or lack of response) ^1,2  . Recent, clinically important examples of these include mutant Kras / epidermal growth factor inhibitors; ALK gene rearrangements/crizotinib; ROS1 gene rearrangements/crizotinib; Met amplification/tivantinib. Nevertheless, these remain relative rarities as the vast majority of anti-cancer drugs do not have clinically validated predictive markers.

In clinical trial analysis, it is important to discriminate whether apparent differences between treatments might be due merely to random allocation of more of the good-prognosis patients to one treatment than to the other treatment. Obviously, anything known about the major determinants of prognosis can answer this question correctly, and determine whether, given the different numbers on each treatment in various prognostic categories, there is any residual relationship of treatment with survival.

 If biology is king, then it is at least logical to consider how molecular markers that describe the natural history of the cancer, and therefore prognosis, could be used to segregate patients according to risk of recurrence and for treatment to be tailored accordingly (perhaps delivering more intensive therapy poor prognostic patients), or to ensure balance  across both arms of a prospective, randomised trial. Significant progress has been made in the characterisation of prognostic biomarkers, through; the application of robust molecular methodologies; improved study design, especially with regard to statistical power; better access to well curated  biobanks, often linked to large clinical trial databases . These markers are now ready to be incorporated into clinical trial design to improve fidelity, relevance and efficiency.