A multi-step classifier addressing cohort heterogeneity improves performance of prognostic biomarkers in three cancer types

Cancer research continues to highlight the extensive genetic diversity that exists both between and within tumors. This intrinsic heterogeneity poses one of the central challenges to predicting patient clinical outcome and the personalization of treatments. Despite progress in some individual tumor types, it is not yet possible to prospectively, accurately classify patients by expected survival. One hypothesis proposed to explain this is that the prognostic classifiers developed to date are insufficiently sensitive and specific; however it is also possible that patients are not equally easy to classify by any given biomarker. We demonstrate in a cohort of 45 AJCC stage III melanoma patients that clinico-pathologic biomarkers can identify those patients that are most likely to be misclassified by a molecular biomarker. The process of modelling the classifiability of patients was then replicated in a cohort of 49 stage II breast cancer patients and 53 stage III colon cancer patients. A multi-step procedure incorporating this information not only improved classification accuracy but also indicated the specific clinical attributes that had made classification problematic in each cohort. These findings show that, even when cohorts are of moderate size, including features that explain the patient-specific performance of a prognostic biomarker in a classification framework can improve the modelling and estimation of survival.