Background: High-throughput analyses yielded a large number of predictive biomarkers in prostatic cancer (PCa) patients. Combinations of these biomarkers and with clinical features could improve on prediction.
Materials and methods: Tissue microarrays (640 patients) with triplicate cores of non-neoplastic prostate, benign prostatic hyperplasia (BPH), and index tumor were immunostained with antibodies to numerous biomarkers, digitized, and quantified. We used tree-based classification algorithms to stratify patients into 3 risk strata on the basis of their clinical and pathologic data. Markers were tested for prognostic ability in each stratum (stratum 1 had <10% risk of recurrence; stratum 3 had >60% likelihood of recurrence over a period >12 years). Sub stratification of the clinico-pathologic strata was also pursued.
Results: We identified a number of significant predictors for PSA recurrence free survival, which were used to construct a predictive model that combines clinical and biomarker data. In the low-risk clinico-pathologic stratum, the markers were predominantly related to non-neoplastic tissues, in the moderate-risk stratum to stromal-epithelial interactions and angiogenesis, while those in the high-risk stratum were mostly oncogenes. Substratification of the intermediate risk group using stromal quantitation and proliferative index successfully, up or down, staged the risk strata for most patients.
Conclusions: The fact that different biomarkers are most predictive of disease recurrence within different risk subgroups suggests an association between biological processes and prognostic ability. This is the first time that subgroup analysis of markers finds that prognostic ability is associated with biological processes and is proof of concept that distinct phenotypes are associated with risk of recurrence in different types of cancer.