Development and validation of a mutivariate model combining cell cycle progression score with CAPRA to predict prostate cancer mortality in a conservatively managed cohort.

Abstract

67 Background: Prostate cancer outcomes are variable and difficult to predict. Improved tools are needed to appropriately match treatment to a patient’s risk of progression. We developed and validated a multivariate model to predict disease−specific mortality (DSM) by combining clinical parameters (CAPRA score) with a score based on measuring the expression level of cell cycle progression (CCP) genes. Methods: A multivariate prediction model was trained using patients from 4 retrospective cohorts with median clinical follow up of 7.6 years. We used 200 men from the UK diagnosed after TURP, 353 from Scott & White and 388 from UCSF treated with radical prostatectomy, and 118 men from Durham VA treated with EBRT. CCP score was derived from fixed tumor tissue (biopsy or surgical resection). Outcome was either time from treatment to biochemical recurrence (US cohorts) or time from diagnosis to disease specific mortality (UK cohort). The model was validated for predicting time from diagnosis to DSM in 180 men from the UK diagnosed by needle biopsy with clinically localized prostate cancer and managed conservatively (mean/median CAPRA score = 6). Results: A model combining CAPRA with CCP score was fit in the training set by a Cox Proportional Hazards analysis stratified by cohort. The Combined score was defined as 0.39*CAPRA+0.57*CCP score. There were no significant interactions between cohort and CAPRA or CCP score. This suggests that both CCP score and CAPRA confer similar prognostic information regardless of cohort composition, treatment, or specific outcome. In the validation cohort the Combined score was highly prognostic (HR= 2.27, 95%CI: (1.63, 3.16), p = 1.2 x 10−7). By likelihood ratio testing, the Combined score was a better predictor of DSM than CAPRA alone (p = 0.0028). The c−index of the Combined score was 0.75, which was an improvement over CAPRA (c−index 0.71). Conclusions: This multivariate model predicts DSM in a conservatively treated cohort. The model provides prognostic information beyond clinical variables, and can be used to help differentiate aggressive from indolent cancer at diagnosis.