Stratification of high-risk prostate cancer into prognostic categories: a European multi-institutional study

BACKGROUND: High-risk prostate cancer (PCa) is an extremely heterogeneous disease. A clear definition of prognostic subgroups is mandatory.

OBJECTIVE: To develop a pretreatment prognostic model for PCa-specific survival (PCSS) in high-risk PCa based on combinations of unfavorable risk factors.

DESIGN, SETTING, AND PARTICIPANTS: We conducted a retrospective multicenter cohort study including 1360 consecutive patients with high-risk PCa treated at eight European high-volume centers.

INTERVENTION: Retropubic radical prostatectomy with pelvic lymphadenectomy.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS: Two Cox multivariable regression models were constructed to predict PCSS as a function of dichotomization of clinical stage (< cT3 vs cT3-4), Gleason score (GS) (2-7 vs 8-10), and prostate-specific antigen (PSA; ≤ 20 ng/ml vs > 20 ng/ml). The first "extended" model includes all seven possible combinations; the second "simplified" model includes three subgroups: a good prognosis subgroup (one single high-risk factor); an intermediate prognosis subgroup (PSA >20 ng/ml and stage cT3-4); and a poor prognosis subgroup (GS 8-10 in combination with at least one other high-risk factor). The predictive accuracy of the models was summarized and compared. Survival estimates and clinical and pathologic outcomes were compared between the three subgroups.

RESULTS AND LIMITATIONS: The simplified model yielded an R(2) of 33% with a 5-yr area under the curve (AUC) of 0.70 with no significant loss of predictive accuracy compared with the extended model (R(2): 34%; AUC: 0.71). The 5- and 10-yr PCSS rates were 98.7% and 95.4%, 96.5% and 88.3%, 88.8% and 79.7%, for the good, intermediate, and poor prognosis subgroups, respectively (p = 0.0003). Overall survival, clinical progression-free survival, and histopathologic outcomes significantly worsened in a stepwise fashion from the good to the poor prognosis subgroups. Limitations of the study are the retrospective design and the long study period.

CONCLUSIONS: This study presents an intuitive and easy-to-use stratification of high-risk PCa into three prognostic subgroups. The model is useful for counseling and decision making in the pretreatment setting.