Presentation Authors: Elio Mazzone*, Giorgio Gandaglia, Milan, Italy, Sophie Knipper, Markus Graefen, Derya Tilki, Hamburg, Germany, Giuseppe Rosiello, Giuseppe Fallara, Marco Bandini, Armando Stabile, Carlo Bravi, Paolo Dell'Oglio, Nicola Fossati, Francesco Montorsi, Alberto Briganti, Milan, Italy
Introduction: Most of the preoperative risk tools predicting prostate cancer (PCa) recurrence after radical prostatectomy (RP) considered biochemical recurrence (BCR) as the main endpoint. However, not all patients experiencing BCR ultimately progress to clinical recurrence (CR). We aimed at developing and externally validate a novel risk classification predicting long-term CR in European patients treated with RP.
Methods: Overall, 19,825 PCa patients treated with RP between 1991 and 2018 at two European referral centers were identified. The study outcome was 8-year CR defined as positive imaging after the onset of BCR. Kaplan-Meier analysis assessed time to CR. A multivariable Cox regression model-derived nomogram predicting 8-year CR was developed in one institution cohort and externally validated in the second institution cohort. Predictors were: PSA level at RP ( < 10 vs. 10-20 vs. >20), biopsy grade group (1 vs. 2 vs. 3 vs. 4-5), clinical stage (T1 vs T2 vs. T3) and percentage of positive cores at biopsy ( < 33% vs. 34-66% vs. >66%). The accuracy of our model was compared to the D&[prime]Amico classification and CAPRA score in predicting 8-year CR. Finally, nomogram-derived coefficients were used to develop a novel risk score.
Results: Overall, 6640 (33%), 8823 (44%) and 4362 (23%) had low, intermediate and high-risk disease according to the D&[prime]Amico classification. The 8-year overall CR-free survival rate was 92.9%. No significant differences were observed in the 8-year CR-free survival rates between D&[prime]Amico low- and intermediate-risk patients (98.9 vs. 94.1%; P=0.1). The 8-year CR-free survival rate was 78% for D&[prime]Amico high-risk patients. The developed model predicting CR depicted excellent discrimination at external validation (area under the curve [AUC] 89%). After testing the accuracy of the D&[prime]Amico risk groups and the CAPRA score in our cohort, the novel model showed higher AUC (89% vs. 76% vs. 80%). Finally, we assigned a specific coefficient to each covariate of the model and the sum of the derived score identified three risk groups: low (0), intermediate (1-2) and high (>2). The 8-year CR-free survival was 99%, 85%, and 62% for patients in the novel low-, intermediate- and high-risk group, respectively (P=0.01).
Conclusions: We developed and externally validated a novel preoperative risk tool to predict long-term CR after RP. Our model exhibited higher accuracy as compared to available tools in the prediction of stronger oncologic endpoints at long-term follow-up. This stratification might assist physicians in preoperative counselling.