Head-to-head comparison of the most relevant integrated prognostic systems predicting cancer-specific survival in clear cell renal cell carcinoma

Ficarra V.1, Sun M.2, Karakiewicz P.I.2, Novara G.1, Antonelli A.3, Bertini R.4, Carini M.5, Carmignani G.6, Longo N.7, Martignoni G.8, Martorana G.9, Minervini A.5, Mirone V.7, Artibani W.8, Zattoni F.1, Simionato A.6, Siracusano S.10, Terrone C.11 1University of Padua, Dept. of Urology, Padua, Italy, 2University of Montreal, Dept. of Urology, Montreal, Canada, 3University of Brescia, Dept. of Urology, Brescia, Italy, 4Vita-Salute University San Raffaele, Dept. of Urology, Milan, Italy, 5University of Florence, Dept. of Urology, Florence, Italy, 6University of Genova, Dept. of Urology, Genova, Italy, 7University Federico II, Dept. of Urology, Naples, Italy, 8University of Verona, Dept. of Urology, Verona, Italy, 9University of Bologna, Dept. of Urology, Bologna, Italy, 10University of Trieste, Dept. of Urology, Trieste, Italy, 11University of Eastern Piedmont, Dept. of Urology, Novara, Italy Introduction & Objectives: To compare the performances of UISS, SSIGN score, and Karakiewicz nomogram in a large multi-institutional series of patients with clear cell RCC. Materials & Methods: We collected retrospectively the preoperative, pathological and follow-up data of 1871 patients treated in 16 academic centers. The predictive accuracy of the three prognostic models for prediction of CSS was quantified according to Harrell’s concordance index, whereas differences were estimated using the DeLong test. Decision curve analyses were used to determine the optimal benefit derived from the use of the 3 models. Results: At a median follow-up of 40 months, 272 (15%) had died of disease. Overall 3- and 5-year CSS estimates were 87.5% and 83.1%, respectively. The predictive accuracy estimates for prediction of CSS at 3 and 5 years were 88.9% and 88.8% for the Karakiewicz nomogram; 85.0%, and 84.1% for the UISS; and 87.9% and 82.5% for the SSIGN score, respectively. Most of the differences in predictive accuracy were statistically significant. Calibration plots demonstrated substantial departures from ideal predictions for all the model. The Karakiewicz nomogram demonstrated the highest net benefit up to threshold probability of 25% and 37% at 3 and 5 yrs, respectively. Following those thresholds, the UISS resulted with the highest net benefits. Conclusions: All three models demonstrated substantially sensible predictive accuracy, but all tend to underestimate the CSS probabilities. The Karakiewicz nomogram demonstrated the highest net benefit up to threshold probabilities of 25% at 3 yr and 37% at 5 yr, whereas, the UISS resulted with the highest netbenefit in patients with higher risks of deaths.