External validation of a prediction model of survival after cytoreductive nephrectomy (CN) for metastatic renal cell carcinoma (mRCC).

Abstract

556 Background: In 2013 a model to predict cancer-specific survival (CSS) after CN was published (Eur Urol 63:947-52, 2013), consisting of a preoperative nomogram to predict 6-month (mo) probability of death (PoD) based on serum albumin and serum lactate dehydrogenase and a postoperative nomogram (12-mo PoD) to which were added pathologic primary tumour stage, nodal stage and intraoperative blood transfusion. We externally validated the model with a contemporary European dataset of patients treated in the VEGF-targeted therapy era. Methods: After having received the original calibration indices, data from 205 mRCC patients who underwent CN between 2006 and 2013 at 3 European centres in Amsterdam, Edinburgh and Munich were used for external validation. Next to the variables for the nomogram, age, MSKCC, IMDC, number and metastatic sites and subtype were collected. Both nomograms were used to obtain predictions for the subjects. Based on the quantiles of the distribution of predictions subjects were grouped into 4 categories and Kaplan-Meier estimates calculated as were calibration plots of observed versus predicted PoD. Since all subjects had follow-up for > 1.25 years CSS at 6 and 12 mo as a dichotomous outcome and the ROC curves were plotted. Results: With a median follow-up of 29 mo and a median OS of 15 (95% CI 12-22) mo for the entire cohort, 32 and 82 patients died from RCC at 6 and 12 mo after CN respectively. For the pre- and postoperative nomogram the respective ROC curve had an AUC of 0.667 and 0.712. In the preoperative calibration plot the observed PoD at 6 mo was higher than the predicted, resulting in a lower discrimination when compared to the original validation (0.76 and 0.74 for pre- and postoperative nomograms respectively). Conclusions: In external validation the pre- and postoperative nomograms perform well in patients with primary mRCC. Their discrimination is in the reported range of MSKCC and IMDC risk scores, which were not developed to address the decision of CN (concordance index 0.657 and 0.71, Lancet Oncol 14:141-8, 2013). Although discrimination of the preoperative nomogram was lower in external validation, it retains ability to predict early death after CN.