Development and external validation of a new prognostic model for metastatic clear cell renal cell carcinoma: a preregistered analysis using data from randomised clinical trials in the checkpoint inhibitor era.

Abstract

Prognostic classification of patients with metastatic clear cell renal cell carcinoma (mccRCC) is instrumental for clinical trials and treatment decisions. Hence, optimal performance of prognostic models is crucial. This study aimed to develop and externally validate the new Clinical Prognostic Index in the Checkpoint Inhibitor era (CPI2) model using clinical trial data, and to externally validate the International Metastatic Renal Cell Carcinoma Database Consortium (IMDC) model. Individual patient data from the clinical trials CheckMate-214 and CheckMate-9ER were used for development and external validation of the CPI2 model, respectively. Both trials included previously untreated mccRCC patients of all IMDC prognostic groups and Karnofsky performance status of at least 70%, and randomly allocated them to receive either sunitinib or nivolumab plus ipilimumab (CheckMate-214) or nivolumab plus cabozantinib (CheckMate-9ER). Using a Cox proportional hazards model, predictors were selected from a list of patient, disease, and laboratory parameters using Akaike information criterion-based backwards selection. Bootstrapping was used to estimate coefficients and a shrinkage factor to correct for optimism. Calibration plots and Uno's C-index for discriminative accuracy were calculated for the CPI2 model and IMDC model in both trial populations. Data from 1096 patients included in CheckMate-214 between October, 2014, and February, 2016, and 651 patients included in CheckMate-9ER between September, 2017, and May, 2019, were used. Median follow-up was 40 months (IQR 13-94) in CheckMate-214 data and 38 months (15-53) in CheckMate-9ER data. In total, 665 (38%) of 1747 patients were aged 65 years or older; 1289 (74%) were male, and 458 (26%) were female. A prognostic model based on age, Karnofsky performance status, previous nephrectomy, metastatic locations (liver, lung, bone, and lymph node), and seven laboratory parameters (calcium, alkaline phosphatase, albumin, lactate dehydrogenase, absolute neutrophil count, lymphocyte count, and white blood cell count) was developed based on CheckMate-214 data. Classification into three prognostic groups provided discriminative accuracy at 36 months of 0·72 (95% CI 0·69-0·75) in CheckMate-214 and 0·72 (0·68-0·76) in CheckMate-9ER, which was 0·65 (0·62-0·68) and 0·61 (0·57-0·65) for the IMDC classification, respectively. Calibration showed adequate agreement between predicted and observed mortality risks. PD-L1 expression did not improve prognostic accuracy. CPI2 favourable, intermediate, and poor risk groups included 472 (43%), 366 (33%), and 258 (24%) of 1096 CheckMate-214 patients and 257 (39%), 196 (30%), and 198 (30%) of 651 CheckMate-9ER patients, respectively. The CPI2 model, using 14 readily available clinical parameters, demonstrated improvement in discriminative accuracy compared with the IMDC model. This classification should be further validated in other trial and observational populations and be used to reanalyse heterogeneity of treatment effects of immune checkpoint inhibitor-based combination regimens in trials shaping the current and future treatment landscape. Josephine Nefkens Foundation.