CPTAC Consensus Subtypes Add No Incremental Prognostic Value Beyond Stage and Grade in Clear Cell Renal Cell Carcinoma: A Proteomic Reanalysis

Molecular subtypes from multi-omic profiling have refined the biological characterization of clear cell renal cell carcinoma (ccRCC), but whether they add prognostic information to clinicopathological variables already in use is unclear. We evaluated the incremental prognostic value of proteomically assigned CPTAC consensus subtypes (NMF1–NMF3) against stage and grade, and asked whether subtype identity is carried by a biologically specific set of proteins. Subtypes were predicted from tumor proteomes in 211 patients under nested cross-validation and a locked hold-out design, and the analysis was repeated using methylation subtypes defined without proteomic input. A transcriptome classifier was transferred to 533 TCGA-KIRC tumors, and the transferred subtypes entered nested Cox models containing age, stage and Fuhrman grade. A curated set of 139 metabolic enzymes was compared against 500 random protein sets of the same size and against random sets additionally matched on abundance, missingness, variance and co-regulation. Classification reached a macro-average AUROC of 0.901 (95% CI 0.871–0.933) under cross-validation and 0.862 in the untouched expansion cohort; methylation subtypes were classified with comparable accuracy (0.905). Subtype proportions in TCGA-KIRC differed from CPTAC by at most 2.5 percentage points. Among 518 patients with complete clinical data, grade improved a model containing age and stage (p = 0.001), whereas subtype added nothing once grade was included (p = 0.703; change in C-index +0.0019, 95% CI −0.0023 to +0.0116); with 171 deaths, the study could detect a hazard ratio of 1.64 at 80% power, so small effects remain possible. The metabolic panel did not outperform size-matched (p = 0.649) or covariate-matched random sets (p = 0.762), whereas a housekeeping set performed significantly worse. CPTAC consensus subtype identity is highly predictable from the proteome but redundantly encoded rather than concentrated in a functional class, and adds no measurable prognostic information beyond stage and grade in this cohort. High classification accuracy for a molecular label is not evidence of clinical utility; these conclusions apply to the CPTAC NMF scheme as evaluated here and should not be extended to proteomic subtyping in general.