BACKGROUND

Risk stratification for localized renal cell carcinoma (RCC) relies heavily on retrospective models, limiting their generalizability to contemporary cohorts.

OBJECTIVE

To introduce a contemporary RCC prognostic model, developed using prospective, highly annotated data from a phase III adjuvant trial.

DESIGN, SETTING, AND PARTICIPANTS: The model utilizes outcome data from the ECOG-ACRIN 2805 (ASSURE) RCC trial.

OUTCOME MEASUREMENTS AND STATISTICAL ANALYSIS

The primary outcome for the model is disease-free survival (DFS), with overall survival (OS) and early disease progression (EDP) as secondary outcomes. Model performance was assessed using discrimination and calibration tests.

RESULTS AND LIMITATIONS

A total of 1735 patients were included in the analysis, with 887 DFS events occurring over a median follow-up of 9.6 yr. Five common tumor variables (histology, size, grade, tumor necrosis, and nodal involvement) were included in each model. Tumor histology was the single most powerful predictor for each model outcome. The C-statistics at 1 yr were 78.4% and 81.9% for DFS and OS, respectively. Degradation of the DFS, DFS validation set, and OS model's discriminatory ability was seen over time, with a global c-index of 68.0% (95% confidence interval or CI [65.5, 70.4]), 68.6% [65.1%, 72.2%], and 69.4% (95% CI [66.9%, 71.9%], respectively. The EDP model had a c-index of 75.1% (95% CI [71.3, 79.0]).

CONCLUSIONS

We introduce a contemporary RCC recurrence model built and internally validated using prospective and highly annotated data from a clinical trial. Performance characteristics of the current model exceed available prognostic models with the added benefit of being histology inclusive and TNM agnostic.

PATIENT SUMMARY

Important decisions, including treatment protocols, clinical trial eligibility, and life planning, rest on our ability to predict cancer outcomes accurately. Here, we introduce a contemporary renal cell carcinoma prognostic model leveraging high-quality data from a clinical trial. The current model predicts three outcome measures commonly utilized in clinical practice and exceeds the predictive ability of available prognostic models.