Efficiency of Existing Prognostic Models in Renal Cell Carcinoma
Posted: Monday, August 12, 2019
The predictive ability of prognostic models for renal cell carcinoma seemed to decrease when prospective trial data were used compared with retrospective validations, according to a trial-based validation study by Andres F. Correa, MD, of Fox Chase Cancer Center, Philadelphia, and colleagues from various other institutions. Their results, which were published in the Journal of Clinical Oncology, suggest that prospective validation of any predictive model is needed before it is implemented into clinical practice or clinical trial design.
“This study provides the highest level of validation to date for the most commonly used renal cell carcinoma prediction models, some currently used in the design of costly adjuvant clinical trials,” the investigators commented.
To validate the models, a total of eight recurrence models were chosen based on their use in clinical practice. They were affirmed with data from 1,647 patients with resected localized high-grade or locally advanced disease from the ASSURE (ECOG-ACRIN E2805) adjuvant trial. The investigators looked at the discriminatory and calibration capacity to understand the models’ predictive performance.
When compared with the original results, prospective validation of the prognostic models showed diminished predictive ability. Based on their findings, the SSIGN (Stage, Size, Grade, and Necrosis) model performed the best (C index = 0.688), and the UISS (University of California at Los Angeles Integrated Staging System) model performed the worst (C index = 0.556). In addition, most models only “marginally outperformed” standard staging when compared with the 2002 TNM stage system (C index = 0.60). The models were most efficient within the first 2 years after diagnosis, and their predictive ability varied significantly over time.
Disclosure: The study authors’ disclosure information may be found at ascopubs.org.