Oncology - Bladder, Renal, Test
Mid-Atlantic Section 76th Annual Meeting
Introduction & Objective :
With the rise of nephron-sparing management for renal cell carcinoma (RCC), quality of life (QOL) metrics may provide prognostic value above and beyond traditional demographic and disease parameters. We evaluate the utility of self-reported QOL results in predicting mortality among RCC patients and test the findings in a prospectively-maintained external database.
Predictive variables were predefined and analyzed using the Surveillance, Epidemiology, and End Results – Medicare Health Outcomes Survey (SEER-MHOS) database. QOL metrics were comprised of mental component summary (MCS) and physical component summary (PCS) scores. For each multivariable Cox proportional hazards regression, the Harrell’s concordance statistic (C-index) and Akaike Information Criteria (AIC) were calculated to determine predictive accuracy and parsimony, respectively. A lower AIC indicates a more parsimonious model. Findings from the SEER-MHOS database were tested in the prospectively-maintained Delayed Intervention and Surveillance for Small Renal Masses (DISSRM) database.
In SEER-MHOS, 1494 patients with a median age of 73.4 years and follow-up time of 5.6 years were included. There were 747 deaths, 139 of which were due to RCC. Cox regression demonstrated that each additional MCS and PCS point reduced the hazard of all-cause mortality by 1.3% (95% CI 0.981-0.993, P<0.001) and 2.3% (95% CI 0.971-0.984, P<0.001), respectively. Regression models with QOL metrics demonstrated higher predictive accuracy (C-index 72.3% vs 70.1%) and parsimony (AIC 9376.5 vs 9454.5) than models without QOL metrics. In DISSRM, 479 patients with a median age of 65.3 years and follow-up time of 3.9 years were included. There were 49 deaths, 2 of which were due to RCC. In agreement with the SEER-MHOS analysis, regression models including QOL metrics demonstrated maximum predictive ability (C-index 77.8% vs 74.1%) and parsimony (AIC 494.9 vs 496.4) compared to those without QOL metrics. Further testing demonstrated that the single best question producing maximum predictive ability (C-index 76.9%) and parsimony (AIC 335.2) was one of physical functioning limitations in the context of “moderate activities such as moving a table, pushing a vacuum cleaner, bowling, or playing golf.”
Models with self-reported QOL metrics predict all-cause mortality in RCC patients with higher accuracy and parsimony than those without QOL metrics in two separate database tests. Physical health in particular was a stronger predictor of mortality than mental health.