Competing mortality in advanced head and neck cancer: A population-based study.

Abstract

5578 Background: Competing (noncancer) mortality (CM) is a common event in head/neck cancer (HNC) that complicates the interpretation of treatment effects. Population-based models are needed to estimate the impact of CM on power of HNC clinical trials. Methods: We analyzed 22,729 patients diagnosed with advanced nonmetastatic HNC between 1993 and 2004 using SEER data. We identified factors associated with cancer-specific mortality (CSM) and CM using the Fine-Gray proportional hazards model with backward stepwise regression. Risk scores for CM were derived from coefficient estimates and values of covariates retained in the model. The model was validated using bootstrap and split sample techniques. Power estimates assumed exponential hazards, 500 patients accrued over 3 years, 2 years follow-up, hazard ratio (HR) for all-cause mortality (ACM) of 0.67, and primary endpoint (PE) evaluation at 5 years. The HR for competing risk power estimates (0.63) was calibrated such that power was equivalent in the low CM risk tertile for both competing risk (using CSM as the PE) and standard methods (using ACM as the PE). Results: Median follow-up for surviving patients was 59 months. The 5-year cumulative incidence of ACM, CSM, and CM were 57%, 45%, and 12%, respectively. Risk factors for increasing CM were male sex (subdistribution HR [SHR] 1.17; 95% CI, 1.08-1.27), black race (SHR 1.12; 95%, CI 1.02- 1.24), low socioeconomic status (SHR 1.07; 95% CI, .99-1.16), age (SHR 1.04; 95% CI, 1.04-1.04), unmarried (SHR 1.27; 95% CI 1.19-1.37), grade 1-2 tumor (SHR 1.08; 95% CI 1.01-1.17), larynx subsite (SHR 1.15; 95% CI 1.07-1.24), and nonsurgical treatment (SHR 1.18; 95% CI 1.10-1.26). The 5-year cumulative incidence of CM for patients at low, medium, and high risk was 9%, 11%, and 17%, respectively. The Table shows that power is overestimated with standard methods as the risk for CM increases. Conclusions: Multiple demographic and tumor characteristics affect CM risk in the population, suggesting that competing risk methods should be considered for power calculation in HNC clinical trials. CM risk group 5-year cumulative incidence Power Additional patientsrequired (%) ACM CM ACM Competing risks Low 57% 9% 80% 80% 0 Medium 53% 11% 77% 74% 34 (7%) High 62% 17% 84% 78% 74 (15%) No significant financial relationships to disclose.