Competing mortality (CM) risk in patients with resected stage I and II non-small cell lung cancer (NSCLC): A Surveillance, Epidemiology, and End Results (SEER) analysis.

Abstract

7563 Background: CM risk has been identified as a potential confounder in the interpretation of treatment effects in head and neck cancer (Rose et al. J Clin Oncol. 2011). Lung cancer patients (pts) are at considerable risk for CM due to their advanced age at diagnosis and smoking related chronic diseases. We plan to identify risk factors for CM in pts with early stage NSCLC and develop a statistical model to estimate the effect of CM on power calculation for lung cancer clinical trials. Methods: Using SEER registry we identified 32104 pts who had undergone surgical resection with or without radiation for stage I and II NSCLC between 1994 and 2006. The data set was split into two groups: training set (75%) and testing set (25%). Risk factors for lung cancer-specific mortality (LCSM) and CM were identified using training data by Gray’s sub-distribution regression of competing risk. Pts from the testing data were then stratified according to CM risk and the impact of this risk on power loss was evaluated. Results: The 5-year cumulative incidence of death from lung cancer, other causes and overall mortality was 32.7%, 14.2% and 46.9% respectively. Risk factors for CM were: age (hazard ratio [HR] 1.05), male gender (HR 1.43), divorced (HR 1.30), widowed (HR 1.23) or single (HR 1.29) marital status, squamous (HR 1.40) or not-otherwise-specified (HR 1.22) histology, stage I NSCLC (HR 1.27) and sublobar resection (HR 1.23). The 5-year cumulative incidence of CM in low, mid and high-risk tertiles was 7%, 14% and 21% respectively. Sample size calculations based on all-cause mortality (ACM) result in over-estimation of power as the risk for CM increases. In order to restore the underestimated power in LCSM, 19% and 35% more pts are required in the mid and high CM risk groups respectively (Table). Conclusions: Our findings indicate that conventional sample size calculation methods can result in significant loss of power and incorporating CM risk models in power estimation should be considered for clinical trials involving early stage NSCLC pts. [Table: see text]