Comparison of comorbidity measures to predict postoperative lung cancer survival in the National Cancer Database (AFT-03).

Abstract

6519 Background: Comprehensive assessment of comorbidity in cancer registries is critical for comparative effectiveness research. The National Cancer Database (NCDB) measures comorbidity with a diagnosis code-based Charlson Comorbidity Index (CCI) abstracted from discharge abstracts or billing face sheets. However, the prognostic performance of this code-based CCI has not been compared with a medical chart-based CCI or individual comorbid conditions in a nationally representative sample of patients with lung cancer. Methods: Through a special study of the NCDB, cancer registrars performed chart abstraction for 18 perioperative comorbid conditions for 9,640 randomly selected patients with stage I-III non-small cell lung cancer resected in 2006-07 at 1,150 Commission on Cancer-accredited facilities. We compared the prognostic performance of the NCDB code-based categorical CCI (0, 1, 2+), special study chart-based continuous CCI, and individual comorbid conditions in 3 separate Cox proportional hazards models for 5-year postoperative overall survival. All models adjusted for demographic and clinical characteristics. Results: Median age was 67 (IQR 60-74). The most common comorbidities were COPD (40%) and CAD (21%). Five-year postoperative overall survival was 55.5%. Agreement between the code- and chart-based CCIs was 51.9% with the code-based CCI underestimating comorbidity for 36.2% patients. The model including individual comorbid conditions had the best prognostic performance (R2 0.196, C index 0.654). COPD, CAD, CHF, dementia, diabetes, moderate/severe renal and liver disease, peripheral vascular disease, psychiatric disorder, and substance abuse were independently associated with decreased survival. The chart-based CCI model (R2 0.189, C index 0.650) predicted postoperative survival better than the code-based CCI model (R2 0.181, C index 0.645). Conclusions: The NCDB code-based CCI underestimates comorbidity in patients with surgically resected lung cancer. The chart-based CCI and data on individual comorbid conditions improved prognostic performance and would be valuable additions to the NCDB to strengthen comparative effectiveness research.