A Comparison of the NCDB and SEER Database for Research Involving Head and Neck Cancer

Abstract

Objective To determine whether structural differences in data sampling between the National Cancer Database (NCDB), a non-population-based cancer registry, and Surveillance, Epidemiology, and End Results (SEER), a population-based cancer registry, result in differences in patient characteristics or oncologic outcomes. Study Design Retrospective cohort study. Setting NCDB and SEER database. Subjects and Methods Patients with head and neck cancer (HNC) were included from 2004 to 2014. The primary outcome, weighted differences in characteristics between the databases, was evaluated for each head and neck subsite (oral cavity [OC], oropharynx [OP], hypopharynx [HP], and larynx [LX]). The secondary outcome measure, overall survival (OS), was evaluated using Kaplan-Meier (KM) estimates of survival and Cox proportional hazards (PH) regression modeling. Results In total, 112,007 and 340,420 HNC cases were registered in SEER and the NCDB, respectively. The mean age at diagnosis for the 4 head and neck subsites differed by no more than 1.1 years between the 2 databases. The largest difference in patient or tumor characteristics was the frequency of OC subsite lip cancer (weighted proportional difference, 6.9%; 95% confidence interval, 6.5%-7.3%). Unadjusted KM estimates of 5-year OS differed by no more than 2% (OP, HP, and LX subsites). On Cox PH modeling, adjusted hazard ratios ranged from 0.89 to 0.91 for patients of different head and neck subsites in the NCDB relative to SEER. Conclusions Patients with HNC in the SEER database and NCDB do not greatly differ in terms of demographics, treatment, and survival. Decisions to use either database should be driven by the data fields, which vary between the registries.