Assessing the quality of race/ethnicity, tumor, and breast cancer treatment information in a non-SEER state registry.

INTRODUCTION: The quality of population-based cancer registries are largely defined by the completeness, accuracy, and timeliness of incident cases and demographics reported. However, both Surveillance, Epidemiology, and End Results (SEER) cancer registries and non-SEER population-based state cancer registries have been regularly used to examine treatment patterns. While the quality of treatment data in SEER cancer registries has often been examined and improved, the quality of such data in non-SEER state registries has rarely been assessed.
METHODS: We used self-reported (SR) and medical record (MR) abstracted data from a population-based breast cancer study for comparison with information contained in the Illinois State Cancer Registry (registry). Using either MR or SR as the gold standard, we estimated concordance, kappa, and sensitivity for the presence or absence of surgery and initiation of chemotherapy, radiation and hormone therapy, as well as tumor characteristics, race/ethnicity and insurance status.
RESULTS: The accuracy of most of the data elements examined was generally high. For instance, there was almost perfect agreement between SR race/ethnicity and registry documentation (k = 0.92). MR and registry data on tumor stage, grade, ER/PR status, and node status had substantial agreement (k = 0.78-0.88). In regard to treatment information, surgery was rarely underdocumented in registry data, while radiation and chemotherapy were modestly underdocumented (8 percent-16 percent). On the other hand, per SR or MR, the registry generally failed to document hormonal treatment in a large proportion of cases (0.38 and 0.52, respectively). Health insurance information in the registry was also not well documented. There was only moderate agreement (k = 0.41) between SR and registry health insurance status, with uninsured patients being the least likely to be documented as such in the registry (sensitivity = 0.37 vs 0.96 and 0.63 for public and private insurance status, respectively). DISCUSSION: While some registry data elements are quite reliable, others warrant concern and must be interpreted with great caution. Understanding the strengths and limitations of a population-based non-SEER state cancer registry data can be useful to researchers who use these data sources to examine population cancer patterns or carry out cancer studies.