Using Multivariate Capture-Recapture Techniques and Statewide Hospital Discharge Data to Assess the Validity of a Cancer Registry for Epidemiologic Use

Background: Cancer surveillance is essential for assessing patterns of cancer. State cancer registries do not capture all cases. We used a statewide hospital discharge file to estimate the capture rate and determine biases in capture by a state cancer registry. Methods: We merged the Virginia hospital discharge file and cancer registry for 1995, then used multivariate two- source capture-recapture techniques to control for heterogeneity and more accurately estimate the number and characteristics of missing breast, lung, colorectal and prostate cancer cases. Results: Results suggest heterogeneity of capture, with rates a function of demographics, surgery and being in a hospital with a cancer program certified by the American College of Surgeons(ACOS). Overall registry capture rates ranged from 66% (prostate)–79% (breast). Capture rates varied by subgroup, with differences larger according to surgery and ACOS certification status than for demographic subgroups. While the registry captured most cases who had surgery (85–89%), capture rates for those without surgery was much lower (37–71%). Conclusion: We conclude that hospital discharge data and multivariate capture-recapture techniques are useful to registries to estimate the number of missing cases and assess bias in capture. Epidemiologic research based on registry data alone would likely provide biased, misleading results.