Seminar by Danping Liu

Challenges and Developments in Cancer Outcome Ascertainment: Applications in the U.S. Radiologic Technologists Study

Traditional methods of cancer ascertainment by self-report and/or medical record validation are time-consuming, costly and often results in under-ascertainment of incidence cancer cases. Cancer registry linkage has emerged as an appealing alternative for identifying incident cancer cases, but it may not have complete coverage of the study population. My talk will present methods for addressing analytic challenges of using cancer outcome data to ensure valid inference of cancer risk. Incomplete coverage by cancer registries introduces observation gaps in follow-up, during which incident cancers may have been missed by the registries. To address this issue, we propose to use a mover-stayer model to estimate the posterior probability of an individual living outside the registry coverage area each year, and then developed a proper weight for the person-years at risk incorporated in the survival data likelihood. In addition, when the self-reported cancer outcomes are collected as a supplement to registry-identified cancers, we develop a semiparametric working imputation model to combine the registry-identified outcome with imperfect coverage and the error-prone self-reported outcome.

These methodological developments are motivated from the U.S. Radiologic Technologists (USRT) Study cohort, with the linkage to 43 population-based cancer registries.