Examining bias due to method of follow-up for cancer incidence in a large U.S. cohort: Self-report versus registry linkage

Abstract

Background

Cancer incidence follow-up of many U.S. cohort studies relies on self-report due to the lack of a national cancer registry.

Methods

In the U.S. Radiologic Technologists (USRT) Study, cancer incidence has been captured via self-report approximately every 10 years between the mid-1980s and mid-2010s and via linkages with state cancer registries during 1999–2012. We compared associations of smoking and lung cancer (rapidly fatal) and body mass index and thyroid cancer (rarely fatal) according to method of cancer incidence follow-up: self-report versus registry-based. Simulation studies were performed to quantify bias and assess relative efficiency of cancer ascertainment methods.

Results

During 1999–2012, hazard ratio (HR) estimates were similar regardless of cancer follow-up method, although self-reported data yielded the widest confidence intervals (CIs). Simulation studies showed that registry data yielded the least-biased HRs. The usefulness of self-reported data depends mostly on its specificity: with perfect specificity, self-reported data led to nearly unbiased inference. However, with imperfect specificity, self-reported data resulted in 3–9 % bias and lower than nominal CI coverage.

Conclusions

These findings highlight the benefits of registry-based cancer incidence follow-up compared to self-report in contemporary U.S. cohort studies assuming high coverage by registries of the study population.