Testing for misclassified deaths in cancer registries by comparing relative and cause-specific survival

Abstract

Historically, the cause of death reported in cancer registries was often unreliable, and this crucial information could not be exploited by competing-risks methods. To circumvent this issue, the relative survival methodology leveraged the knowledge of general population mortality from national life tables, together with a few assumptions, to still estimate the cancer-specific mortality. The literature contains several empirical studies of the quality of the cause-of-death reported in real cancer registries datasets, but none are seriously formalized. We propose for the first time a formal statistical test for the quality of the cause of death reported in the databases, exploiting techniques from both the relative survival and competing risks methodologies. Our test can be used by practitioners to assess the alignment of their reported causes of death, under the standard relative survival hypotheses. We study the behavior and calibration of the tests on simulated datasets and finally provide an illustration using the SEER registry.

comments powered by Disqus