Commentary and Review Series 2-91
The influence of death-certificate errors on cancer mortality trends
Ron E, Hoel DG, Carter RL, Mabuchi KJ Natl Cancer Inst 85(13):1063-8, 1993
Summary
Over the past few years, several reports have suggested a recent increase in cancer mortality based on death-certificate diagnoses. To explore the effect of death-certificate errors on temporal trends in cancer mortality rates, we analyzed the data from the Atomic Bomb Casualty Commission/Radiation Effects Research Foundation’s autopsy program in Hiroshima and Nagasaki. This series includes 5886 autopsies conducted between 1961 and 1987. Our analyses were focused on lymphoma, cancer of the breast, neoplasms of the brain, multiple myeloma, and melanoma (172 cases, total) because of concern over reports of their increased mortality. These 172 autopsy cases were referred to as Cancers of Interest. A significant increase in detection rates was observed for these Cancers of Interest primarily due to a large rise in mortality between 1976 and 1987. For the remaining cancers excluding stomach and lung (defined as Other), the pattern was similar to that seen for Cancers of Interest, but the fluctuation over time was not statistically significant. Confirmation rates generally increased with time except for Cancers of Interest.
As a measure of bias in mortality rates due to death-certification errors and as a method to quantify under- or overestimation of death-certificate-based mortality rates, an adjustment factor (confirmation rate divided by detection rate) was calculated. The higher the adjustment factor, the greater the need to compensate for underreporting. For Cancers of Interest the adjustment factor decreased dramatically over time, but it did not change significantly for Other cancers. When the adjustment factors for Cancers of Interest and Other were compared, a statistically significant difference was found. For Cancers of Interest, a significant interaction between type of cancer and period was seen. Our findings indicate that considerable care must be shown when interpreting temporal trends in cancer vital statistics.
Over the past few years, several reports have suggested a recent increase in cancer mortality based on death-certificate diagnoses. To explore the effect of death-certificate errors on temporal trends in cancer mortality rates, we analyzed the data from the Atomic Bomb Casualty Commission/Radiation Effects Research Foundation’s autopsy program in Hiroshima and Nagasaki. This series includes 5886 autopsies conducted between 1961 and 1987. Our analyses were focused on lymphoma, cancer of the breast, neoplasms of the brain, multiple myeloma, and melanoma (172 cases, total) because of concern over reports of their increased mortality. These 172 autopsy cases were referred to as Cancers of Interest. A significant increase in detection rates was observed for these Cancers of Interest primarily due to a large rise in mortality between 1976 and 1987. For the remaining cancers excluding stomach and lung (defined as Other), the pattern was similar to that seen for Cancers of Interest, but the fluctuation over time was not statistically significant. Confirmation rates generally increased with time except for Cancers of Interest.
As a measure of bias in mortality rates due to death-certification errors and as a method to quantify under- or overestimation of death-certificate-based mortality rates, an adjustment factor (confirmation rate divided by detection rate) was calculated. The higher the adjustment factor, the greater the need to compensate for underreporting. For Cancers of Interest the adjustment factor decreased dramatically over time, but it did not change significantly for Other cancers. When the adjustment factors for Cancers of Interest and Other were compared, a statistically significant difference was found. For Cancers of Interest, a significant interaction between type of cancer and period was seen. Our findings indicate that considerable care must be shown when interpreting temporal trends in cancer vital statistics.