Influence of death certificate errors on cancer mortality trends.

BACKGROUND: Evaluation of mortality data is an important tool in assessing both disease time trends and differences in populations. However, the reasons for changes in cancer mortality rates have been controversial. Questions have been raised concerning whether these increases are real or simply the result of changing diagnoses and death certificate reporting.
PURPOSE: Our purpose was to determine on the basis of autopsy data if death certificate reporting varies over age, time, and cancer type and to explore the effect of death certificate error on recent cancer mortality trends.
METHODS: Data were analyzed from 5886 autopsies collected by the Radiation Effects Research Foundation in Hiroshima, Japan, between 1961 and 1987. Death certificates were linked to autopsy data for analyses. An adjustment factor was calculated as a ratio of the accuracy of these death certificates in identifying cancer (detection rate) and listing cancer (confirmation rate) as the cause of death. This adjustment factor measures bias in mortality rates due to death certification errors and quantifies underestimation and overestimation of cancer mortality rates.
RESULTS: Our analyses focused on lymphoma, breast cancer, neoplasms of the brain, multiple myeloma, and melanoma because of reported mortality increases. For these cancers, the adjustment factor decreased significantly (P = .02) over time, implying that death certificate accuracy has improved. This change appears to account for 60% of the observed increase in these cancers during the time period of study. For total cancer, persons 75 years or older have a high adjustment factor but it decreased over time. This decrease suggests an artifactual increase in total cancer mortality rates of about 1% per year for this older group.
CONCLUSIONS: The quality of death certificate reporting has generally remained constant for most groups during the period 1961-1987. In the study population, there was a consistent underestimation of total cancer mortality of about 18%. For the five types of cancer studied, there has been less underestimation of cancer mortality since 1976. For some specific cancers and for persons 75 years or older, improvements in death certificate accuracy have occurred over time. These data imply that the total death certificate error rate varies considerably by cancer type, time period, and age at death. IMPLICATIONS: The changes in death certificate reporting for some sites and the elderly appear to have artifactually created increases in cancer mortality rates. These artifacts need to be considered when using mortality data for prevention research and health care planning.