Technical Report No. 4-91
The effect of diagnostic misclassification on noncancer and cancer mortality dose response in the RERF Life Span Study
Sposto R, Preston DL, Shimizu Y, Mabuchi KEditor’s note: A publication based on this report was published in Biometrics 48:605-17, 1992.
Summary
We performed analyses of cancer and noncancer mortality in the Radiation Effects Research Foundation (RERF) Life Span Study (LSS) to determine whether the observed increased risk of noncancer death due to radiation exposure could be attributed solely to misclassification of causes of death on death certificates. Cancer and noncancer misclassification rates and their dependence on age at death were estimated from a series of autopsies performed on LSS participants between 1961 and 1975. The crude misclassification rates were 20% for cancer and 2.8% for noncancer. Although the noncancer dose response remained significant, correcting for this amount of misclassification reduced the estimated noncancer excess relative risk (ERR) at 1 Gy exposure by 21% and the number of excess noncancer deaths in the cohort by 23%. The estimated cancer ERR at 1 Gy was increased by 12% and the excess cancer deaths by 16% as a result of the correction. The statistical significance of the noncancer dose response was relatively insensitive to underestimating the cancer misclassification rate, but was sensitive to assuming that cancer misclassification was positively associated with dose. We discuss the implementation of the EM algorithm for adjusting for misclassification and extensions of the method to more than two causes of death.
We performed analyses of cancer and noncancer mortality in the Radiation Effects Research Foundation (RERF) Life Span Study (LSS) to determine whether the observed increased risk of noncancer death due to radiation exposure could be attributed solely to misclassification of causes of death on death certificates. Cancer and noncancer misclassification rates and their dependence on age at death were estimated from a series of autopsies performed on LSS participants between 1961 and 1975. The crude misclassification rates were 20% for cancer and 2.8% for noncancer. Although the noncancer dose response remained significant, correcting for this amount of misclassification reduced the estimated noncancer excess relative risk (ERR) at 1 Gy exposure by 21% and the number of excess noncancer deaths in the cohort by 23%. The estimated cancer ERR at 1 Gy was increased by 12% and the excess cancer deaths by 16% as a result of the correction. The statistical significance of the noncancer dose response was relatively insensitive to underestimating the cancer misclassification rate, but was sensitive to assuming that cancer misclassification was positively associated with dose. We discuss the implementation of the EM algorithm for adjusting for misclassification and extensions of the method to more than two causes of death.