ing a decrease with time and older survivors showing an increase with time.
Preston and colleagues (2002b) investigated tumors of the nervous system and pituitary gland based on cases ascertained through the Hiroshima and Nagasaki Tumor and Tissue Registries and through medical records from RERF and major medical institutions in Hiroshima and Nagasaki. Histologic diagnoses were obtained by having four pathologists independently review slides and medical records. The majority of the 228 central nervous system tumors included in the study were benign. A statistically significant dose-response association was observed for all nervous system tumors with an estimated ERR/Sv of 1.2 (95% CI 0.6, 2.1). The ERR/Sv was highest for schwannomas (4.5; 95% CI 1.9, 9.2), but the dose-response for all other central nervous system tumors evaluated as a group was also statistically significant. The dose-responses for all nervous system tumors and for schwannomas were both statistically significant when limited to subjects with doses of less than 1 Sv, and there was no evidence that the slope for this low-dose range was different from that for the full range. Modification of risk by sex, age at exposure, and attained age was also investigated.
A statistically significant dose-response relationship with mortality from nonneoplastic disease in A-bomb survivors was demonstrated by Shimizu and colleagues (1992) based on mortality data for 1950–1985. The addition of five years of mortality data (through 90) strengthened the evidence for this effect and allowed a more detailed evaluation (Shimizu and others 1999). In these analyses, statistically significant associations were seen for the categories of heart disease, stroke, and diseases of the digestive, respiratory, and hematopoietic systems.
Preston and colleagues (2003) updated these results and present analyses of deaths from all causes excluding neoplasms, blood diseases, and external causes such as accidents or suicide. They give considerable attention to the fact that for a few years after the atomic bomb explosions, baseline risks for noncancers in proximal survivors (within 3000 m of the hypocenter) were markedly lower than those in distal survivors. They refer to this as the “healthy survivor effect” and note that it could lead to distortion of the dose-response, particularly in the early years of follow-up. They also note that a small difference (2%) in baseline risks for proximal and distal survivors persisted in later years, which they consider likely to be due to demographic factors such as urban-rural differences. They address this potential source of bias by conducting analyses restricted to the period 1968–1997 and by including an adjustment for differences in proximal and distal survivors (although results without the adjustment are also presented).
The estimated ERR/Sv for noncancers based on a linear model with no dependence on age at exposure or sex was 0.14, generally lower than that for all solid cancers (where the ERR/Sv depends on age and sex). There was no evidence of a statistically significant dependence on either age at exposure or sex, but the data were compatible with effects similar to those estimated for solid cancers. A linear dose-response function fitted the data well, but it was not possible to rule out a pure quadratic model or a model with a threshold as high as 0.5 Sv. Similar to Shimizu and colleagues (1999), significant dose-response relationships were found for heart disease, stroke, respiratory disease, and digestive disease. There was no evidence of radiation effects for infectious diseases or all other noncancer diseases in the group evaluated. Lifetime noncancer risks for people exposed to 1 Sv were estimated to be similar to those for solid cancer for those exposed as adults, and about half those for solid cancer for those exposed as children. Because baseline risks for the noncancer category evaluated are greater than those for all solid cancers, even the relatively small ERR/Sv leads to a fairly large absolute lifetime risk.
Because small ERRs can easily arise from bias, Shimizu and colleagues (1999) evaluated several potential sources of bias, including misclassification of cause of death, confounding, and cohort selection effects. Although Preston and coworkers (2003) discuss cohort selection effects in detail, they did not reevaluate other sources of bias. The committee summarizes the discussion provided by Shimizu and colleagues in the remainder of this section.
With regard to misclassification, they note that Sposto and coworkers (1992) investigated the possibility of bias from this source using mortality data through 1985. These investigators used estimated age-dependent misclassification probabilities obtained from RERF autopsy data to conduct analyses that corrected for misclassification and found that estimates for noncancer mortality were reduced by 20%, but remained highly statistically significant. Shimizu and colleagues (1999) used mail survey and interview data to examine the possible effect of several potential confounders including educational history and smoking. Although most of the factors evaluated were found to affect noncancer mortality, they were not found to be associated strongly with dose. Analyses adjusted for various confounders, based on survivors with available data, resulted in ERRs/Sv that were very similar to the unadjusted values.
Shimizu and colleagues (1999) also evaluated noncancer diseases of the blood, benign neoplasms, and deaths from external causes. Because these categories were not reevaluated by Preston and coworkers (2003), the committee summarizes these findings. The ERR/Sv for the 191 deaths from noncancer diseases of the blood was estimated to be 1.9 (90% CI 1.2, 2.9), larger than the estimated values for most solid cancers. The accuracy of death certificate diagnosis is known