Analysis of Hanford Proportional Mortality Data, Adjusted for Age and Year of Death

Page 1

Analysis o f Hanford Proportional Mortality Data, Adjusted for Age The analysis sumarized i n the attached tables was done t o investigate and Year of Death the suggestion that some of the results reported by Mancuso, Stewart, and Kneale may- be explainable as artifacts of their analysis. The approach has been 'to adjust for certain possible confoundings, viz., association o f 'cumulative dose with age of death and year of death, and possible selection factors between exposed and non-exposed workers; t o compute test statistics that do not assume a symmetric distribution for dose; and t o analyze the death certificate data provided by ORNL insufficient detail that some judgment can be made asto the extent of the bias introduced by a proportional mortality analysis, and as t o whether the reported radiation effects could be. extreme results to be expected by chance alone from a large number of The basic statistical procedure has been a contingency table method, adjusted for age a t death (<49,- 50-64, 65+) and year of death (1943-60, A t each level of these factors (e.g., 549 and 1943-60, rows corresponding t o death from a specified disease o r from other causes and columns t o non-exposed and t o dose intervals within the exposed class. - Expected cell frequencies, variances, and covariances were computed from row and column sums, and then summed over tables to provide details of the nullhypothesis distribution of the summary tables. The main statistical results are the p-values for the exposed vs. non-exposed comparison and for the test of linear trend with dose adjusted

Page 2

for exposure; a value near one of the latter indicates a positive association of the corresponding cause of death with cumulative dose, a value near zero indicates a negative association. Extreme p-values f o r the exposed - non-exposed comparison, i n the absence of a dose response, suggest selection of the exposed or non-exposed population f o r reasons with more moderate p-values for the main two statistics,might suggest a true dose response, especially i f the adjusted trend statistic is at a included as a check on the possible effects of skewness of the dose distribution. A n extreme p-value for linear trend unadjusted for exposure, The trend statistics for the cube root of dose were The finer dose intervals in the l l - interval tables provide greater power against increasing trends; the 6-interval tables provide a check against. numerical or distribution al instability of the test statistics. Analyses were performed for cumulative dose a t death, and at3,5, 10, 15, and 20 years prior t o death. 3-digit ICD code, and i n broader groups. Cancers were treated separately for each Non-cancers were treated in broad groups, except for chronic bronchitis and emphysema, which were treated Several causes of death were present much more, or much l e s s , i n the exposed than in the non-exposed, indicating that the groups probably were selected on some basis related t o health. and multiple myeloma, were significantly related t o dose among males. non-cancer, chronic MI, consistently decreased with increasing dose. Since Two cancers, pancreatic cancer the results for pancreatic cancer and multiple myeloma are based on relatively few cases, i t seems unlikely that these results could be ascribed toa defect

Page 3

of the proportional mortality approach; selection of the exposed population, even differential ly by dose, could not be expected to create an apparent dose relationship for these diseases alone. However, a closer examination o f each o f the individual diagnoses would be i n order, since only a few misdiagnoses could have created the effects. Pancreatic cancer, i n .part j cula r, has a relatively low percentage of histologically proven cases compared with other cancer sites, and therefore may be unusually subject t o misdiagnosis on death certificates. breast appeared t o be related to dose i n unadjusted contingency table analyses, but these apparent relationships did not survive adjustment for Cancers ofthe lung, brain, and female age and year of death. [The possibility that the results for pancreatic cancer and mu1 tiple myeloma are statistical flukes, which might be expected t o be obtained i n an analysis of many different causes of death, cannot be ruled out. less likely for multiple myelona than for pancreatic cancer, however, since multiple myeloma isa very rare disease and the dose distribution of the cases is-extreme. Excess risks o f multiple myeloma and pancreatic cancer, among other cancers, have been reported among U.S. and British radiologists.On ' the other hand, the evidence fora dose response for these two diseases among Japanese A-bomb survivors i s very weak. evidence fora dose effect among Hanford workers should involve those two diseases and not leukemia or malignant lymphomas other than mu1 tiple Itis someHhat surprising that