describing the handling of out-of-area participants in exposure-outcome analyses, and the modifications made to the list and definitions of the primary and alternative thyroid disease outcomes in appendix 2 of the HTDS analysis plan. In addition, specific responses to comments from 5 members of the HTDS Advisory Committee and the 4 external reviewers are summarized.
Executive summary of the Hanford Thyroid Disease Study study protocol, May 26, 1993.
Executive summary of the pilot-study final report, January 24, 1995.
When the committee met, Drs. Scott Davis and Kenneth Kopecky, members of the staff of the Fred Hutchinson Cancer Research Center and the principal investigators in the HTDS, enlarged on the description of the proposed analysis and the changes in the original plan prompted by the need to address the additional source of exposures and the uncertainties in dose. The original plan has been reviewed by the Radiation Studies Branch of the Centers for Disease Control and Prevention, by several ad hoc peer reviewers, and by the Hanford Thyroid Disease Study Advisory Committee. The comments of those varied reviewers were shared with the committee, and the HTDS investigators' responses to them were presented in writing.
Overall, the committee was impressed by the clarity and thoroughness of the analytic plan and compliments the investigators on the care with which they have responded to the need to develop such a plan. Moreover, the study has been thoughtfully and competently executed, and the degree of attention to detail will no doubt provide data of high quality. The investigators have successfully traced 94% of the historical study cohort established 50 years ago–a very impressive achievement. However, the investigators are now faced with the daunting task of analyzing this large and complex data set, which offers almost unlimited possibilities for model-building and hypothesis-testing. The very richness of the data can produce problems, specifically problems that arise from multiple comparisons, some of which will inevitably be statistically significant. To better distinguish between an effect of interest and a chance finding, we recommend that the investigators delineate with more rigor the choices to be made about variable specification and by setting forth a priori a subset of primary analyses (perhaps 10) believed to test the research question(s) directly. A focused selection and description of the primary hypotheses to be tested will provide guidance both to investigators and to reviewers at the conclusion of the data analysis and will avoid to some extent the statistical problems arising from multiple comparisons.
The basic analytic plan for the study envisages a stratified linear exposure-response model to accommodate dichotomous and continuous variables and to account for uncertainty in dose estimates. The presentation of the plan to the committee included a discussion of the proposed revisions regarding the handling of uncertainty for both the HTDS and NTS doses. For the HTDS doses, a dose-estimation program, CIDER, was developed to perform a Monte Carlo simulation of individual doses, taking into account the uncertainties in various components of the exposure model. Dose estimates and