72nd Annual International Meeting of the Radiation Research Society, Waikoloa Village, HI, October 4-7, 2026
USTUR faculty gave three presentations as the 2026 Annual Radiation Research Society Meeting. These presentations focused on recent work to explore uncertainties in radiation doses and disease outcome, and the impact that these uncertainties may have on epidemiological studies.
Bayesian uncertainty analysis of plutonium organ doses in former nuclear workers
Maia Avtandilashvili (USTUR), Martin Šefl (AMEDIS, Czechia), Joey Y. Zhou (DOE), Sergey Y. Tolmachev (USTUR)
Bayesian internal dosimetry methods are increasingly used in radiation epidemiology to represent uncertainty in organ dose estimates derived from bioassay data. Organ dose is proportional to the time integrated activity retained in that organ. Any systematic bias in long term plutonium organ activity propagates into organ dose bias. Hence, uncertainties in biokinetic model predictions of organ activities can be used as surrogates for uncertainties in dose. Thirty-five voluntary tissue donors to the United States Transuranium and Uranium Registries (USTUR), the former workers from Los Alamos, Rocky Flats, and Hanford were used to evaluate uncertainties in plutonium organ activity predictions. Data included urine bioassay and post-mortem measurements of 239Pu in the liver, skeleton, and respiratory tract. Bayesian models assuming complex exposure scenarios involving chronic and acute intakes were applied under two inference regimes: (i) an all-data model incorporating both urine bioassay and post-mortem organ measurements, and (ii) a urine-only model with autopsy data withheld for validation. Model performance was assessed by comparing predicted organ activities with measured post-mortem values and by evaluating consistency in intake timing. Urine-only inference produced reasonably accurate and precise estimates of systemic plutonium retention (liver plus skeleton). However, respiratory tract estimates were highly unreliable, exhibiting large errors and frequent failure to capture measured organ activities. Using post-mortem organ measurements in combination with urine data improved predictions for all three organs, although this represents retrospective reconciliation rather than independent predictive validation. Intake timing was robust for only a subset of cases; in many individuals, acute intake timing was weakly identifiable and functioned as a nuisance parameter. The analysis confirms that urine-only based Bayesian internal dosimetry methods are suitable for systemic organ dose reconstruction but not for respiratory tract dose estimation. These results underscore the need for organ-specific evaluation of dose uncertainty and caution against the uncritical use of urine-only lung dose estimates in epidemiological risk models.[USTUR-0743-26A]
Factor analysis of impact of outcome misclassification on dose-response conclusions in radiation epidemiology
Xirui Liu (USTUR), Stacey L. McComish* (USTUR), Joey Y. Zhou (DOE), Sergey Y. Tolmachev (USTUR)
Radiation dose-response epidemiological studies commonly use death certificates as the source of health outcomes. However, among an all-autopsied population of former nuclear workers from the United States Transuranium and Uranium Registries (USTUR), an overall misclassification rate of 25.4% was observed when comparing causes of death on death certificates to those on autopsy reports. A subsequent simulation study quantified how such misclassification affects dose response conclusions. That study first generated a reference non-misclassified dataset that paired radiation exposure doses with cancer deaths and was selected to have a borderline non-significant p-value. Cancer and non-cancer outcomes were then randomly reassigned at over and under misclassification rates ranging from 0% to 30%, with 20,000 iterations per rate combination. For each iteration, misclassified odds ratio and corresponding p-value were calculated and compared to the reference dataset. And for each rate combination, the percentage of misclassified odds ratios that moved away from null with the p-values that shifted from non-significant to significant was calculated to reflect the impact of conclusions. Results revealed that misclassification can artificially alter the statistical conclusion between 9.4% and 42.8% of the time in borderline scenarios, and lower misclassification rates led to a higher percentage of altered conclusions. The present study further analyzed how specific factors influence the percentage of datasets where misclassification led to a conclusion change from non-significant to significant. For over-misclassification rates greater than 5%, increasing the baseline disease rate increased the likelihood of a conclusions change. Raising the significance threshold, for example from p < 0.05 to p < 0.10, increased the percentage of affected conclusions across all rate combinations, indicating a greater overall potential impact on conclusions when other conditions were the same. These findings indicate that the high significance thresholds and higher baseline disease rates pose the greatest risk to dose-response conclusions, and provide a necessary caution for interpreting radiation epidemiological studies that rely on death certificate data. *Presenting author [USTUR-0740-26A]
Correcting outcome misclassification in epidemiological studies: a simulation study
Stacey L. McComish (USTUR), Xirui Liu (USTUR), Joey Y. Zhou (DOE), Sergey Y. Tolmachev (USTUR)
Death certificate errors are a common source of uncertainty in radiation epidemiological studies. Approximately 25% of death certificate causes of death from US Transuranium and Uranium Registries donors do not match their corresponding autopsy reports. Recent research has indicated that the impact of these errors on dose-response modeling may not be trivial. This simulation study applied a correction method published by Rogan and Gladen in 1978 to examine how misclassification and correction affected the significance of simulated dose–response associations. Preset underlying odds ratios ranging from 1 to 4.5 were used to generate 10,000 population, misclassified, and corrected datasets. Misclassified datasets were corrected by fitting each dataset using a logistic regression model, and determining the probability of disease as a function of dose. These probabilities were then corrected and used to generate corrected datasets. Results indicated that while the correction method successfully recovered the underlying odds ratio associations, the corrected odds ratio distributions became wider than misclassified odds ratio distributions, leading to inflated false positive and false-negative rates. For example, the false-positive rate can be simulated by generating datasets with an underlying odds ratio of 1, and determining the proportion that have significant p-values. When this was done, the false positive rate was 5% among both population and misclassified datasets, but increased to 21% as a consequence of the correction method. Similarly, false negative rates were simulated using underlying odds ratios that were greater than 1, and determining the proportion of non-significant p-values. Misclassification consistently led to false negative rates that were too high. The correction method attenuated this, resulting in false negative rates that were closer to the true population values, though they could still be too large. False negative rates decreased as the underlying odds ratio increased. This trend held true for the corrected false negative rates, which decreased as the underlying odds ratio increased. [USTUR-0742-26A]