Testing prediction capabilities of an 131I terrestrial transport model by using measurements collected at the Hanford nuclear facility

A model describing transport of 131I in the environment was developed by SENES Oak Ridge, Inc., for assessment of radiation doses and excess lifetime risk from 131I atmospheric releases from Oak Ridge Reservation in Oak Ridge, Tennessee, and from Idaho National Engineering and Environmental Laboratory in southeast Idaho. This paper describes the results of an exercise designed to test the reliability of this model and to identify the main sources of uncertainty in doses and risks estimated by this model. The testing of the model was based on materials published by the International Atomic Energy Agency BIOMASS program, specifically environmental data collected after the release into atmosphere of 63 curies of 131I during 2-5 September 1963, after an accident at the Hanford PUREX Chemical Separations Plant, in Hanford, Washington. Measurements of activity in air, vegetation, and milk were collected in nine counties around Hanford during the first couple of months after the accident. The activity of 131I in the thyroid glands of two children was measured 47 d after the accident. The model developed by SENES Oak Ridge, Inc., was used to estimate concentrations of 131I in environmental media, thyroid doses for the general population, and the activity of 131I in thyroid glands of the two children. Predicted concentrations of 131I in pasture grass and milk and thyroid doses were compared with similar estimates produced by other modelers. The SENES model was also used to estimate excess lifetime risk of thyroid cancer due to the September 1963 releases of 131I from Hanford. The SENES model was first calibrated and then applied to all locations of interest around Hanford without fitting the model parameters to a given location. Predictions showed that the SENES model reproduces satisfactorily the time-dependent and the time-integrated measured concentrations in vegetation and milk, and provides reliable estimates of 131I activity in thyroids of children. SENES model generated concentrations of 131I closer to observed concentrations, as compared to the predictions produced with other models. The inter-model comparison showed that variation of thyroid doses among all participating models (SENES model included) was a factor of 3 for the general population, but a factor of 10 for the two studied children. As opposed to other models, SENES model allows a complete analysis of uncertainties in every predicted quantity, including estimated thyroid doses and risk of thyroid cancer. The uncertainties in the risk-per-unit-dose and the dose-per-unit-intake coefficients are major contributors to the uncertainty in the estimated lifetime risk and thyroid dose, respectively. The largest contributors to the uncertainty in the estimated concentration in milk are the feed-to-milk transfer factor (F(m)), the dry deposition velocity (V(d)), and the mass interception factor (r/Y)dry for the elemental form of iodine (I2). Exposure to the 1963 PUREX/Hanford accident produced low doses and risks for people living at the studied locations. The upper 97.5th percentile of the excess lifetime risk of thyroid cancer for the most extreme situations is about 10(-4). Measurements in pasture grass and milk at all locations around Hanford indicate a very low transfer of 131I from pasture to cow's milk (e.g., a feed-to-milk transfer coefficient, F(m), for commercial cows of about 0.0022 d L(-1)). These values are towards the low end of F(m) values measured elsewhere and they are low compared to the F(m) values used in other dose reconstruction studies, including the Hanford Environmental Dose Reconstruction.     

Age-specific uncertainty of the 131I ingestion dose conversion factor

The production of weapons-grade nuclear materials and their by-products has resulted in a number of releases from United States Department of Energy facilities. 131I, a fission by-product, is one of the most common radionuclides generated and released to the environment. It is known that there are differences in various physiological parameters over all age groups when considering biokinetic modeling of iodine. The establishment of age-specific dose conversion factor uncertainty is necessary for accurate internal dose assessment. The 131I dose conversion factor determined herein is log-normally distributed with varying age-specific distribution characteristics. The two most important parameters for determination of the dose conversion factor, in all age groups, are thyroid mass and iodine uptake fraction. These parameters are assumed to be highly correlated with a relationship that is quite important to dose conversion factor uncertainty. Dose estimates to individuals exposed to radioiodine can be determined more accurately with an increased understanding of the correlation between thyroid mass and uptake fraction. Improved dose estimates following oral intakes of 131I can be made from the consideration of age-specific dose conversion factors and their input parameters.     

Uncertainties in dose coefficients from ingestion of 131I, 137Cs, and 90Sr

Quantification of uncertainties in doses from intakes of radionuclides is important in risk assessments and epidemiologic studies of individuals exposed to radiation. In this study, the uncertainties in the doses per unit intake (i.e., dose coefficients) for ingestion of 131I, 137Cs, and 90Sr by healthy individuals have been determined. Age-dependent thyroid dose coefficients were derived for 131I. The analysis for 131I uses recent measurements of thyroid volume obtained by ultrasonography, which indicate a thyroid mass lower than that previously obtained using autopsy measurements. The coefficients for 137Cs are determined using the relationship between the biological half-lives and the amount of potassium in the human body. The most recent International Commission on Radiological Protection biokinetic model was employed to determine the uncertainties for 90Sr. For 137Cs and 90Sr, the dose coefficients represent exposure in adulthood and they were determined for all organs of radiological importance. The uncertainty in the estimated dose coefficients represent state of knowledge estimates for a reference individual, and they are described by lognormal distributions with a specified geometric mean (GM) and geometric standard deviation (GSD). The estimated geometric means vary only slightly from the dose coefficients reported by ICRP publications. The largest uncertainty is observed in the dose coefficients for bone surface (GSD = 2.6), and red bone marrow (GSD = 2.4) in the case of ingestion of 90Sr. For most other organs, the uncertainty in the 90Sr dose coefficients is characterized by a GSD of 1.8 (or less for some organs). For 131I, the uncertainty in the thyroid dose coefficients is well represented by a GSD of 1.7 for both sexes and all ages other than infants for whom a GSD of 1.8 is more appropriate. The lowest uncertainties are obtained for the dose coefficients from ingestion of 137Cs (GSD = 1.24 for males; 1.4 for females). A dominant source of uncertainty in the ingestion dose coefficients is the variation of the biokinetic parameters. For 131I, the largest contribution to the uncertainty comes from the variation in the thyroid mass, but the contribution of the biokinetic parameters is comparable. The biokinetic parameters with the largest contribution to the uncertainty are (a) the fractional uptake from blood to thyroid in the case of ingestion of 131I, (b) the absorbed fraction from the gastrointestinal tract (f1) in the case of 90Sr, and (c) the amount of potassium in the body for 137Cs. The contribution to the uncertainty of the absorbed fraction (which accounts for the fraction of energy deposited in the target organ) is the smallest contributor to the uncertainty in the dose coefficients for most organs. To reduce the uncertainty in the dose estimated for a real individual, one should determine the above-mentioned parameters for the specified individual rather than to rely on assumptions for a reference individual.     

Reconstruction of doses from radionuclide inhalation for nuclear power plant workers using air concentration measurements and associated uncertainties.

This paper presents a case study to illustrate the influence of parameter uncertainties on calculating an internal radiation dose of one actual nuclear plant worker, alias Mr. X, as well as the utility of air sampling for internal dose reconstruction. Input probability distributions for air concentrations of radionuclides were derived from empirical air measurements taken by fixed area air samplers. The total internal dose was calculated by multiplying radionuclide intake by dose conversion factors in Monte-Carlo simulations. There is significant variability in dose conversion factors and uncertainty in the estimated concentrations of radionuclides in the air to which Mr. X was exposed. The high variability and uncertainty of the model parameters contributed to large ranges of predicted internal doses for Mr. X. Two-dimensional Monte-Carlo simulations were conducted to separate contributions of the uncertainty and variability. Sensitivity analysis was conducted to determine which of the input parameters contributed most to uncertainty in internal dose estimates. Our analysis suggests that the uncertainty resulting from use of general air surveys contributes more to the internal dose ranges than the variability from DCFs and other population-derived parameters. Reduction of the uncertainty in reconstructed internal dose can be achieved by using personal air sampling and/or individual bioassays and regular whole-body counting.     

A re-evaluation of the 131I atmospheric releases from the Hanford site

The atmospheric release of 131I from the Hanford site for the 1950's and 1960's, focused on the period of releases after the year 1950, has been re-evaluated using processing plant stack monitoring data to address a series of questions and concerns that have arisen related to the source term. Historical stack monitoring data have been used to re-assess the releases by creating either a release factor to use with the calculated plant throughput or using the stack monitoring results as the basic estimate, and the results have been verified using historical atmospheric monitoring data from a location several kilometers distant. Uncertainties in all of the historical data have been addressed in the re-assessment. Compared to the original estimate between 1950 and 1971 of 2.46 +/- 0.71 PBq, the stack monitoring results show a release of 131I to the atmosphere of 1.55 +/- 0.23 PBq. The concurrent atmospheric monitoring results imply a release of 1.75 +/- 0.11 PBq over the same period, but this result is inflated by inclusion of global fallout The total effective dose estimated to a full-time, nearby adult resident from 131I using the Heeb source term from 1950 through 1972 is 0.73 mSv; using the source term based on stack monitoring data in the Hanford Environmental Dose Reconstruction project models, it is 0.51 mSv.     

Incorporating parameter uncertainty into Quantitative Microbial Risk Assessment (QMRA)

Modern statistical models and computational methods can now incorporate uncertainty of the parameters used in Quantitative Microbial Risk Assessments (QMRA). Many QMRAs use Monte Carlo methods, but work from fixed estimates for means, variances and other parameters. We illustrate the ease of estimating all parameters contemporaneously with the risk assessment, incorporating all the parameter uncertainty arising from the experiments from which these parameters are estimated. A Bayesian approach is adopted, using Markov Chain Monte Carlo Gibbs sampling (MCMC) via the freely available software, WinBUGS. The method and its ease of implementation are illustrated by a case study that involves incorporating three disparate datasets into an MCMC framework. The probabilities of infection when the uncertainty associated with parameter estimation is incorporated into a QMRA are shown to be considerably more variable over various dose ranges than the analogous probabilities obtained when constants from the literature are simply ‘plugged’ in as is done in most QMRAs. Neglecting these sources of uncertainty may lead to erroneous decisions for public health and risk management.     

A radioiodine speciation, deposition, and dispersion model with uncertainty propagation for the Oak Ridge dose reconstruction

Between 1944 and 1956, radioactive 131I was released into the atmosphere from operations at the Oak Ridge National Laboratory in Oak Ridge, TN. The releases occurred from stacks and from building vents and openings in three different chemical forms: elemental, organic, and particulate. During their transport in the atmosphere, different forms of iodine react differently with other atmospheric chemicals and moisture, and are removed from the plume at different rates by the processes of dry and wet deposition. A modified Gaussian plume model was developed to address the processes of radioiodine speciation, deposition, depletion, and dispersion in the atmosphere, and to propagate uncertainties in input parameter values through to the ground-level concentrations of 131I in air. A unique approach was used to develop an implicitly correlated set of hourly meteorological parameters for any day of a month for each month of the year from ten years of available data between 1987 and 1996. The model was validated for both annual average and short-term releases. For the annual average releases, the predictions of ground-level concentrations of 131I from the model were within a factor of 2 of measured field data. For two of the three sets of available weekly data, the measurements fell within the 95% subjective confidence interval of model predictions. Predictions of ground-level air concentrations were made on an annual average basis for the entire period of release and on a short-term, episodic basis for a 1954 accident.     

An assessment of the cord blood:maternal blood methylmercury ratio: implications for risk assessment

In the current U.S. Environmental Protection Agency reference dose (RfD) for methylmercury, the one-compartment pharmacokinetic model is used to convert fetal cord blood mercury (Hg) concentration to a maternal intake dose. This requires a ratio relating cord blood Hg concentration to maternal blood Hg concentration. No formal analysis of either the central tendency or variability of this ratio has been done. This variability contributes to the overall variability in the dose estimate. A ratio of 1.0 is implicitly used in the model, but an uncertainty factor adjustment is applied to the central tendency estimate of dose to address variability in that estimate. Thus, incorporation of the cord:maternal ratio and its variability into the estimate of intake dose could result in a significant change in the value of the RfD. We analyzed studies providing data on the cord:maternal blood Hg ratio and conducted a Monte Carlo-based meta-analysis of 10 studies meeting all inclusion criteria to generate a comprehensive estimate of the central tendency and variability of the ratio. This analysis results in a recommended central tendency estimate of 1.7, a coefficient of variation of 0.56, and a 95th percentile of 3.4. By analogy to the impact of the similar hair:blood Hg ratio on the overall variability in the dose estimate, incorporation of the cord:maternal ratio may support a 3-fold uncertainty factor adjustment to the central tendency estimate of dose to account for pharmacokinetic variability. Whether the information generated in this analysis is sufficient to warrant a revision to the RfD will depend on the outcome of a comprehensive reanalysis of the entire one-compartment model. We are currently engaged in such an analysis.     

Test of CAP88-PC's predicted concentrations of tritium in air at Lawrence Livermore National Laboratory

Based on annual tritium release rates from the five sources of tritium at Lawrence Livermore National Laboratory and the Tritium Research Laboratory at Sandia National Laboratory, the regulatory dispersion and dose model, CAP88-PC, was used to predict tritium concentrations in air at perimeter and offsite air surveillance monitoring locations for 1986 through 2001. These predictions were compared with mean annual measured concentrations, based on biweekly sampling. Deterministic predictions were compared with deterministic observations using predicted-to-observed ratios. In addition, the uncertainty on observations and predictions was assessed: when the uncertainty bounds of the observations overlapped with the uncertainty bounds of the predictions, the predictions were assumed to agree with the observations with high probability. Deterministically, 54% of all predictions were higher than the observations, and 96% fell within a factor of three. Accounting for uncertainty, 75% of all predictions agreed with the observations; 87% of the predictions either matched or exceeded the observations. Predictions equaled or exceeded observations at those sampling locations towards which the wind blows most frequently, except those in the hills. Under-predictions were seen at locations towards which the wind blows infrequently when released tritium was from elevated sources. When a high fraction of tritium was from area (diffuse) sources, predictions matched observations.     

非线性因子分析模型参数估计研究

目的研究非线性因子分析模型的参数估计方法.方法利用MCECM算法对模型参数进行极大似然估计,利用Louis公式计算标准误.结果统计模拟结果显示估计值与真实值比较吻合,误差较小.结论基于MCECM算法的极大似然估计方法可用于估计非线性因子分析模型的参数.     

Uncertainty analysis of doses from inhalation of depleted uranium

Measurements of uranium excreted in urine have been widely used to monitor possible exposures to depleted uranium (DU). This paper describes a comprehensive probabilistic uncertainty analysis of doses determined retrospectively from measurements of DU in urine. Parametric uncertainties in the International Commission on Radiological Protection (ICRP) Human Respiratory Tract Model (HRTM) and ICRP systemic model for uranium were considered in the analysis, together with uncertainties in an alternative model for particle removal from the lungs. Probability distributions were assigned to HRTM parameters based on uncertainties documented in ICRP Publication 66 and elsewhere, including the Capstone study of aerosols produced after DU penetrator impacts. Uncertainties in the uranium systemic model were restricted to transfer rates having the greatest effect on urinary excretion, and hence retrospective dose assessments, over the measurement times considered (10-10(4) d). The overall uncertainty on dose (the ratio of the upper and lower quantiles, q0.975/q0.025) was estimated to be about a factor of 50 at 10 days after intake and about a factor of 10 at 10(3)-10(4) d. The dose to the lung dominated the committed effective dose, with the lung absorption parameters, particularly the slow dissolution rate, ss, dominating the overall uncertainty. The median dose determined from a measurement of 1 ng DU, collected in urine in a 24-h period, varied from 0.1 microSv at 10 d to about 1 mSv at 10(4) d. Despite the large uncertainties, the upper q0.975 quantile for the assessed dose was below 1 mSv up to 5,000 d.     

Markov chain Monte Carlo estimation of a multiparameter decision model: consistency of evidence and the accurate assessment of uncertainty.

Decision models are usually populated 1 parameter at a time, with 1 item of information informing each parameter. Often, however, data may not be available on the parameters themselves but on several functions of parameters, and there may be more items of information than there are parameters to be estimated. The authors show how in these circumstances all the model parameters can be estimated simultaneously using Bayesian Markov chain Monte Carlo methods. Consistency of the information and/or the adequacy of the model can also be assessed within this framework. Statistical evidence synthesis using all available data should result in more precise estimates of parameters and functions of parameters, and is compatible with the emphasis currently placed on systematic use of evidence. To illustrate this, WinBUGS software is used to estimate a simple 9-parameter model of the epidemiology of HIV in women attending prenatal clinics, using information on 12 functions of parameters, and to thereby compute the expected net benefit of 2 alternative prenatal testing strategies, universal testing and targeted testing of high-risk groups. The authors demonstrate improved precision of estimates, and lower estimates of the expected value of perfect information, resulting from the use of all available data.     

Influences of parameter uncertainties within the ICRP-66 respiratory tract model: a parameter sensitivity analysis

An important aspect in model uncertainty analysis is the evaluation of input parameter sensitivities with respect to model outcomes. In previous publications, parameter uncertainties were examined for the ICRP-66 respiratory tract model. The studies were aided by the development and use of a computer code LUDUC (Lung Dose Uncertainty Code) which allows probabilities density functions to be specified for all ICRP-66 model input parameters. These density functions are sampled using Latin hypercube techniques with values subsequently propagated through the ICRP-66 model. In the present study, LUDUC has been used to perform a detailed parameter sensitivity analysis of the ICRP-66 model using input parameter density functions specified in previously published articles. The results suggest that most of the variability in the dose to a given target region is explained by only a few input parameters. For example, for particle diameters between 0.1 and 50 microm, about 50% of the variability in the total lung dose (weighted sum of target tissue doses) for 239PuO2 is due to variability in the dose to the alveolar-interstitial (AI) region. In turn, almost 90% of the variability in the dose to the AI region is attributable to uncertainties in only four parameters in the model: the ventilation rate, the AI deposition fraction, the clearance rate constant for slow-phase absorption of deposited material to the blood, and the clearance rate constant for particle transport from the AI2 to bb1 compartment. A general conclusion is that many input parameters do not significantly influence variability in final doses. As a result, future research can focus on improving density functions for those input variables that contribute the most to variability in final dose values.     

Probability modeling applied to CAD systems for mammography

A practical model based on basic probability theory is developed to evaluate the operational and financial performance of mammography systems. The model is intended to be used by decision makers to evaluate overall sensitivity, overall specificity, positive and negative predictive values, and expected cost. As an illustration, computer aided detection (CAD) systems that support a radiologist's diagnosis are compared with standard mammography to determine conditions that would support their use. The model's input parameters include the operational performance of mammography (with and without CAD), the age of the patient, the cost of administering the mammogram and the expected costs associated with false positive and false negative outcomes. Sensitivity analyses are presented that show the CAD system projecting financial benefit over ranges of uncertainty associated with each model parameter.     

Regression approaches to derive generic and fish group-specific probability density functions of bioconcentration factors for metals

In the framework of environmental multimedia modeling studies dedicated to environmental and health risk assessments of chemicals, the bioconcentration factor (BCF) is a parameter commonly used, especially for fish. As for neutral lipophilic substances, it is assumed that BCF is independent of exposure levels of the substances. However, for metals some studies found the inverse relationship between BCF values and aquatic exposure concentrations for various aquatic species and metals, and also high variability in BCF data. To deal with the factors determining BCF for metals, we conducted regression analyses to evaluate the inverse relationships and introduce the concept of probability density function (PDF) for Cd, Cu, Zn, Pb, and As. In the present study, for building the regression model and derive the PDF of fish BCF, two statistical approaches are applied: ordinary regression analysis to estimate a regression model that does not consider the variation in data across different fish family groups; and hierarchical Bayesian regression analysis to estimate fish group-specific regression models. The results show that the BCF ranges and PDFs estimated for metals by both statistical approaches have less uncertainty than the variation of collected BCF data (the uncertainty is reduced by 9%-61%), and thus such PDFs proved to be useful to obtain accurate model predictions for environmental and health risk assessment concerning metals.     

Diagnostic thyroid procedures and corresponding radiation doses in Manitoba: 1981-1985

Data on nuclear medicine thyroid examinations performed in Manitoba (population 1 million) from 1981-1985 were collected, with more detailed demographic data obtained on 1,100 consecutive patients between June 1987 and January 1988. An average of 2,081 patients were examined per year, 81% female and 19% male, representing 8.4% of all nuclear medicine procedures. Typical administered activity and associated HE per patient were 238.0 MBq and 1.5 mSv for 99mTc, 7.4 MBq and 1.2 mSv for 123I, and 0.33 MBq and 3.9 mSv for 131I. Based on NCRP risk estimates with explicit corrections for age, sex, and radionuclide used, it is estimated that the rate of thyroid cancer induction is unlikely to exceed 0.56 y-1, of which about 10% would be fatal. This estimate is about a factor of 4 less than that generated using more generally applicable radiation protection risk estimates averaged over both sexes and all ages in the general population. The replacement of 131I with the present mix of radiotracers used for thyroid evaluation has resulted in a reduction of the estimated population detriment by a factor of 3.6.     

Credibility of Chernobyl thyroid doses exceeding 10 Gy based on in-vivo measurements of 131I in Belarus

Many estimates of individual thyroid doses to children and adults in Belarus have been based on the results of direct thyroid measurements made using survey meters soon after the Chernobyl accident in 1986. Thyroid doses from internal exposure to 131I that are estimated using such measurements are usually considered to be better than estimates obtained by environmental transport modeling of concentrations expected in milk. Nonetheless, some of the estimated doses, primarily those to children, were high enough to raise questions about their credibility. Questions about high thyroid doses, taken here to be those exceeding 10 Gy, identified the need for further analysis, which is reported in this article. The overall conclusion is that the initial dose estimates exceeding 10 Gy based on direct thyroid measurements in Belarus are credible estimates and not mistakes. While the possibility of copying and data entry errors cannot be completely ruled out, the consistency of multiple measurements for many individuals supports the high dose estimates.     

A revised probabilistic estimate of the maternal methyl mercury intake dose corresponding to a measured cord blood mercury concentration

In 2001, the U.S. Environmental Protection Agency (EPA) adopted a revised reference dose (RfD) for methyl mercury (MeHg) of 0.1 microg/kg/day. The RfD is based on neurologic developmental effects measured in children associated with exposure in utero to MeHg from the maternal diet. The RfD derivation proceeded from a point of departure based on measured concentration of mercury in fetal cord blood (micrograms per liter). The RfD, however, is a maternal dose (micrograms per kilogram per day). Reconstruction of the maternal dose corresponding to this cord blood concentration, including the variability around this estimate, is a critical step in the RfD derivation. The dose reconstruction employed by the U.S. EPA using the one-compartment pharmacokinetic model contains two areas of significant uncertainty: It does not directly account for the influence of the ratio of cord blood: maternal blood Hg concentration, and it does not resolve uncertainty regarding the most appropriate central tendency estimates for pregnancy and third-trimester-specific model parameters. A probabilistic reassessment of this dose reconstruction was undertaken to address these areas of uncertainty and generally to reconsider the specification of model input parameters. On the basis of a thorough review of the literature and recalculation of the one-compartment model including sensitivity analyses, I estimated that the 95th and 99th percentiles (i.e., the lower 5th and 1st percentiles) of the maternal intake dose corresponding to a fetal cord blood Hg concentration of 58 microg/L are 0.3 and 0.2 microg/kg/day, respectively. For the 99th percentile, this is half the value previously estimated by the U.S. EPA.     

Estimating uncertainty ranges for costs by the bootstrap procedure combined with probabilistic sensitivity analysis.

When an economic evaluation incorporates patient-level data, there are two types of uncertainty over the results: uncertainty due to variation in the sampled data, and uncertainty over the choice of modelling parameters and assumptions. Previously statistical methods have been used to estimate the extent of the former, and sensitivity analysis to estimate the extent of the latter. Ideally interval estimates for economic variables should reflect both types of uncertainty. This paper describes a method for combining bootstrapping with probabilistic sensitivity analysis to estimate a total 'uncertainty range' for incremental costs. The approach is illustrated using cost data from a randomized controlled trial of endoscopy for Helicobactor pylori negative young dyspeptic patients. The trial failed to demonstrate any clinical benefit from endoscopy, which was on average pound 395 more costly. The combined 95% uncertainty range for incremental costs (-pound 236 to pound 931) was wider than 95% intervals estimated by either probabilistic sensitivity analysis (pound 43 to pound 592) or the non-parametric bootstrap method (-pound 95 to pound 667) alone. The method can easily be extended to the calculation of uncertainty ranges for incremental cost-effectiveness ratios.