Colorado Plateau Uranium Miners Research Articles

General Relative Rate Models for the Analysis of Studies Using Case-Cohort Designs.

A standard approach to analysis of case-cohort data involves fitting log-linear models. In this paper, we describe how standard statistical software can be used to fit a broad class of general relative rate models to case-cohort data and derive confidence intervals. We focus on a case-cohort design in which a roster has been assembled and events ascertained but additional information needs to be collected on explanatory variables. The additional information is ascertained just for persons who experience the event of interest and for a sample of the cohort members enumerated at study entry. One appeal of such a case-cohort design is that this sample of the cohort may be used to support analyses of several outcomes. The ability to fit general relative rate models to case-cohort data may allow an investigator to reduce model misspecification in exposure-response analyses, fit models in which some factors have effects that are additive and others multiplicative, and facilitate estimation of relative excess risk due to interaction. We address model fitting for simple random sampling study designs as well as stratified designs. Data on lung cancer among radon-exposed men (Colorado Plateau uranium miners, 1950-1990) are used to illustrate these methods.

Using the entire history in the analysis of nested case cohort samples

Countermatching designs can provide more efficient estimates than simple matching or case-cohort designs in certain situations such as when good surrogate variables for an exposure of interest are available. We extend pseudolikelihood estimation for the Cox model under countermatching designs to models where time-varying covariates are considered. We also implement pseudolikelihood with calibrated weights to improve efficiency in nested case-control designs in the presence of time-varying variables. A simulation study is carried out, which considers four different scenarios including a binary time-dependent variable, a continuous time-dependent variable, and the case including interactions in each. Simulation results show that pseudolikelihood with calibrated weights under countermatching offers large gains in efficiency if compared to case-cohort. Pseudolikelihood with calibrated weights yielded more efficient estimators than pseudolikelihood estimators. Additionally, estimators were more efficient under countermatching than under case-cohort for the situations considered. The methods are illustrated using the Colorado Plateau uranium miners cohort. Furthermore, we present a general method to generate survival times with time-varying covariates. Copyright © 2016 John Wiley & Sons, Ltd.

Healthy worker survivor bias in the Colorado Plateau uranium miners cohort.

Cohort mortality studies of underground miners have been used to estimate the number of lung cancer deaths attributable to radon exposure. However, previous studies of the radon-lung cancer association among underground miners may have been subject to healthy worker survivor bias, a type of time-varying confounding by employment status. We examined radon-mortality associations in a study of 4,124 male uranium miners from the Colorado Plateau who were followed from 1950 through 2005. We estimated the time ratio (relative change in median survival time) per 100 working level months (radon exposure averaging 130,000 mega-electron volts of potential α energy per liter of air, per working month) using G-estimation of structural nested models. After controlling for healthy worker survivor bias, the time ratio for lung cancer per 100 working level months was 1.168 (95% confidence interval: 1.152, 1.174). In an unadjusted model, the estimate was 1.102 (95% confidence interval: 1.099, 1.112)-39% lower. Controlling for this bias, we estimated that among 617 lung cancer deaths, 6,071 person-years of life were lost due to occupational radon exposure during follow-up. Our analysis suggests that healthy worker survivor bias in miner cohort studies can be substantial, warranting reexamination of current estimates of radon's estimated impact on lung cancer mortality.

Healthy Worker Survivor Bias in the Colorado Plateau Uranium Miners Cohort

Healthy Worker Survivor Bias in the Colorado Plateau Uranium Miners Cohort

Occupational radon exposure and lung cancer mortality: estimating intervention effects using the parametric g-formula.

Traditional regression analysis techniques used to estimate associations between occupational radon exposure and lung cancer focus on estimating the effect of cumulative radon exposure on lung cancer. In contrast, public health interventions are typically based on regulating radon concentration rather than workers' cumulative exposure. Estimating the effect of cumulative occupational exposure on lung cancer may be difficult in situations vulnerable to the healthy worker survivor bias. Workers in the Colorado Plateau Uranium Miners cohort (n = 4,134) entered the study between 1950 and 1964 and were followed for lung cancer mortality through 2005. We use the parametric g-formula to compare the observed lung cancer mortality to the potential lung cancer mortality had each of 3 policies to limit monthly radon exposure been in place throughout follow-up. There were 617 lung cancer deaths over 135,275 person-years of follow-up. With no intervention on radon exposure, estimated lung cancer mortality by age 90 was 16%. Lung cancer mortality was reduced for all interventions considered, and larger reductions in lung cancer mortality were seen for interventions with lower monthly radon exposure limits. The most stringent guideline, the Mine Safety and Health Administration standard of 0.33 working-level months, reduced lung cancer mortality from 16% to 10% (risk ratio = 0.67 [95% confidence interval = 0.61 to 0.73]). This work illustrates the utility of the parametric g-formula for estimating the effects of policies regarding occupational exposures, particularly in situations vulnerable to the healthy worker survivor bias.

Correcting for Healthy-Worker Survivor Bias of the Radon-lung Cancer Dose-response in a Cohort of Uranium Miners using Structural Nested Accelerated Failure Time Models

Radon is estimated to cause over 20,000 lung cancer deaths per year in the United States. Occupational studies have been used to estimate population risks. These studies may be subject to healthy worker survivor bias, though this potential bias has not been thoroughly explored. Healthy worker survivor bias can be conceptualized as time-varying confounding by employment status. This confounding is problematic for regression models when radon exposure influences employment status. We estimate the time ratio (the relative decrease in median survival per 100 working level months, a unit of occupational radon exposure) using data from the Colorado Plateau uranium miners cohort. We estimated the association between radon and lung cancer, with and without control for the healthy-worker survivor bias using G-estimation of a structural nested accelerated failure time model. In a structural nested model not adjusted for healthy worker survivor bias, we estimate a time ratio (95% confidence intervals) per 100 working level months of 1.055 (1.055, 1.055). When we control this bias, the estimate is 1.168 (1.152, 1.174). We found that cumulative radon exposure was associated with leaving work, consistent with employment status being a confounder and intermediate variable, such that standard regression methods are biased. These findings suggest that healthy worker survivor bias leads to an underestimate of the effects of radon. Further, standard analyses of the radon-lung cancer dose-response cannot adequately control this bias, warranting re-examination of the current estimates of radon's carcinogenic effects.

0216 Controlling healthy worker survivor bias of the radon-lung cancer dose-response in a cohort of uranium miners

ObjectivesRadon is a leading cause of lung cancer and is estimated to cause nearly 20 000 deaths per year in the United States. Studies of underground miners, who often were...

Efficient estimation of the censored linear regression model

In linear regression or accelerated failure time models, complications in efficient estimation arise from the multiple roots of the efficient score and density estimation. This paper proposes a one-step efficient estimation method based on a counting process martingale, which has several advantages: it avoids the multiple-root problem, the initial estimator is easily available and the variance estimator can be obtained by employing plug-in rules. A simple and effective data-driven bandwidth selector is provided. The proposed estimator is proved to be semiparametric efficient, with the same asymptotic variance as the efficient estimator when the error distribution is known up to a location shift. Numerical studies with supportive evidence are presented. The proposal is applied to the Colorado Plateau uranium miners data. Copyright 2013, Oxford University Press.

Analysis of cohort survival data with transformation model

Existing methods for analysis of survival data arising from cohort sam- pling are largely based on Cox's model and pertained to a certain type of sampling design. This paper applies the general linear transformation model, which includes Cox's model and proportional odds model as special cases, to a class of sampling designs including nested case-control, case-cohort and classical case-control designs. A simple likelihood-based method is developed, and the resulting estimator of the regression coefficient is shown to be consistent and asymptotic normal. The compu- tation and inference procedures are straightforward. In addition to the simplicity and generality of the method, it also has minimal loss of efficiency as the observa- tions with missing covariates that are not used contain little information about the regression parameter. The proposed estimation performs well in simulation studies and is applied to analyze the Colorado Plateau uranium miners cohort data.

Fitting General Relative Risk Models for Survival Time and Matched Case-Control Analysis

Cox proportional hazards regression analysis of survival data and conditional logistic regression analysis of matched case-control data are methods that are widely used by epidemiologists. Standard statistical software packages accommodate only log-linear model forms, which imply exponential exposure-response functions and multiplicative interactions. In this paper, the authors describe methods for fitting non-log-linear Cox and conditional logistic regression models. The authors use data from a study of lung cancer mortality among Colorado Plateau uranium miners (1950-1982) to illustrate these methods for fitting general relative risk models to matched case-control control data, countermatched data with weights, d:m matching, and full cohort Cox regression using the SAS statistical package (SAS Institute Inc., Cary, North Carolina).

Are Nested Case-Control Studies Biased?

It has been recently asserted that the nested case-control study design, in which case-control sets are sampled from cohort risk sets, can introduce bias ("study design bias") when there are lagged exposures. The bases for this claim include a theoretical and an "empirical evaluation" argument. We examined both of these arguments and found them to be incorrect. We describe an appropriate empirical evaluation method to explore the performance of nested case-control study designs and analysis methods from an existing cohort. This empirical evaluation approach relies on simulating case-control outcomes from risk sets in the cohort from which the case-control study is to be performed. Because it is based on the underlying cohort structure, the empirical evaluation can provide an assessment that is tailored to the specific characteristics of the study under consideration. The methods are illustrated using samples from the Colorado Plateau uranium miners cohort.

Risks Associated with Low Doses and Low Dose Rates of Ionizing Radiation: Why Linearity May Be (Almost) the Best We Can Do

Deterministic and stochastic effects associated with high-dose ionizing radiation (x-ray) exposure have been known for almost as long as ionizing radiation itself (1–3). At lower doses, radiation risks are primarily stochastic effects, in particular, somatic effects (cancer) rather than the deterministic effects characteristic of higher-dose exposure (4–6). In contrast to deterministic effects, for stochastic effects, scientific committees generally assume that at sufficiently low doses there is a positive linear component to the dose response—that is, that there is no threshold (4–6). This does not preclude there being higher-order (eg, quadratic) powers of dose in the dose response that may be of importance at higher doses. It is on this basis that models linear (or linear-quadratic) in dose are often used to extrapolate the experience of the Japanese atomic bomb survivors (who were typically exposed at a high dose rate to moderate doses [average, 0.1 Sv]) to estimate risks from low doses and low dose rates (4–6). Most population-based cancer risk estimates are based primarily on the Japanese atomic bomb survivor Life Span Study (LSS) cohort data (4–6). However, evidence of excess risks comes from a large number of other studies as well. In the parallel editorial (7), evidence is presented for possible real (or at least “practical”) thresholds or “hormetic” (beneficial) effects of low doses of ionizing radiation. As we summarize here, and in contrast to the arguments of Tubiana et al (7), we judge that there is little epidemiologic or biologic evidence for these for cancer. The arguments are of three forms: (a) assessment of the degree of curvature in the cancer dose response within the Japanese atomic bomb survivors and other exposed groups (in particular, departure from linear or linear-quadratic curvature), (b) consistency of risks between the Japanese and other moderate- and low-dose cohorts, and (c) assessment of biologic data on mechanisms. Most of the information on radiation-induced cancer risk comes from (a) the Japanese atomic bomb survivors, (b) medically exposed populations, (c) occupationally exposed groups, and (d) environmentally exposed groups (6). In the higher-dose radiation therapy studies, where doses received are very much higher than in the LSS, sometimes in the range at which cell sterilization occurs, excess cancer risks per unit dose tend to be less than in comparable subsets of the LSS (8,9). However, as we show, risks in moderate- and low-dose medically and occupationally exposed groups are generally consistent with those in the LSS.

Using tensor product splines in modeling exposure-time-response relationships: application to the Colorado Plateau Uranium Miners cohort.

An adequate depiction of exposure-time-response relationships is important in assessing public health implications of an occupational or environmental exposure. Recent advances have focused on flexible modeling of the overall shape of latency. Methods are needed to allow for varying shapes of latency under different exposure profiles. A tensor product spline model is proposed for describing exposure-response relationships for protracted time-dependent occupational exposure histories in epidemiologic studies. The methods use flexible multi-dimensional techniques to jointly model age, latency and exposure-response effects. In analyzing data from the Colorado Plateau Uranium Miners cohort, a model that allows for varying exposure-dependent latency shapes is found to be superior to models that only allowed for an overall latency curve. Specifically, the model suggests that, at low exposure levels risk increased at short latencies followed by a slow decline for longer latency periods. On the other hand, risk was higher but did not change much by latency for higher exposure levels. The proposed methodology has the advantage of allowing for latency functions that vary by exposure levels and, conversely, exposure-response relationships that are influenced by the latency structure.

Fitting the two-stage model of carcinogenesis to nested case-control data on the Colorado Plateau uranium miners: dependence on data assumptions.

This paper investigates some problems with the Colorado Plateau uranium miners cohort when fitting the mechanistic model of carcinogenesis of Moolgavkar et al. (MVK model) to nested case-control data for lung cancers. The influence of data for hard rock mining and work histories on the model fitting is examined and found to be highly influential. The question of selecting the most appropriate number of controls per case is also considered. Analyses were carried out assuming that the hard rock mining exposure occurred prior to all other work histories and that miners received no exposure between work histories. The use of less than 15 controls per case was found to seriously restrict the quality of fit of the models. The best fitting, most reliable model contained linear effects of radon exposure on the first mutation rate, the rate of differentiation, and the rate of death of the intermediate cells. An effect of smoking on the growth of the intermediate cells was also included in the model. It is concluded that owing to the complexity of the MVK model and the limited amount of information in this dataset, the reliability of the Colorado Plateau dataset for fitting this type of model, particularly in a case-control format, is questionable.

Modelling lung tumour risk in radon-exposed uranium miners using generalizations of the two-mutation model of Moolgavkar, Venzon and Knudson

Purposes : To model radon-induced lung cancer in uranium miners using a quasi-biological model of carcinogenesis. Materials and methods : Fitting of generalizations of the stochastic two-mutation carcinogenesis model of Moolgavkar, Venzon and Knudson to a case-control dataset nested within the cohort and to the full cohort of lung cancer mortality in the Colorado Plateau uranium miners, taking account of exposure to cigarette smoke and to radon daughters. Results : Models with three mutations gave adequate descriptions of the time and age patterns of radon-daughter-induced excess lung tumour mortality. The overall fit of the two-mutation model to the case-control data was somewhat worse than that of the three-mutation model. For both the optimal two- and threemutation models radon daughters and cigarette smoke were assumed to act on the first mutation rate. In the optimal twomutation model, radon daughters also modified the intermediate cell death or differentiation rate. In the optimal three-mutation model, radon daughters modified the second mutation rate. In all models, the action of radon daughters and cigarette smoke was markedly non-linear, particularly in their action on the mutation rates. The optimal two- and three-mutation models fitted to the cohort data were of slightly different form to those fitted to the case-control data. The model fits to the cohort data are preferred to those to the case-control data on grounds of plausibility. Conclusions : Quasi-biological carcinogenesis models with three mutations give adequate descriptions of the time and age patterns of radon-daughter-induced excess lung tumour mortality. The overall fit of the two-mutation model is somewhat worse than that of the three-mutation model.

Comparisons of lung tumour mortality risk in the Japanese A-bomb survivors and in the Colorado Plateau uranium miners: support for the ICRP lung model

Purposes : To estimate the ratio of risks for exposure to radon progeny relative to low-LET radiation based on human lung cancer data, taking account of possible time and age variations in radiation-induced lung cancer risk. Materials and methods : Fitting two sorts of time- and age-adjusted relative risk models to a case-control dataset nested within the Colorado Plateau uranium miner cohort and to the Japanese atomic (A)-bomb survivor mortality data. Results : If all A-bomb survivors are compared with the Colorado data, there are statistically significant (two-sided p < 0.05) differences between the two datasets in the pattern of the variation of relative risk with time after exposure, age at exposure and attained age. The excess relative risk decreases much faster with time, age at exposure and attained age in the Colorado uranium miners than in the Japanese A-bomb survivors. If only male A-bomb survivors are compared with the Colorado data, there are no longer statistically significant differences between the two datasets in the pattern of variation of relative risk with time after exposure, age at exposure or attained age. There are no statistically significant differences between the male and female A-bomb survivors in the speed of reduction of relative risk with time after exposure, age at exposure or attained age, although there are indications of rather faster reduction of relative risk with time and age among male survivors than among female survivors. The implicit risk conversion factor for exposure to radon progeny relative to the A-bomb radiation in the male survivors is 1.8 ×10-2 Sv WLM-1 (95% CI 6.1 ×10-3, 1.1 ×10-1) using a model with exponential adjustments for the effects of radiation for time since exposure and age at exposure, and 1.9 ×10-2 Sv WLM-1 (95% CI 6.2 ×10-3, 1.6 ×10-1) using a model with adjustments for the effects of radiation proportional to powers of time since exposure and attained age. Estimates of the risk conversion factor calculated using variant assumptions as to the definition of lung cancer in the Colorado data, or by excluding miners for whom exposure estimates may be less reliable, are very similar. The absence of information on cigarette smoking in the Japanese A-bomb survivors, and the possibility that this may confound the time trends in radiation-induced lung cancer risk in that cohort, imply that these findings should be interpreted with caution. Conclusions : There are no statistically significant differences between the male A-bomb survivors data and the Colorado miner data in the pattern of variation of relative risk with time after exposure and age at exposure. The risk conversion factor is very close to the value suggested by the latest ICRP lung model, albeit with substantial uncertainties.

Using splines to analyse latency in the Colorado Plateau uranium miners cohort.

Different approaches have been proposed to investigate latency in epidemiologic studies where detailed exposure histories are available. We demonstrate the application of a flexible, yet parsimonious, spline function model to investigate latency patterns for radon progeny exposure and lung cancer in the Colorado Plateau uranium miners cohort. The model extends a previously proposed bilinear model. The excess relative risk (ERR) reached a maximum of 0.6 per 100 working level months, for exposures received 14 years previously. The ERR then declined, and was estimated to approach zero for exposures received 35 years and more in the past. The point-wise 95% confidence intervals supported ERRs > 0 for the period 9-32 years before the event. The estimated latency curve was homogeneous across categories of attained age, duration of exposure, rate of exposure, and smoking. The proposed spline model is a flexible tool for latency analyses, and extends previously used methods.

Correcting for exposure measurement error in a reanalysis of lung cancer mortality for the Colorado Plateau Uranium Miners cohort.

The exposure estimates used to date for the analysis of lung cancer mortality in the Colorado Plateau Uranium Miners cohort were developed from radon progeny measurements taken in mines beginning in 1951. Since uranium miners were often exposed over long periods of time and since mines were not continuously monitored, much extrapolation and/or interpolation of measured dose-rates was needed in order to develop estimates of exposure for each of the miners in the cohort. We have recently re-examined the interpolation scheme used to create the histories in the light of the fit of a statistical model for the radon progeny measurements taken in mines within the Plateau, and we have computed revised exposure estimates for the large majority of miners in the cohort. This report describes the use of these new model-based revised exposure estimates in the analysis of lung cancer mortality, using follow-up data current through 1990. Specific issues addressed here are (1) the strength of the association between exposure and risk of lung cancer mortality; (2) effects of attained age and time since exposure upon risk of lung cancer mortality; and (3) exposure-rate effects upon risk. Results using the revised exposure estimates are compared to those obtained fitting the same models using the original Public Health Service (PHS) exposure estimates. We found evidence that the new exposure histories provide a better fit to the lung cancer mortality data than do the histories based upon the original PHS dose-rate estimates. In general, the new results show a stronger overall relationship (larger slope estimate) between lung cancer mortality and exposure per unit exposure compared to those obtained with the original estimates, while displaying similar age at exposure and time since exposure effects. In the reanalysis the impact of low dose-rate exposure is found to be relatively unchanged before and after exposure error correction, while the estimate of the effect of high dose-rate exposure is considerably increased. Even after applying our measurement error corrections, evidence of inverse dose-rate effects is found, since the estimate of the impact of high dose-rate exposure is still below that of the low dose-rates. The magnitude and statistical significance, however, of the dose-rate effect estimates are diminished when fit using the revised exposure estimates.

Latency analysis in epidemiologic studies of occupational exposures: Application to the Colorado plateau uranium miners cohort

Latency analysis in epidemiologic studies of occupational exposures: Application to the Colorado plateau uranium miners cohort

Latency analysis in epidemiologic studies of occupational exposures: application to the Colorado Plateau uranium miners cohort.

Latency effects are an important factor in assessing the public health implications of an occupational or environmental exposure. Usually, however, latency results as described in the literature are insufficient to answer public health related questions. Alternative approaches to the analysis of latency effects are warranted. A general statistical framework for modeling latency effects is described. We then propose bilinear and exponential decay latency models for analyzing latency effects as they have parameters that address questions of public health interest. Methods are described for fitting these models to cohort or case-control data; statistical inference is based on standard likelihood methods. A latency analysis of radon exposure and lung cancer in the Colorado Plateau uranium miners cohort was performed. We first analyzed the entire cohort and found that the relative risk associated with exposure increases for about 8.5 years and thereafter decreases until it reaches background levels after about 34 years. The hypothesis that the relative risk remains at its peak level is strongly rejected (P < 0.001). Next, we investigated the variation in the latency effects over subsets of the cohort based on attained age, level and rate of exposure, and smoking. Age was the only factor for which effect modification was demonstrated (P = 0.014). We found that the decline in effect is much steeper at older ages (60+ years) than younger. The proposed methods can provide much more information about the exposure-disease latency effects than those generally used.