Estimation of a Benchmark Dose in the Presence or Absence of Hormesis Using Posterior Averaging

U.S. Environment Protection Agency benchmark doses for dichotomous cancer responses are often estimated using a multistage model based on a monotonic dose‐response assumption. To account for model uncertainty in the estimation process, several model averaging methods have been proposed for risk assessment. In this article, we extend the usual parameter space in the multistage model for monotonicity to allow for the possibility of a hormetic dose‐response relationship. Bayesian model averaging is used to estimate the benchmark dose and to provide posterior probabilities for monotonicity versus hormesis. Simulation studies show that the newly proposed method provides robust point and interval estimation of a benchmark dose in the presence or absence of hormesis. We also apply the method to two data sets on carcinogenic response of rats to 2,3,7,8‐tetrachlorodibenzo‐p‐dioxin.     

Multistage Models of Carcinogenesis: An Approximation for the Size and Number Distribution of Late-Stage Clones

Multistage models have become the basic paradigm for modeling carcinogenesis. One model, the two-stage model of carcinogenesis, is now routinely used in the analysis of cancer risks from exposure to environmental chemicals. In its most general form, this model has two states, an initiated state and a neoplastic state, which allow for growth of cells via a simple linear birth-death process. In all analyses done with this model, researchers have assumed that tumor incidence is equivalent to the formation of a single neoplastic cell and the growth kinetics in the neoplastic state have been ignored. Some researchers have discussed the impact of this assumption on their analyses, but no formal methods were available for a more rigorous application of the birth-death process. In this paper, an approximation is introduced which allows for the application of growth kinetics in the neoplastic state. The adequacy of the approximation against simulated data is evaluated and methods are developed for implementing the approximation using data on the number and size of neoplastic clones.     

Could Removing Arsenic from Tobacco Smoke Significantly Reduce Smoker Risks of Lung Cancer?

If a specific biological mechanism could be determined by which a carcinogen increases lung cancer risk, how might this knowledge be used to improve risk assessment? To explore this issue, we assume (perhaps incorrectly) that arsenic in cigarette smoke increases lung cancer risk by hypermethylating the promoter region of gene p16INK4a, leading to a more rapid entry of altered (initiated) cells into a clonal expansion phase. The potential impact on lung cancer of removing arsenic is then quantified using a three‐stage version of a multistage clonal expansion (MSCE) model. This refines the usual two‐stage clonal expansion (TSCE) model of carcinogenesis by resolving its intermediate or “initiated” cell compartment into two subcompartments, representing experimentally observed “patch” and “field” cells. This refinement allows p16 methylation effects to be represented as speeding transitions of cells from the patch state to the clonally expanding field state. Given these assumptions, removing arsenic might greatly reduce the number of nonsmall cell lung cancer cells (NSCLCs) produced in smokers, by up to two‐thirds, depending on the fraction (between 0 and 1) of the smoking‐induced increase in the patch‐to‐field transition rate prevented if arsenic were removed. At present, this fraction is unknown (and could be as low as zero), but the possibility that it could be high (close to 1) cannot be ruled out without further data.     

Logistic Regression Models with Unspecified Low Dose–Response Relationships and Experimental Designs for Hormesis Studies

Hormesis refers to a nonmonotonic (biphasic) dose–response relationship in toxicology, environmental science, and related fields. In the presence of hormesis, a low dose of a toxic agent may have a lower risk than the risk at the control dose, and the risk may increase at high doses. When the sample size is small due to practical, logistic, and ethical considerations, a parametric model may provide an efficient approach to hypothesis testing at the cost of adopting a strong assumption, which is not guaranteed to be true. In this article, we first consider alternative parameterizations based on the traditional three‐parameter logistic regression. The new parameterizations attempt to provide robustness to model misspecification by allowing an unspecified dose–response relationship between the control dose and the first nonzero experimental dose. We then consider experimental designs including the uniform design (the same sample size per dose group) and the $c$‐optimal design (minimizing the standard error of an estimator for a parameter of interest). Our simulation studies showed that (1) the $c$‐optimal design under the traditional three‐parameter logistic regression does not help reducing an inflated Type I error rate due to model misspecification, (2) it is helpful under the new parameterization with three parameters (Type I error rate is close to a fixed significance level), and (3) the new parameterization with four parameters and the $c$‐optimal design does not reduce statistical power much while preserving the Type I error rate at a fixed significance level.     

Optimal Experimental Design Strategies for Detecting Hormesis

Hormesis is a widely observed phenomenon in many branches of life sciences, ranging from toxicology studies to agronomy, with obvious public health and risk assessment implications. We address optimal experimental design strategies for determining the presence of hormesis in a controlled environment using the recently proposed Hunt‐Bowman model. We propose alternative models that have an implicit hormetic threshold, discuss their advantages over current models, and construct and study properties of optimal designs for (i) estimating model parameters, (ii) estimating the threshold dose, and (iii) testing for the presence of hormesis. We also determine maximin optimal designs that maximize the minimum of the design efficiencies when we have multiple design criteria or there is model uncertainty where we have a few plausible models of interest. We apply these optimal design strategies to a teratology study and show that the proposed designs outperform the implemented design by a wide margin for many situations.     

Two-Event Model for Carcinogenesis: Biological, Mathematical, and Statistical Considerations

A two-mutation model for carcinogenesis is reviewed. General principles in fitting the model to epidemiologic and experimental data are discussed, and some examples are given. A general solution to the model with time-dependent parameters is developed, and its use is illustrated by application to data from an experiment in which rats exposed to radon developed lung tumors.     

Extending the Stochastic Two-Stage Model of Carcinogenesis to Include Self-Regulation of the Nonmalignant Cell Population

One of the challenges of introducing greater biological realism into stochastic models of cancer induction is to find a way to represent the homeostatic control of the normal cell population over its own size without complicating the analysis too much to obtain useful results. Current two-stage models of carcinogenesis typically ignore homeostatic control. Instead, a deterministic growth path is specified for the population of "normal" cells, while the population of "initiated" cells is assumed to grow randomly according to a birth-death process with random immigrations from the normal population. This paper introduces a simple model of homeostatically controlled cell division for mature tissues, in which the size of the nonmalignant population remains essentially constant over time. Growth of the nonmalignant cell population (normal and initiated cells) is restricted by allowing cells to divide only to fill the "openings" left by cells that die or differentiate, thus maintaining the constant size of the nonmalignant cell population. The fundamental technical insight from this model is that random walks, rather than birth-and-death processes, are the appropriate stochastic processes for describing the kinetics of the initiated cell population. Qualitative and analytic results are presented, drawn from the mathematical theories of random walks and diffusion processes, that describe the probability of spontaneous extinction and the size distribution of surviving initiated populations when the death/differentiation rates of normal and initiated cells are known. The constraint that the nonmalignant population size must remain approximately constant leads to much simpler analytic formulas and approximations, flowing directly from random walk theory, than in previous birth-death models.(ABSTRACT TRUNCATED AT 250 WORDS)     

Using Average Lifetime Dose Rate for Intermittent Exposures to Carcinogens

The effect of using the average dose rate over a lifetime as a representative measure of exposure to carcinogens is investigated by comparing the true theoretical multistage intermittent-dosing lifetime low-dose excess risk to the theoretical multistage continuous-dosing lifetime risk corresponding to the average lifetime dose rate. It is concluded that low-dose risk estimates based on the average lifetime dose rate may overestimate the true risk by several orders of magnitude, but that they never underestimate the true risk by more than a factor of k/r, where k is the total number of stages in the multistage model and r is the number of stages that are dose-related.     

A Nonidentifiability Aspect of the Two-Stage Model of Carcinogenesis

This paper discusses identifiability of the two-stage birth-death-mutation model of carcinogenesis. It is shown that the homogeneous version of the model is nonidentifiable; the same is all the more evident for its nonhomogeneous versions. This result implies that the model parameters cannot be uniquely estimated from time-to-tumor observations.     

Calculating the Hazard Function and Probability of Tumor for Cancer Risk Assessment When the Parameters Are Time-Dependent

The probability of tumor and hazard function are calculated in a stochastic two-stage model for carcinogenesis when the parameters of the mode are time-dependent. The method used is called the method of characteristics.     

The Multistage Model with a Time-Dependent Dose Pattern: Applications to Carcinogenic Risk Assessment1

A cancer risk assessment methodology based upon the Armitage–Doll multistage model of cancer is applied to animal bioassay data. The method utilizes the exact time-dependent dose pattern used in a bioassay rather than some single measure of dose such as average dose rate or cumulative dose. The methodology can be used to predict risks from arbitrary exposure patterns including, for example, intermittent exposure and short-term exposure occurring at an arbitrary age. The methodology is illustrated by applying it to a National Cancer Institute bioassay of ethylene dibromide in which dose rates were modified several times during the course of the experiment.     

An Exact Analysis of the Multistage Model Explaining Dose-Response Concavity

The traditional multistage (MS) model of carcinogenesis implies several empirically testable properties for dose-response functions. These include convex (linear or upward-curving) cumulative hazards as a function of dose; symmetric effects on lifetime tumor probability of transition rates at different stages; cumulative hazard functions that increase without bound as stage-specific transition rates increase without bound; and identical tumor probabilities for individuals with identical parameters and exposures. However, for at least some chemicals, cumulative hazards are not convex functions of dose. This paper shows that none of these predicted properties is implied by the mechanistic assumptions of the MS model itself. Instead, they arise from the simplifying "rare-tumor" approximations made in the usual mathematical analysis of the model. An alternative exact probabilistic analysis of the MS model with only two stages is presented, both for the usual case where a carcinogen acts on both stages simultaneously, and also for idealized initiation-promotion experiments in which one stage at a time is affected. The exact two-stage model successfully fits bioassay data for chemicals (e.g., 1,3-butadiene) with concave cumulative hazard functions that are not well-described by the traditional MS model. Qualitative properties of the exact two-stage model are described and illustrated by least-squares fits to several real datasets. The major contribution is to show that properties of the traditional MS model family that appear to be inconsistent with empirical data for some chemicals can be explained easily if an exact, rather than an approximate model, is used. This suggests that it may be worth using the exact model in cases where tumor rates are not negligible (e.g., in which they exceed 10%). This includes the majority of bioassay experiments currently being performed.     

Dose-Response Analysis Using Spreadsheets

The task of fitting dose-response models to experimental data can be performed using a spreadsheet with a built-in optimization engine. This paper shows how the task of point and interval estimation can be performed using Microsoft EXCEL. A case study is presented on the carcinogenic dose-response behavior of chloroform.     

Multistage Modeling of Lung Cancer Mortality Among Arsenic-Exposed Copper-Smelter Workers

Multistage modeling incorporating a time-dependent exposure pattern is applied to lung cancer mortality data obtained from a cohort of 2802 arsenic-exposed copper-smelter workers who worked 1 or more years during the period 1940-1964 at a copper smelter at Tacoma, Washington. The workers were followed for death through 1976. There were 100 deaths due to lung cancer during the follow-up period. Exposures to air arsenic levels measured in micrograms/m3 were estimated from departmental air arsenic and workers urinary arsenic measurements. Relationships of different temporal variables with excess death rates are examined to judge qualitatively the implications of the multistage cancer process. Analysis to date indicates a late stage effect of arsenic although an additional early stage effect cannot be ruled out.     

Estimation of the Joint Risk from Multiple-Compound Exposure Based on Single-Compound Experiments

In the absence of data from multiple-compound exposure experiments, the health risk from exposure to a mixture of chemical carcinogens is generally based on the results of the individual single-compound experiments. A procedure to obtain an upper confidence limit on the total risk is proposed under the assumption that total risk for the mixture is additive. It is shown that the current practice of simply summing the individual upper-confidence-limit risk estimates as the upper-confidence-limit estimate on the total excess risk of the mixture may overestimate the true upper bound. In general, if the individual upper-confidence-limit risk estimates are on the same order of magnitude, the proposed method gives a smaller upper-confidence-limit risk estimate than the estimate based on summing the individual upper-confidence-limit estimates; the difference increases as the number of carcinogenic components increases.     

Parameters of a Dose‐Response Model Are on the Boundary: What Happens with BMDL?

It is well known that, under appropriate regularity conditions, the asymptotic distribution for the likelihood ratio statistic is χ². This result is used in EPA's benchmark dose software to obtain a lower confidence bound (BMDL) for the benchmark dose (BMD) by the profile likelihood method. Recently, based on work by Self and Liang, it has been demonstrated that the asymptotic distribution of the likelihood ratio remains the same if some of the regularity conditions are violated, that is, when true values of some nuisance parameters are on the boundary. That is often the situation for BMD analysis of cancer bioassay data. In this article, we study by simulation the coverage of one- and two-sided confidence intervals for BMD when some of the model parameters have true values on the boundary of a parameter space. Fortunately, because two-sided confidence intervals (size 1–2α) have coverage close to the nominal level when there are 50 animals in each group, the coverage of nominal 1−α one-sided intervals is bounded between roughly 1–2α and 1. In many of the simulation scenarios with a nominal one-sided confidence level of 95%, that is, α= 0.05, coverage of the BMDL was close to 1, but for some scenarios coverage was close to 90%, both for a group size of 50 animals and asymptotically (group size 100,000). Another important observation is that when the true parameter is below the boundary, as with the shape parameter of a log-logistic model, the coverage of BMDL in a constrained model (a case of model misspecification not uncommon in BMDS analyses) may be very small and even approach 0 asymptotically. We also discuss that whenever profile likelihood is used for one-sided tests, the Self and Liang methodology is needed to derive the correct asymptotic distribution.     

不允许卖空情况下M-VaR和M-SA投资组合比较研究

为了验证投资组合理论在中国证券市场的有效性,针对不允许卖空情况,文章分别研究了均值-VaR(M- VaR)和均值-半绝对偏差(M-SA)投资组合模型,并分别结合序列二次规划法和不等式组的旋转算法以及线性规划的旋转算法进行求解。文章选取1998-2000年沪市六只业绩比较好的股票,依据1998-1999年的数据作为样本数据,分别求出两个模型在不同期望收益率下的最优投资策略,将得出的最优投资策略应用到2000年,进行模拟投资,从而计算出各模型的总收益率。以等比例投资为标准,比较两个模型的绩效。最后,证明了两个模型对于中国证券市场是适用。     

Comparison of cancer slope factors using different statistical approaches.

The U.S. Environmental Protection Agency's cancer guidelines ( USEPA, 2005 ) present the default approach for the cancer slope factor (denoted here as s*) as the slope of the linear extrapolation to the origin, generally drawn from the 95% lower confidence limit on dose at the lowest prescribed risk level supported by the data. In the past, the cancer slope factor has been calculated as the upper 95% confidence limit on the coefficient (q*₁) of the linear term of the multistage model for the extra cancer risk over background. To what extent do the two approaches differ in practice? We addressed this issue by calculating s* and q*₁ for 102 data sets for 60 carcinogens using the constrained multistage model to fit the dose‐response data. We also examined how frequently the fitted dose‐response curves departed appreciably from linearity at low dose by comparing q₁, the coefficient of the linear term in the multistage polynomial, with a slope factor, s_c, derived from a point of departure based on the maximum liklihood estimate of the dose‐response. Another question we addressed is the extent to which s* exceeded s_c for various levels of extra risk. For the vast majority of chemicals, the prescribed default EPA methodology for the cancer slope factor provides values very similar to that obtained with the traditionally estimated q*₁. At 10% extra risk, q*₁/s* is greater than 0.3 for all except one data set; for 82% of the data sets, q*₁ is within 0.9 to 1.1 of s*. At the 10% response level, the interquartile range of the ratio, s*/s_c, is 1.4 to 2.0.     

Model Uncertainty and Risk Estimation for Experimental Studies of Quantal Responses

Experimental animal studies often serve as the basis for predicting risk of adverse responses in humans exposed to occupational hazards. A statistical model is applied to exposure-response data and this fitted model may be used to obtain estimates of the exposure associated with a specified level of adverse response. Unfortunately, a number of different statistical models are candidates for fitting the data and may result in wide ranging estimates of risk. Bayesian model averaging (BMA) offers a strategy for addressing uncertainty in the selection of statistical models when generating risk estimates. This strategy is illustrated with two examples: applying the multistage model to cancer responses and a second example where different quantal models are fit to kidney lesion data. BMA provides excess risk estimates or benchmark dose estimates that reflects model uncertainty.     

Identification of Marginal Effects in Nonseparable Models Without Monotonicity

Nonseparable models do not impose any type of additivity between the unobserved part and the observable regressors, and are therefore ideal for many economic applications. To identify these models using the entire joint distribution of the data as summarized in regression quantiles, monotonicity in unobservables has frequently been assumed. This paper establishes that in the absence of monotonicity, the quantiles identify local average structural derivatives of nonseparable models.