Tests for trend in developmental toxicity experiments with correlated binary data

Toxicity studies of acetoin and 2,3-pentanedione administered by inhalation to Wistar Han [Crl:WI(Han)] rats and B6C3F1/N mice

Acetoin and 2,3-pentanedione are highly volatile components of artificial butter flavoring (ABF). Concerns over the inhalation toxicity of these compounds originate from the association between occupational exposures to ABF and adverse fibrotic lung effects, specifically obliterative bronchiolitis (OB) in the distal airways. 2,3-Pentanedione has been used as a replacement for 2,3-butanedione (diacetyl) in some ABF due to concerns about the respiratory toxicity of 2,3-butanedione. However, 2,3-pentanedione is structurally similar to 2,3-butanedione and has been shown to exhibit potency similar to 2,3-butanedione regarding airway toxicity following acute inhalation (whole-body) exposure. This report describes a series of studies to evaluate the 2-week inhalation toxicity of acetoin and the 3-month inhalation toxicity of acetoin and 2,3-pentanedione. (Abstract Abridged).

Toxicity Studies of Sodium Metavanadate and Vanadyl Sulfate Administered in Drinking Water to Sprague Dawley (Hsd:Sprague Dawley SD) Rats and B6C3F1/N Mice

Oral human exposure to vanadium may occur due to its presence in food and drinking water and its use in dietary supplements. The most prevalent oxidation states of vanadium in food and drinking water have been characterized as tetravalent and pentavalent. Vanadyl sulfate and sodium metavanadate were selected as representative tetravalent (V4+) and pentavalent (V5+) test articles for these studies, respectively. To assess the potential for oral toxicity of vanadium compounds with differing oxidation states under similar test conditions, the 3-month National Toxicology Program (NTP) toxicity studies of sodium metavanadate and vanadyl sulfate in male and female Sprague Dawley (Hsd:Sprague Dawley SD) rats (including perinatal exposure) and in B6C3F1/N mice. Drinking water concentrations for sodium metavanadate (0, 31.3, 62.5, 125, 250, and 500 mg/L) and vanadyl sulfate (0, 21.0, 41.9, 83.8, 168, and 335 mg/L) were selected on the basis of previously published 14-day drinking water studies conducted as part of the NTP vanadium research program. (Abstract Abridged).

Analysis of incidence data in developmental toxicity studies: Statistical tests to account for litter effects in fetal defect data

BACKGROUND

When analyzing fetal defect incidence in laboratory animal studies, correlation in responses within litters (i.e., litter effects) can lead to increased false-positive rates if litter effects are not incorporated into the analysis. Studies of fetal defects require analysis methods that are robust across a broad range of defect types, including those with zero or near-zero incidence rates in control groups.

METHODS

A simulation study compared power and false-positive rates for six approaches across a range of background defect rates and litter size distributions. Statistical methods evaluated included ignoring the litter effect as well as parametric and nonparametric approaches based on litter proportions, generalized linear mixed models (GLMMs), the Rao-Scott Cochran-Armitage (RSCA) trend test, and a modification to the RSCA (mRSCA) introduced here to improve estimation at low background rates. These methods were also applied to a common and a rare defect from two prenatal developmental toxicology studies conducted by the National Toxicology Program (NTP).

RESULTS

At background defect rates of 1%, the mRSCA and parametric litter proportion methods provided gains in power over the nonparametric litter proportion method, the GLMM method, and the RSCA method. Simulations involving litter loss in high-dose groups showed loss of power for both litter proportion methods.

CONCLUSIONS

The mRSCA test developed here compares favorably with other litter-based approaches and is robust across a range of background defect rates and litter size distributions, making it a practical choice for prenatal developmental toxicology studies involving both common and rare fetal defects.

Dose-Response Modeling with Summary Data from Developmental Toxicity Studies

Dose-response analysis of binary developmental data (e.g., implant loss, fetal abnormalities) is best done using individual fetus data (identified to litter) or litter-specific statistics such as number of offspring per litter and proportion abnormal. However, such data are not often available to risk assessors. Scientific articles usually present only dose-group summaries for the number or average proportion abnormal and the total number of fetuses. Without litter-specific data, it is not possible to estimate variances correctly (often characterized as a problem of overdispersion, intralitter correlation, or "litter effect"). However, it is possible to use group summary data when the design effect has been estimated for each dose group. Previous studies have demonstrated useful dose-response and trend test analyses based on design effect estimates using litter-specific data from the same study. This simplifies the analysis but does not help when litter-specific data are unavailable. In the present study, we show that summary data on fetal malformations can be adjusted satisfactorily using estimates of the design effect based on historical data. When adjusted data are then analyzed with models designed for binomial responses, the resulting benchmark doses are similar to those obtained from analyzing litter-level data with nested dichotomous models.

A systematic evaluation of the potential effects of trichloroethylene exposure on cardiac development

The 2011 EPA trichloroethylene (TCE) IRIS assessment, used developmental cardiac defects from a controversial drinking water study in rats (Johnson et al. [51]), along with several other studies/endpoints to derive reference values. An updated literature search of TCE-related developmental cardiac defects was conducted. Study quality, strengths, and limitations were assessed. A putative adverse outcome pathway (AOP) construct was developed to explore key events for the most commonly observed cardiac dysmorphologies, particularly those involved with epithelial-mesenchymal transition (EMT) of endothelial origin (EndMT); several candidate pathways were identified. A hypothesis-driven weight-of-evidence analysis of epidemiological, toxicological, in vitro, in ovo, and mechanistic/AOP data concluded that TCE has the potential to cause cardiac defects in humans when exposure occurs at sufficient doses during a sensitive window of fetal development. The study by Johnson et al. [51] was reaffirmed as suitable for hazard characterization and reference value derivation, though acknowledging study limitations and uncertainties.

Statistical analysis of histopathological endpoints

Histopathological assessments of fish from aquatic ecotoxicology studies are being performed with increasing frequency. Aquatic ecotoxicology studies performed for submission to regulatory agencies are usually conducted with multiple subjects (e.g., fish) in each of multiple vessels (replicates) within a water control and within each of several concentrations of a test substance. A number of histopathological endpoints are evaluated in each fish, and a severity score is generally recorded for each endpoint. The severity scores are often recorded using a nonquantitative scale of 0 to 4, with 0 indicating no effect, 1 indicating minimal effect, through 4 for severe effect. Statistical methods often used to analyze these scores suffer from several shortcomings: computing average scores as though scores were quantitative values, considering only the frequency of abnormality while ignoring severity, ignoring any concentration-response trend, and ignoring the possible correlation between responses of individuals within test vessels. A new test, the Rao-Scott Cochran-Armitage by Slices (RSCABS), is proposed that incorporates the replicate vessel experimental design and the biological expectation that the severity of the effect tends to increase with increasing doses or concentrations, while retaining the individual subject scores and taking into account the severity as well as frequency of scores. A power simulation and examples demonstrate the performance of the test. R-based software has been developed to carry out this test and is available free of charge at www.epa.gov/med/Prods_Pubs/rscabs.htm. The SAS-based RSCABS software is available from the first and third authors.

Comparison of tests for categorical data from a stratified cluster randomized trial

Two features commonly exhibited by randomized trials of health promotion interventions are cluster randomization and stratification. Ignoring correlations between individuals within clusters can lead to an inflated type I error rate and hence a P-value which overstates the significance of the result. This paper compares several methods for analysing categorical data from a stratified cluster randomized trial. We propose an extension of a method from survey sampling that uses the design effect to reduce the effective sample size. We compare this with three methods from Zhang and Boos that extend the standard Cochran-Mantel-Haenszel (CMH) statistic by using appropriate covariance matrices, and with a bootstrap method. The comparison is based on empirical type I error rates from a simulation study, in which the number of clusters randomized is small, as in most public health intervention studies. The method that performs consistently well is one of the Zhang and Boos extensions of the standard CMH statistic.

Bayesian analysis of mutational spectra

Studies that examine both the frequency of gene mutation and the pattern or spectrum of mutational changes can be used to identify chemical mutagens and to explore the molecular mechanisms of mutagenesis. In this article, we propose a Bayesian hierarchical modeling approach for the analysis of mutational spectra. We assume that the total number of independent mutations and the numbers of mutations falling into different response categories, defined by location within a gene and/or type of alteration, follow binomial and multinomial sampling distributions, respectively. We use prior distributions to summarize past information about the overall mutation frequency and the probabilities corresponding to the different mutational categories. These priors can be chosen on the basis of data from previous studies using an approach that accounts for heterogeneity among studies. Inferences about the overall mutation frequency, the proportions of mutations in each response category, and the category-specific mutation frequencies can be based on posterior distributions, which incorporate past and current data on the mutant frequency and on DNA sequence alterations. Methods are described for comparing groups and for assessing dose-related trends. We illustrate our approach using data from the literature.

A simple method for analysing overdispersion in clustered Poisson data

A simple method is proposed for analysing grouped count data exhibiting overdispersion relative to a Poisson model. The method is similar to the approach suggested for the analysis of clustered binary data in Rao and Scott (1992). It requires no specific model for the overdispersion and it can be implemented easily using standard programs designed to handle independent Poisson counts, after a small amount of preprocessing.

Procedures for calculating benchmark doses for health risk assessment

Safety assessment for noncancer health effects generally has been based upon dividing a no observed adverse effect (NOAEL) by uncertainty (safety) factors to provide an acceptable daily intake (ADI) or reference dose (RfD). Since the NOAEL does not utilize all of the available dose-response data, allows higher ADI from poorer experiments, and may have an unknown, unacceptable level of risk, the benchmark dose (BD) with a specified, controlled low level of risk has become popular as an adjunct to the NOAEL or the low observed adverse effect level (LOAEL) in the safety assessment process. The purpose of this paper is to summarize statistical procedures available for calculating BDs and their confidence limits for noncancer endpoints. Procedures are presented and illustrated for quantal (binary), quasicontinuous (proportion), and continuous data. Quasicontinuous data arise in developmental studies where the measure of an effect for a fetus is quantal (normal or abnormal) but the experimental unit is the mother (litter) so that results can be expressed as the proportion of abnormal fetuses per litter. However, the correlation of effects among fetuses within a litter poses some additional statistical problems. Also, developmental studies usually include some continuous measures, such as fetal body weight or length. With continuous data there generally is not a clear demarcation between normal and adverse measurements. In such cases, extremely high and/or low measurements at some designated percentile(s) can be considered abnormal. Then the probability (risk) of abnormal individuals can be estimated as a function of dose. The procedure for estimating a BD with continuous data is illustrated using neurotoxicity data. When multiple measures of adverse effects are available, a BD can be estimated based on a selected endpoint or the appearance of any combination of endpoints. Multivariate procedures are illustrated using developmental and reproductive toxicity data.

A comparison of methods for estimating the benchmark dose based on overdispersed data from developmental toxicity studies

Developmental anomalies resulting from prenatal toxicity can be manifested in terms of both malformations among surviving offspring and prenatal death. Although these two endpoints have traditionally been analyzed separately in the assessment of risk, multivariate methods of risk characterization have recently been proposed. We examined this and other issues in developmental toxicity risk assessment by evaluating the accuracy and precision of estimates of the effective dose (ED05) and the benchmark dose (BMD05) using computer simulation. Our results indicated that different variance structures (Dirichlet-trinomial and generalized linear model) used to characterize overdispersion yielded comparable results when fitting joint dose response models based on generalized estimating equations. (The choice of variance structure in separate modeling was also not critical.) However, using the Rao-Scott transformation to eliminate overdispersion tended to produce estimates of the ED05 with reduced bias and mean squared error. Because joint modeling ensures that the ED05 for overall toxicity (based on both malformations and prenatal death) is always less than the ED05 for either malformations or prenatal death, joint modeling is preferred to separate modeling for risk assessment purposes.

Use of generalized linear mixed models in analyzing mutant frequency data from the transgenic mouse assay

The transgenic mouse assay is now widely used for the study of mutagenesis in diverse rodent tissues and to test chemicals for genotoxic potential. This kind of assay generally involves nested observations at several levels of sampling, e.g., animals, packaging reactions, and plates. Due to the common origin, the mutant frequency (MF) in tissues from the same animal are likely to be positively correlated, inducing extra variation relative to the common binomial variation. In this article, a generalized linear mixed model is used to analyze the overdispersed binomial data on mutant frequency from the transgenic mouse assay, with a random effect for each level of the sampling hierarchy. This is a comprehensive framework within which different sources of variation in the data can be evaluated in nested factorial experiments and treatment effects can be assessed simultaneously. It avoids the current practice of repeated testing for excess binomial variability at each level of the sampling hierarchy and aggregating data up the levels, but fits the data with one single model. Parameters associated with the fixed effects, particularly dose, and the variance components for the random effects (e.g., animals, packages, and plates) can be estimated and tested for significance. Data previously reported in the literature involving the lacl gene from the Big Blue mouse are used to illustrate the proposed method.

Statistical analysis of the lacI transgenic mouse mutagenicity assay

The transgenic mouse assay is now widely used to test chemicals for genotoxic potential. In this article, we consider statistical tests for increasing trend in mutant frequency with increasing dose, along with statistical models that may be used to describe the observed dose-response relationships. The application of these methods is illustrated using data on 2-acetylaminofluorene, di(2-ethylhexyl)phthalate, heptachlor, and sodium phenobarbital. No strong evidence of extra-binomial variation was detected at the plate level, but greater evidence was noted when the data were aggregated to the package or animal level in liver, necessitating the use of statistical methods that allow for overdispersion relative to binomial variation. Clear increase on mutant frequency induced by 2-acetylaminofluorene was detected in both liver and bladder, but no apparent trends were noted with di(2-ethylhexyl)phthalate, heptachlor, and sodium phenobarbital. The exponential model provides a good fit to the observed dose-response relationship in liver, whereas a Weibull model provides a better fit for bladder.

A computer program for the analysis of over-dispersed counts and proportions

Over-dispersed binary and count data occur frequently in many fields of application. Examples include occurrence of cavities in one or more teeth, and development of tumors in one or more animals of a litter. Methods of statistical analyses that ignore correlation between observations underestimate the standard errors. Consequently, coverage proportions of confidence intervals and significance levels of tests are distorted. To implement methods for the analysis of correlated binary or count data requires a level of sophistication for data analysis such that one can specify a model for over-dispersion and the correlation between observations. To analyze the over-dispersed binary or count data, one could postulate a specific statistical model and use maximum likelihood methods for the estimation of parameters. However, it may be preferable to employ an approach that does not rely on modeling because the true model is hard to know with confidence. Rao and Scott (J.N.K. Rao and A.J. Scott, Biometrics 48 (1992) 577-585)y and Scott and Rao (A.J. Scott and J.N.K. Rao, submitted for publication, 1995) proposed simple methods for analyzing correlated binary and count data exhibiting over-dispersion relative to a binomial and homogeneous Poisson model. This paper presents the SAS program to implement their methods to analyze over-dispersed binary and count data. To demonstrate the implementation and the usefulness of their methods, we present an application involving sensitivity of a monoclonal antibody and the number of mammary tumors developing in rats.

Sources of variability in data from a positive selection lacZ transgenic mouse mutation assay: an interlaboratory study

Experimental features of a positive selection transgenic mouse mutation assay based on a lambda lacZ transgene are considered in detail, with emphasis on results using germ cells as the target tissue. Sources of variability in the experimental protocol that can affect the statistical nature of the observations are examined, with the goal of identifying sources of excess variation in the observed mutant frequencies. The sources include plate-to-plate (within packages), package-to-package (within animals), and animal-to-animal variability. Data from five laboratories are evaluated in detail. Results suggest only scattered patterns of excess variability below the animal-to-animal level, but, generally, significant excess variability at the animal-to-animal level. Using source of variability analyses to guide the choice of statistical methods, control-vs-treatment comparisons are performed for assessing the male germ cell mutagenicity of ethylnitrosourea (ENU), isopropyl methanesulfonate (iPMS), and methyl methanesulfonate (MMS). Results on male germ cell mutagenesis of ethyl methanesulfonate (EMS) and methylnitrosourea (MNU) are also reported.

Variance estimation for complex surveys using replication techniques

The analysis of survey data requires the application of special methods to deal appropriately with the effects of the sample design on the properties of estimators and test statistics. The class of replication techniques represents one approach to handling this problem. This paper discusses the use of these techniques for estimating sampling variances, and the use of such variance estimates in drawing inferences from survey data. The techniques of the jackknife, balanced repeated replication (balanced half-samples), and the bootstrap are described, and the properties of these methods are summarized. Several examples from the literature of the use of replication in analysing large complex surveys are outlined.

Statistical considerations in the design and analysis of community intervention trials

Community intervention trials are often characterized by the allocation of intact social units to different intervention groups. The assessment of adequate sample size for such trials must take into account the statistical dependencies among responses observed within an allocated unit. However, the small numbers of units typically involved in such trials imply that many methods of analysis that have been proposed for analyzing correlated data, particularly in the case of a dichotomous outcome variable, are not applicable to such designs. In this article we investigate this issue and determine the minimum number of units required per group, for the case of both a dichotomous and a continuous outcome variable, needed to provide adequate statistical power for detecting various levels of treatment effect. The use of significance testing as a method of detecting intracluster correlation is also investigated, and, in general, discouraged.

Risk estimation--an overview: disease risk estimation based upon animal bioassays

Much more could be written about the issues addressed here, as well as about issues that are not even mentioned. The goal was to present a brief overview of some of the techniques and issues in quantitative health risk assessment based upon animal data. Hopefully, this overview will provoke some attention to specific in risk assessment that require more research. Perhaps the bibliographic references given will lead to other papers.