Repeated-measure analyses: Which one? A survey of statistical models and recommendations for reporting.

Repeated-measure analysis of variance is a general term that can imply a number of different statistical models used to analyze data from studies in which measurements are taken from each subject on more than one occasion. Repeated-measure analyses encompass univariate models (with or without sphericity adjustment), multivariate models, mixed models, analysis of covariance, multilevel models, latent growth models, and hybrids of these models. These models are based on different assumptions, especially regarding correlations (sphericity) between within-subject factors, which comprise the variance-covariance matrix. Violation of this assumption may lead to misleading and erroneous conclusions. Because many papers do not provide enough information about what analysis was really conducted, and about why it was done, the reader is unable to evaluate the validity of the analysis. Here a brief overview of several of the most commonly used models for analyzing data from repeated-measure designs is provided, and guidance is suggested for describing the statistical approach employed. The goals of this paper are (1) to give authors an overview of the diversity of commonly used models and associated assumptions, and (2) to facilitate reporting sufficient information about the tests to allow the reader to evaluate the validity of the tests and the credibility of the inferences made by the authors. Among the available approaches to repeated-measure analyses, the mixed model is recommended for its flexibility in handling different covariance structures and its insensitivity to missing data. Whether or not it is used, the overall guiding principles in reporting should always be Accuracy, Completeness, and Transparency (ACT principles): tell the reader precisely all what you did and why.

Oral gavage subchronic neurotoxicity and inhalation subchronic immunotoxicity studies of ethylbenzene in the rat.

The potential for neurotoxicological and immunotoxicological effects of ethylbenzene was studied in young adult Crl:CD(SD) rats following 90-day oral (neurotoxicity) or 28-day inhalation (immunotoxicity) exposures. In the neurotoxicity study, ethylbenzene was administered orally via gavage twice daily at 0, 25, 125, or 250 mg/kg per dose (total daily dosages of 0, 50, 250, or 500 mg/kg bwt/day [mg/kg bwt/day]) for 13 weeks and the functional observational battery (FOB), automated tests for motor activity and neuropathological examination were conducted. In the immunotoxicity study, animals were exposed by inhalation to 0, 25, 100, or 500 ppm ethylbenzene (approximately 26, 90, or 342 mg/kg bwt/day as calculated from physiologically based pharmacokinetic modeling). Immunotoxicity was evaluated in female rats using the splenic antibody-forming cell plaque-forming assay in sheep red blood cell sensitized animals. The no-observed-effect level for the oral gavage study was 50mg/kg bwt/day based on increased relative weights of the liver and kidneys in the male rats. The no-observed-adverse-effect level (NOAEL) for adult neurotoxicity was the highest dose tested 500 mg/kg bwt/day. The NOAEL for the immunotoxicity evaluation was the highest tested exposure concentration, 500 ppm (342 mg/kg bwt/day).

Statistical issues and techniques appropriate for developmental neurotoxicity testing: a report from the ILSI Research Foundation/Risk Science Institute expert working group on neurodevelopmental endpoints.

The data from developmental neurotoxicity (DNT) guideline studies present a number of challenges for statistical design and analysis. The importance of specifying the planned statistical analyses a priori cannot be overestimated. A review of datasets submitted to the US Environmental Protection Agency revealed several inadequate approaches, including issues of Type I error control, power considerations, and ignoring gender, time, and litter allocation as factors in the analyses. Since DNT studies include numerous experimental procedures conducted on the dam and offspring at several ages, it is not unusual to have hundreds of significance tests if each was analyzed separately. Two general approaches to control experiment-wise Type I inflation are: 1) statistical/design considerations that reduce the number of p-values, including factorial designs, multivariate techniques, and repeated-measures analyses; and 2) adjustments to the alpha level, including newer approaches that are less conservative than, for example, Bonferroni corrections. The design of the DNT study includes testing of both sexes, and gender must be included in the statistical analysis for the determination of sex-related differences, and, indeed, including both sexes may increase power. The influence of litter must be taken into account in the allocation of test animals as well as the statistical analyses. This manuscript reviews many key considerations in the analysis of DNT studies with recommendations for statistical approaches and reporting of the data.

Effects of intensity and type of prepulse stimulus on prepulse inhibition in scopolamine treated rats.

Prepulse inhibition (PPI) of the auditory startle response (ASR) is a behavioral test that has been used to measure auditory thresholds, to assess sensory-motor integration functions, and its use has been recommended in the United States Environmental Protection Agency Developmental Neurotoxicity Guideline (OPPTS 870.6300). The purpose of the present study was to determine to what extent the intensity and/or type of prepulse stimuli modulate PPI in scopolamine-treated rats. The PPI of the ASR peak amplitude was measured when the intensity of a 10-kHz prepulse tone was varied (69-, 80-, and 90 dB[A]; Experiment 1) and when both the intensity and type of auditory prepulse (a 10-kHz tone vs. a white noise burst) were varied (Experiment 2). Scopolamine treatment attenuated PPI in both experiments and interacted significantly with the prepulse stimulus intensity in Experiment 1. In Experiment 2, the percent of PPI was linearly related to prepulse stimulus intensity for trials using a tone, but was biphasic on trials using a white-noise prepulse stimulus. Prepulse stimuli of certain intensities elicited a response, and this response was greater when the prepulse stimulus was a white noise burst versus a tone of the same intensity. Further, the response to the prepulse altered the amount of inhibition and, therefore, confounded the overall measure of PPI at the higher prepulse stimulus intensity levels. Overall, these results indicate that careful consideration of the intensity and type of prepulse stimuli be taken in the context of their potential to induce a prepulse-elicited response, as well as providing the appropriate measures of such a response, when designing and interpreting PPI experiments.

Differential sensitivity of blood, peripheral, and central cholinesterases in beagle dogs following dietary exposure to chlorpyrifos.

Two studies were performed to find out whether exposure limits that protect brain acetylcholinesterase (AChE) will protect peripheral tissue AChE after exposure to chlorpyrifos (CPF), an organophosphate insecticide. In a methods-development study, male dogs (3/dose) were exposed to 0.0, 0.3, 0.6, or 1.2mg/kg/day CPF in their diets for 4 weeks. Mixed cholinesterase (mChE), AChE, and butyrylcholinesterase (BuChE) activities were measured in plasma, RBC, brain, left atrium and ventricle, diaphragm, quadriceps, and nodose ganglia. Plasma, brain and peripheral tissue BuChE was inhibited at all dose levels. While RBC AChE was inhibited at all doses, brain and peripheral AChE activities were unaffected. In the main study, dogs (4/sex/dose) were exposed to 0.0, 0.5, 1.0, or 2.0mg/kg/day CPF in their diets for six weeks and RBC AChE was significantly inhibited at all doses in both sexes. Diaphragm, quadriceps, and nodose ganglia AChE was unaffected by treatment. Brain AChE was decreased by approximately 6% compared to controls in high-dose groups, and this was considered a threshold effect. Left atrium AChE in high-dose dogs was 25.5% less (males) and 32.1% greater (females) than controls; these differences were attributed to chance. While peripheral tissue and brain AChE were not affected following exposure to 1.0mg/kg/day, RBC AChE was inhibited at all doses. These results show that RBC AChE is more sensitive than brain or peripheral tissue AChE to inhibition by CPF, and that protection of brain AChE would protect peripheral tissue AChE.

Neurotoxicity test validation, positive controls and proficiency: are chemicals necessary?

The USEPA neurotoxicity guidelines require the use of positive control data in support of toxicology studies submitted to the Agency and emphasize the use of chemicals to accomplish this requirement. These guidelines, though, propose a number of different rationales for the use of chemicals as positive control agents. We re-evaluated the potential roles of positive control data in addressing three questions: 1) what does the test measure? 2) is the performing laboratory proficient in the use of the test? 3) do the complementary data submitted in support of neurotoxicity studies conducted with the test material provide enough context for the interpretation of the biological significance of an effect? While, for most types of guideline neurotoxicity tests, the use of test chemicals has been emphasized for positive control testing, the use of non-chemical procedures (i.e., systematic manipulation of the experimental parameters of a test, which poses less risk of adverse effects to the test animals) should be strongly considered as a potential alternative.

Workshop proceedings: managing conflict of interest in science. A little consensus and a lot of controversy.

There is no doubt that participants in the Conflict of Interest (COI) Workshop at the Society of Toxicology (SOT) 2005 Annual Meeting (New Orleans, 6-10 March 2005) engaged in a vigorous and useful exchange of diverse ideas and viewpoints. While there was consensus on the value and interest of this Workshop, there was less consensus and more controversy over many of the issues discussed during the Workshop, which included the distinction between bias and conflict, the success or failure of policies of disclosure, whether waivers should or should not be granted to conflicted individuals in order to seat a "balanced" committee with appropriate expertise, whether conflicted individuals retain the ability to recognize their own conflict, and more. The discussion left no doubt, however, that conflict of interest will remain an important and controversial issue in the scientific community for some time to come.

Validation of an auditory startle response system using chemicals or parametric modulation as positive controls.

Neurotoxicity regulatory guidelines mandate that automated test systems be validated using chemicals. However, in some cases, chemicals may not necessarily be needed to prove test system validity. To examine this issue, two independent experiments were conducted to validate an automated auditory startle response (ASR) system. In Experiment 1, we used adult (PND 63) and weanling (PND 22) Sprague-Dawley rats (10/sex/dose) to determine the effect of either d-amphetamine (4.0 or 8.0 mg/kg) or clonidine (0.4 or 0.8 mg/kg) on the ASR peak amplitude (ASR PA). The startle response of each rat to a short burst of white noise (120 dB SPL) was recorded over 50 consecutive trials. The ASR PA was significantly decreased (by clonidine) and increased (by d-amphetamine) compared to controls in PND 63 rats. In PND 22 rats, the response to clonidine was similar to adults, but d-amphetamine effects were not significant. Neither drug affected the rate of the decrease in ASR PA over time (habituation). In Experiment 2, PND 31 Sprague-Dawley rats (8/sex) were presented with 150 trials consisting of either white noise bursts of variable intensity (70-120 dB SPL in 10 dB increments, presented in random order) or null (0 dB SPL) trials. Statistically significant sex- and intensity-dependent differences were detected in the ASR PA. These results suggest that in some cases, parametric modulation may be an alternative to using chemicals for test system validation.

Factors affecting grip strength testing.

The rodent grip strength test was developed decades ago and is a putative measure of muscular strength. This test has been included in the functional observational battery (FOB) to screen for neurobehavioral toxicity, and changes in grip strength have been interpreted as evidence of motor neurotoxicity. Despite its widespread use, questions remain about what the grip strength test actually measures. In this study, potential confounders of the grip strength test were identified and tested, including operational parameters, disruption of peripheral sensory function and changes in body weight. Operational parameters (sampling rate, system type and trial angle but not trial speed) had dramatic effects on grip strength data. Doxorubicin (DX, 10 mg/kg iv) was used to cause sensory impairment. It decreased forelimb and hindlimb grip strength (by 27% and 32%, respectively, compared with controls), an effect that was correlated with degeneration of peripheral and central sensory components (distal tibial and sural nerves, dorsal funiculus of the spinal cord and dorsal, but not ventral, spinal roots). Feed restriction-induced loss of body weight (26% compared with controls) and muscle mass (20% compared with controls) reversibly decreased both forelimb and hindlimb grip strength (18% and 17%, respectively, compared with controls). Ignoring these confounding factors could potentially lead to increased data variability and inconsistency within single studies, across studies and in historical control data sets. To assist in data interpretation and evaluation of grip strength results, it is suggested that exact conditions of application of the test be reported in greater detail. Furthermore, given that the grip strength test can be influenced by factors other than true muscular strength, use of the term grip performance is proposed to better reflect the apical nature of this test.