Content uploaded by Atsushi Ono
Author content
All content in this area was uploaded by Atsushi Ono on Mar 22, 2016
Content may be subject to copyright.
A preview of the PDF is not available
Transition of Japan’s statistical tools by decision tree for quantitative data obtained from the general repeated dose administration toxicity studies in rodents
Abstract and Figures
Statistical significance is one of important criteria on judgment of regulatory toxicological testing. The decision tree for analysing quantitative data obtained from repeated dose administration studies in rodents has been in use in Japan around 1981. Since then, several authors proposed improved versions of the decision tree incorporating all possible situations of statistical analysis normally encountered in such studies. Recently, a decision tree, which traces a simple route, unlike the previously proposed ones which trace complex routes has been proposed by a few researchers in Japan. While tracing to the most appropriate statistical tool using a decision tree, we propose to consider following points which also play a significant role in selecting the most appropriate statistical tool: (1) statistical tools that fails to detect a significant difference in the low dose group, (2) use of the one-sided test with high power to detect a significant difference compared with two-sided, (3) as far as possible avoid carrying out statistical analysis on the transformed data, since the analytical result of such data is difficult to interpret, (4) it is important to mention what statistical tools of the decision tree are used for the analysis, (5) examine the data for both normality and homogeneity and (6) for testing homogeneity, use Levene's test. Selection of widely accepted statistical tools is usually preferred to less popular and complex statistical analysis. It has been observed that in recent years the preferred statistical tools for analyzing quantitative data obtained from toxicity studied are of simple in nature but with high power to detect a significant difference.
Figures - uploaded by Atsushi Ono
Author content
All figure content in this area was uploaded by Atsushi Ono
Content may be subject to copyright.
... Data on the body weight, feed consumption, organ weights as well as clinical pathology data were analyzed. All the data was checked for normality with Shapiro-Wilk [24]. Data for each group of animals was subjected to analysis of variance (ANOVA). ...
Probiotics are vital bacteria that colonize the intestine and modify its microflora with benefits for the host. Very few members of the Bacillus group are recognized as safe for use and hence only a few strains are available as commercial preparations for application in humans and animals. Acute and subacute studies in rats were conducted to establish safety of Bacillus clausii (B. clausii) UBBC07. In the acute toxicity study, the oral LD50 for B. clausii UBBC07 was found to be >5000 mg/kg (630 billion cfu/kg) body weight. The NOAEL for B. clausii UBBC07 was found to be 1000 (126 billion cfu) mg/kg body weight/day by oral route in the subacute toxicity study. There were no significant differences between control and treated groups in any of the endpoints assessed using an OECD443 or OECD407 protocol.
B. clausii UBBC07 was found to be resistant to three antibiotics −clindamycin, erythromycin and chloramphenicol. Analysis of the whole genome sequence of B. clausii UBBC07 revealed that the antibiotic resistance genes are present in chromosomal DNA which is intrinsic and not transferable. Toxin genes were also found to be absent. These results suggest consumption of B. clausii UBBC07 is safe for humans.
The question of whether study results are significant, relevant and meaningful is the one to be answered before every study summary and presenting conclusions. This paper analyzes and juxtaposes currently used methods to assess statistical significance, effect size, and highlights the value of understating and assessing biological relevance. Many opinions of experts in various fields are cited to demonstrate the ambiguity of merely p-value usage. The answer to the question of the best approach is complex and a 3-step approach is suggested taking into consideration
a. Statistical assessment of differences between groups
b. Effect analysis and
c. Biological relevance assessment.
The paper emphasizes the need to take into account more than just statistical significance in the decision process, or decisions on accepting or rejecting hypotheses. p-values or any other statistical tool is not recommended as the main criterion for decision making. Furthermore, none of the above mentioned 3 steps should be used in isolation to assess the results. Moreover, there is a need for publication of negative results unless directly caused by poor design or low sample size because the current tendency to focus entirely on positive results biases the literature and leads to unnecessary replication of experiment.
It is inappropriate to apply the Student t-test in simultaneous comparison of more than two groups. It is, however, the prevailing custom to apply the test in long term toxicity studies in rats. This often results in misleading and seemingly conflicting results indicating that the mean of a variable in the lower dose groups differs significantly from that in a control, whereas those in the higher groups do not. The authors analysed four recent toxicity studies and found that such meaningless abnormalities in hypothesis testing were almost abolished by application of a new algorithm for statistical analysis. The new method is as follows: if variances among groups are homogenous, the analysis of variance of a one-way layout type is applied. Otherwise, the nonparametric Kruskal-Wallis' test is used. If there are significant differences among the means by these methods, Dunnett type multiple comparisons between a control and each treatment group are performed when the numbers of animals are equal in all groups. The Scheffe type comparisons are applied when they are not. The algorithm was named as ASSIT (Algorithm for Simultaneous Statistical Inference in Toxicology). A new algorithm is presented for random allocation of animals to experimental groups. Animals are selected from the smaller values of the standardized norm calculated from the pre-test mean body weights and body weight gains. After stratified random allocation of animals to groups, homogeneity of variances and group means are confirmed statistically before final group determination. These procedures are provided by programs prepared for a computer system.
A Comparison of One- and Two-Sided Tests for Judging Significant Differences in Quantitative Data Obtained in Toxicological Bioassay of Laboratory Animals. Katsumi KOBAYASHI. Biosafety Research Center, Foods, Drugs and Pesticides—Since there are many ambiguous statements concerning the selection of one- or two-sided tests in the statistical analysis of toxicological data, I examined the rate of appearance of significant differences in the data showing a trend in either a fixed direction or a mixed sided direction compared with the control and the number of significant differences to these two tests by t and Dunnett's tests in a longterm chronic/carcinogenicity study conducted at the An-Pyo Center, in addition to referring to the most widely used statistical analyses by mean of the one or two-sided test in the literature. The results were as follows; (1) Almost all quantitative data (578 out of 700 cases) showed a fixed trend with statistically significant differences (p<0.05) compared with the control value. (2) The number of significant differences obtained with the one-sided test was greater than with the two-sided test in either analysis of the t or Dunnett's test; that is, the percentages of the significant differences in the two-sided test were 85 and 86% of those in the one-sided test by means of t and Dunnett's test, respectively. (3) The frequency of use of the one-sided tests was very low in both Japanese and international publications. Consequently, the one-sided test may be recommended for statistical analyses of toxicological bioassay data that generally show a fixed trend as compared with the control values, since more rigid evaluation of the data of the chemical effects on the living body and the environment is necessary. (J Occup Health 1997; 39: 29-35)
ICR mice and Wistar rats of both sexes were fed a diet containing ZnSO4 at 0, 300, 3,000 and 30,000 ppm for 13 weeks. Animals in the 30,000 ppm group showed retarded growth along with low food intake, abnormal values in a few hematological parameters and regressive changes of the pancreatic exocrine gland. In addition, mice had decreased water intake and significant deviations in biochemical parameters, toxic lesions appeared in the stomach, intestine and spleen of both sexes and in the kidney of females. Four males and one female mouse were found dead or moribund during the study. The maximum no-effect level of ZnSO4 was determined to be 3,000 ppm, which is approximately equivalent to the following milligram doses: mice; male 458 mg/kg/day, female 479 mg/kg/day, rats; male 234 mg/kg/day, female 243 mg/kg/day. © 1981, Pesticide Science Society of Japan. All rights reserved.
The chronic toxicity of bis (2-chloro-l-methylethyl) ether (dichlorodiisopropyl ether: DCIP), a nematicide, was studied in specific pathogen-free (SPF) mice of the ICR strain. DCIP (purity: 98.5%) was fed to each groups of 56 male and 56 female mice for a period of 104 weeks at dietary dose levels of 0, 80, 400, 2,000 or 10,000 ppm. Seven animals of each group at 13, 26 and 52 weeks, 6 animals at 78 weeks and all surviving mice at the end of the experiment were sacrificed to carry out hematological, biochemical and pathological examinations. In both sexes of the 10,000 ppm group, a remarkable depression of body weight gain was noted corresponding with the decrease in food consumption. Mild anemia and hemosiderin deposition and increased extramedullary hematopoiesis of the spleen were observed at 13, 26 and 52 weeks. Only female mice in the 2,000 ppm group showed a slight inhibition of body weight increase and an anemic tendency. No conspicuous changes attributable to the feeding of DCIP were observed less than in the 2,000 ppm group males and the 400 ppm group females. There was also no significant increase in the incidence of age-related lesions and tumors in any treated groups as compared with those in the control group. Based on these results, it was concluded that the maximum no-effect levels of DCIP were 2,000 ppm for male mice (198 mg/kg/day) and 400 ppm for female mice (35.8 mg/kg/day). © 1979, Pesticide Science Society of Japan. All rights reserved.
Let (Xoj, X1j, …, Xkj) be the result of a single trial, where the subscripts o is associated with a control and the subscripts 1, …, k with treatments. To test the joint hypothesis P(Xij – Xoj > 0)=l/2 = P(Xij – Xoj < 0), all i, compute the test criterion (r1, …, rk) where ri is the number of times Xij, – Xoj is negative in n trials. A method for computing the distribution of (r1, …, rk) is illustrated. Exact probability distributions of min ri are given for k = 2, n = 4(1)10 and k = 3, n = 4(1)7. It is conjectured that 2(min r – n/2)/ is distributed approximately as Dunnett's t. Tables based on this conjecture are computed and values are seen to agree well with comparable values from the exact distribution.
The abstract for this document is available on CSA Illumina.To view the Abstract, click the Abstract button above the document title.
This paper considers the use of rank sums from a combined ranking of k independent samples in order to decide which populations differ. Such a procedure is suggested as a convenient alternative to making separate rankings for each pair of samples, and the two methods are compared. Asymptotic use of the normal tables is given and the treatment of ties is discussed. A numerical example is given.