Full document here (in French).
Here in English (Official translation)
[English: Extended quote
For information, the authors describe more and earlier deaths in the female population in all treated groups and „generally'33 earlier and larger tumours. They describe liver damage for all of the GM-treated males.
The main criticism of the study's results concerns the lack of statistical data analysis supporting these findings.
The authors simply note that the treated groups were generally more affected than the control group without testing the possibility that these results may have been due to chance. The authors, when asked about this point, indicated that they had simply wanted to report their results, which they had found disturbing, not in the form of a statistical analysis but rather in the form of a description as practised in human clinical research.
“Gilles-Éric Séralini’s team added that this study corresponded to a research protocol and was not at all intended to be a regulatory test protocol. The results are presented factually for both tumours and mortality. The team did not wish to conduct statistical analyses for these points, as it was keenly aware that with 10 rats per group, tests would not be sufficiently powerful. The team had criticised the Monsanto study for just this. Gilles-Eric Séralini stated that thorough statistical analyses had however been undertaken for biochemical parameters, confirming disturbances that can lead to the observed pathologies” (Extracted from the verbatim report of the hearing with the study’s authors)
Following the hearing, ANSES asked Mr Séralini to submit all of the study‟s raw data. The authors did not grant this request (see hearing) but submitted quantitative data on mortality and the onset of non-regressive tumours. The ECEAG was able to use some of these data (on mortality) in addition to those available in the publication (Figure 1 and Table 2) to determine the significance of certain results.
In order to increase the probability of detecting effects, the ECEAG first undertook one-tailed tests considering Type I error risks of 5% without taking the effect of multiple testing into account. This approach increases statistical power and yet also increases the risk of false positives (false discovery). It is considered the most favourable statistical test for highlighting a maximum number of effects that need to be interpreted biologically. The statistical tests were then corrected so as to limit risks of false discoveries (FDR (False Discovery Rate) control or correction) (Benjamini and Hochberg 1995). Indeed, undertaking multiple statistical tests on a given dataset rapidly increases the rate of Type I errors, i.e. the probability of observing falsely significant differences..]
Table 1. (above) Results of statistical tests (P = probability of error of the first kind) on mortality. P values were not adjusted for multiplicity of tests. If these corrections are applied, no difference is significant.
[English
Three series of statistical tests were undertaken by the ECEAG:
The first series of statistical tests aimed to determine whether there were significant differences in mortality rates at the end of the study between the control group of rats and the GMO and/or ROUNDUP groups of rats37. These tests compared the null hypothesis H0 „Mortality rate in the control group = Mortality rate in the GMO and/or ROUNDUP groups‟ and the alternative hypothesis A „Mortality rate in the control group < Mortality rate in the GMO and/or ROUNDUP groups‟. This series of tests was undertaken using the data extracted from Figure 1 of the study by Séralini et al. (2012). The risk of a Type 1 error (probability of wrongly rejecting H0) was calculated by conducting Fisher's exact test on 2 dead rats out of the 10 rats of the female control group and 3 dead rats out of the 10 rats of the male control group. The results (Table 1) show that the differences in mortality rates are significant at the 5% level before correction (FDR) for two in 18 groups of rats:
* for the female group, GMO at the 22% dose,
* and for the female group, GMO + R at the 22% dose.
When taking multiple testing into account (FDR), no significant differences are found.
The second series of tests aimed to determine whether the rats in either of the GMO and/or ROUNDUP groups died earlier than the rats in the control group. The ECEAG used the Log-Rank test for this purpose. This test compared survival probabilities for the various groups. Three series of comparisons were conducted successively: Control vs. GMO alone, Control vs. GMO treated with ROUNDUP WEATHER MAX, Control vs. ROUNDUP GT PLUS. The tests were undertaken with and without correction (Sidak correction) for the number of comparisons made per series using the raw mortality data submitted by the author after his hearing. The results (Table 2) show that there are two significant differences out of 18 with the uncorrected tests:
· for the female group, GMO at the 22% dose,
· and for the female group, GMO + R at the 22% dose.
Taking into account the adjustment, no difference is significant.
English:
The third series of tests aimed to determine whether the frequency of pathologies was higher in the GMO and/or ROUNDUP groups than in the control group. These tests compared the null hypothesis H0 „Frequency of pathologies in the control group = Frequency of pathologies in the GMO and/or ROUNDUP groups' with the alternative hypothesis A „Frequency of pathologies in the control group ; Frequency of pathologies in the GMO and/or ROUNDUP groups‟. These tests were undertaken using Table 2 of the study by Séralini et al. (2012) for the six listed pathologies with the six GMO treatments (three doses of GMO without ROUNDUP + three doses of GMO with ROUNDUP WEATHER MAX) and the three ROUNDUP GT PLUS treatments. The Type I error risk was calculated separately for each pathology and each diet with Fisher‟s exact test. The results are shown in Table 3 below.
Of the 54 comparisons, five are significant at a level of 5% before FDR correction.
*„hepatic pathologies‟ described by the author as liver congestions, macroscopic spots and microscopic necrotic foci
o for the males in the group fed 22% GMO,
o for the males in the RB group.
* mammary tumours
o for the females in the RB group.
*pathological signs in the mammary glands (other than tumours described by the authors as galactoceles and mammary hyperplasias)
o for the females in the RA group,
o for the females in the RB group.
After FDR correction for multiple testing, there are no significant differences at the 5% level.
Table 3. Results of statistical tests (P = probability of error of the first kind) on disease incidence (percentage of animals infected with at least one tumor lesion or pathology). The odds were not adjusted for multiplicity of tests. If these corrections are applied, no difference is significant at 5%.
In general, more specific information about the observed pathologies is required to determine the biological significance of the statistical results before correcting for multiple testing as recommended, particularly in the ANSES report (ANSES 2011). The ECEAG considers it unfortunate that the definitions of the groups of pathologies described in the publication are unclear and that there are so few useable biochemical data. Nonetheless, the statistical analysis results as a whole show that:
* The increased mortality and reduced life expectancy (Tables 1 and 2) observed for the females in the GMO 22% and GMO 22% + R groups (before correction) are not confirmed by any underlying pathologies (Table 3). This finding is striking and additional information on the cause of death for each animal in these groups would be necessary to interpret it.
*The increase in pathologies highlighted in the publication is significant at a level of 5% (before correction) for only a small number of treatments and is difficult to interpret from a biological standpoint due to the unclear definitions of the pathologies.
* The increase in the incidence of hepatic pathologies in the „GMO 22% male' group (before correction) is not found at the 11% and 33% doses nor for the „GMO 22% + R male‟ group. This result does not appear coherent since it occurs at a single intermediate dose and is not found in a group fed the same percentage of GM maize.
*The increase in hepatic pathologies observed in males for the RB dose of ROUNDUP GT PLUS (before correction) may be consistent with the LOAEL for glyphosate. However, the other pathologies (mammary tumours, galactoceles and mammary hyperplasia) observed in females at the RB dose do not appear consistent with the toxicological data on glyphosate (long-term studies in rodents). Furthermore, none of these effects are found at the highest dose (RC). This finding does not support biological coherence even though it could be expected that this high dose would interfere with the eating behaviour of rats. It would be useful to have data on the water and feed consumption of the treated animals.
* The significant increase in the frequency of mammary gland pathologies (excluding tumours) (Table 3) observed at the lowest dose of ROUNDUP GT PLUS (RA) (before correction) caught the attention of the ECEAG and may suggest an unexpected effect at a very low dose. However, in order to determine a biologically significant effect, it is necessary to have individual data, comprehensive biochemical data and historical data on the SD strain provided by the CRO.
See also other posts
No comments:
Post a Comment