Beispiel, wie man es nicht machen sollte.

Der folgende Text ist dem Kapitel 2 des Buches Rasch. D.. Pilz. J. and Šimeček P. 2010. Optimal Design of Experiments with R. Boca Raton. Chapman and Hall entnommen.

The following text in italics was taken from an information of the Ministry of Health in Austria entitled:

SCIENTIFIC ARGUMENTS FOR AN IMPORT BAN OF GENETICALLY

MODIFIED MAIZE MON 863 (Zea mays L., line MON 863) OF MONSANTO

(NOTIFICATION C/DE/02/9)

On 8th August 2005 the Decision (2005/608/EC) concerning the placing on the market, in accordance with Directive 2001/18/EC of genetically modified maize MON 863 was adopted by the Commission. The product may be placed on the market and put to the same uses as any other maize, with the exception of cultivation and uses as or in food.

On 13th January 2006 the placing on the market of foods and food ingredients derived from genetically modified maize line MON 863 as novel foods or novel food ingredients under Regulation (EC) No 258/97 was authorised.

With regard to the studies on nutritional equivalence assessment in farm animals, which are quoted in HAMMOND et al. (2006) as scientific proof for the safety of maize MON 863, a lot of shortcomings have been detected:

In this document, the scientific arguments, which are justifying the Austrian import ban of this GMO, are described. They focus particularly on the toxicological safety assessment and the antibiotic resistance marker (ARM) gene nptII, which is contained in maize MON 863, but also on the given risk management measures to prevent accidental spillage.

Summarizing the evaluation of the toxicological safety assessment of the dossier, it can be stated that a lot of deficiencies are obvious:

Concerning the experimental design it has to be criticised that reference groups are often contributing 60-80% of the sample size. Statistically significant differences between test and control groups are therefore often masked because group differences between iso- and transgenic diets fall into the broad range of reference groups.

An important factor is also the sensitivity of the animal model: HAMMOND et al. (2006) described the use of an outbred rat model. The study compared a high number of different lines of maize, among them MON863. The data vary considerably in and between the groups. That would allow the assumption that only effects with great deviations from the control would have been detectable with the chosen trial setup.

So far a part of the official text of the ministry. Let us consider the design of Hammond et al. (2006) without the reference groups Then we have two control groups of rats fed with grain free of MON 863 and two groups fed with MON 863 one of them with 11% and one with 33% MON 863. The size of each group was n = 20.

Now let us assume that we wish to test the null hypothesis that there is no decrease in fecundity of rats in the treatment groups compared with the control groups. That means that there is no argument (with regard to fecundity) not to use MON 863. Let us further assume that we use a risk of the first kind of 0.1 or alternatively 0.05 (the probability that we reject the null hypothesis if it is true, the producer risk of Monsanto) and at first a risk of the second kind of 0.01 (the probability that there is a decrease of fecundity even it is present) if the decrease is 2% or larger. Then we will find out by the method described in 2.4.2.2., assuming that the usual fecundity is around 0.8, and that in each of two groups to be compared, the group size should be 10893. An overview about the association between risk of the first and second kind, group size and minimum detectable decrease (from 0.8) as calculated by R is given below (one-sided alternative hypothesis):

Risk b	0.8 - d	Size of each group; a =0.1	Size of each group; a =0.05
0.01	0.78	10893	13177
	0.70	504	607
	0.60	142	171
0.05	0.78	7201	9074
	0.70	338	423
	0.60	97	121
0.1	0.78	5547	7201
	0.7	264	339
	0.60	77	98
0.2	0.78	3338	5277
	0.70	188	250
	0.60	56	74

With 20 animals and with a= 0.1 (one-sided) and a high error rate of 100*b = 20% the null hypothesis of no negative effect is accepted as long as the real fecundity in the group treated with MON 863 lies above 0.4375. But for a= 0.1 (one-sided) and 100*b = 1% a= 0.1 (one-sided) and as long as the real fecundity in the group treated with MON 863 lies above 0.1625. Of course such risks are totally unacceptable.