Control of Echinococcosis is an important, though not always straightforward issue. The potential use of a vaccine in dogs is a welcome development as such as vaccine could make a significant impact in reducing the biomass of E. granulosus in endemic areas. There is much epidemiological evidence to suggest that canids, following natural exposure, develop some sort of protective immunity against reinfection (reviewed in [1]). However, this hypothesis is still unproven. Petavy and co workers in their recent publication [2] claim to have developed a recombinant vaccine that has efficacy against infection and growth of E. granulosus in dogs. Unfortunately there are a number of problems with the design of the experiment and the statistical analysis that seriously undermine this claim. First there is little information about the experimental dogs other than they were purchased locally in Tunisia and Morrocco and they were up to 6 months of age. Even relatively young puppies of just a few months of age can become infected and Echinococcus is an extremely common parasite in these countries. In addition, no mention was made of any attempt to ensure uniformity of breed or type. Individual animals are notorious in their responsiveness to helminth infections. It has long been known that individual sheep, for example, have a large difference in their response to nematode infections even when they are of the same breed (often termed responders and non responders [3]) and it is likely that dogs infected with cestodes will behave likewise. Indeed evidence for this can be seen with experimental infections with E. multilocularis [4]. Because of these issues it is essential to have a reasonable sample size between experimental groups in order to come to logical conclusions. Indeed there is evidence of extremely high variability in the parasite establishment of the 6 control dogs (table 2). From the figures in the table it can be calculated that the numbers of worms found in the six dogs at necropsy were: 390, 6690 and 8640 (from Morocco) and 160, 5,652 and 6,500 from Tunisia. Thus there were 2 out of 6 dogs where there were very low numbers of parasites recovered from the non vaccinated dogs. This suggests that 1 in 3 naïve dogs (if they are naïve) will only have a low parasite burden even when given a large inoculating dose of protoscolices. If this argument is extended to the salmonella and vaccinated groups, it quickly becomes apparent that the chance finding of 2 or 3 animals with such low counts is unremarkable. For example the salmonella alone group from Morocco has burden of 265, 660 and 3570 worms. Thus 2 of the 3 dogs had low burdens. But with the expected numbers of low burdens being 1 in 3, the p value of such an observation is maybe 22%. Far above the significance cited in the article. The problem is because the authors have small sample sizes and use a parametric statistic. Dunnett’s test relies on the underlying distribution of the comparison groups to be normal [5]. With parasite data, even in experimentally infected animals this is rarely the case [6] . Indeed this can be seen by looking at the mean and variance of the control groups. The variance is much bigger than the mean, indicating the data does not even approach a Poisson distribution (the closest thing to a normal distribution for count data). To overcome this problem, it is necessary to employ other techniques such as using other distributions, maximum likelihood analysis or bootstrapping [6]. It then quickly becomes apparent that the observed values in the salmonella or vaccine group could easily have arisen by chance, given the observed distribution of the control groups. The same arguments can be used for the relative sizes of the worms. If the majority of dogs with a better innate response were randomly assigned to the vaccine groups (possible with a small sample size), then this may account for the smaller worm size in the vaccinated animals. This does not mean to say the vaccine does not work. It may do, but at the moment there is insufficient data to support the conclusions of the article. It is accepted that such experiments with dogs are a challenge to undertake and are expensive. However an experiment where there is little chance of producing interpretable results is even more expensive. This is a pity as such incorrect statistical analysis and experimental design risk derailing a very important line of research that could produce a product that would revolutionise Echinococcus-control

References

1. Torgerson, P.R., Canid immunity to Echinococcus spp.: impact on transmission. Parasite Immunology, 2006. 28(7): p. 295-303.
2. Petavy, A.F., et al., An Oral Recombinant Vaccine in Dogs against Echinococcus granulosus, the Causative Agent of Human Hydatid Disease: A Pilot Study. PLoS Neglected Tropical idseases, 2008. 2: p. e125.
3. Riffkin, G.G. and C. Dobson, Predicting resistance of sheep to Haemonchus contortus. Veterinary Parasitology, 1979. 5: p. 365-378.
4. Kapel, C.M.O., et al., Reproductive potential of Echinococcus multilocularis in experimentally infected foxes, dogs, raccoon dogs and cats. International Journal for Parasitology, 2006. 36(1): p. 79-86.
5. Cheung, S.H. and B. Holland, Extension of Dunnett's multiple comparison procedure to the case of several groups. Biometrics, 1991. 47: p. 21-32.
6. Torgerson, P.R., M. Schnyder, and H. Hertzberg, Detection of anthelmintic resistance: a comparison of mathematical techniques. Veterinary Parasitology, 2005. 128(3-4): p. 291-298.