(whistling wind) (ding) – I recently attended a
departmental research seminar and the speaker described
a study that sought to identify microbiome effects
of a treatment in mice. There were 20 mice in
a total of four cages. Each cage was randomized to
receive one of two treatments, either antibiotics delivered in the water, or just plain water. At the end of the treatment period, they did microbiome analyses. As she showed the results, she noted that the control mice were very tight with little variability,
however, in the active treatment, the mice seemed to cluster
into two different groups. The audience of other faculty and residents began hypothesizing about why such a difference could occur, but I asked whether the
results clustered by cage. In fact, they did. The speaker had group-housed the mice, so five animals in each of four cages. What was different about the
cages is not entirely clear. Gut microbes have been shown
to be altered by a large array of factors, and certainly
a shared environment would encourage sharing the microbiota. For instance, one group
of mice may have consumed more antibiotic water than
the other group of mice, but consumption was not
quantified in the study. In addition, in one of the
cages, a mouse was diagnosed with a dry tail, and an
appropriate animal-care standard for treating it was application
of an approved moisturizer. Mice are coprophagic,
meaning they eat feces. If there was something
different about a mouse or mice in one cage, it could spread
to others within the cage. I could think of no better
example of the challenges of group-housing animals
than this, particularly when the outcome can so
easily be contaminated between animals. In the end, she essentially
had n equals two in each group, two cages. Mice were not independent,
only the cages were. – So how can we learn from these mistakes? From a statistical point of view, when treatments are
applied to cages instead of individual animals,
individually housing animals would provide the greatest power. It maximizes the number
of experimental units, but, of course, how animals
are housed depends on more than just statistics, including
animal-care protocols, the design of the experiment, and the exact question being asked. There are at least three
things to consider, then, when conducting experiments
with group-treated animals. First, recognize that the unit of randomization is the group,
not the individual mouse. Second, when designing the experiment, recognize that power and
sample size calculations are functions of cages, not just mice. The number of cages typically
has a much larger effect on the power of the study
than does the number of mice. Third, be sure to incorporate
cages in group housing into the statistical model. Otherwise, the statistical
analyses may tend to reject more null
hypotheses than they should. And of course, if you have any questions, consult your friendly
neighborhood statistician.

Leave a Reply

Your email address will not be published. Required fields are marked *