Animal experiments must be well designed, efficiently executed, correctly analyzed, clearly presented, and correctly interpreted to be ethically acceptable. This page sets out the principles of good design.
The prototype experiment
Need for clearly stated objectives:
Need for a wide range of applicability
Being amenable to statistical analysis
A fancy full lop rabbit (these are rarely used in medical research)
The most basic experiment would involve taking two animals (or other subjects) which are as similar as possible in every respect. One, assigned at random, would receive a “treatment” the other would be kept as a “control” without the treatment. Any difference between them following the treatment is assumed to be caused by that treatment.
But no two animals are exactly alike, so several similar animals must be used to average out any individual differences.
Statistical methods determine the probability that any observed difference between the two groups could be due to the chance allocation of these slightly different animals to each group. If this probability is low (say less than 5%), then it is assumed that any observed difference is due to the treatment.
Optimum group size depends on the degree of similarity between animals and the size of the treatment effect.
There are many motives for doing an experiment, such as the need to do a pilot experiment to test logistics and gain preliminary information or to test a hypothesis about the effect of a treatment or explore the effect of some treatment or estimate parameters such as a dose-response relationship or group means (see Purpose of Study)
No group of individuals should be treated differently from any other (except for the applied interventions), as this may cause bias.
No group should be dosed or measured first, or by a different person. Dosing should be in random order
No group should be housed in a different animal room
No group should be on a different shelf within the animal room. Positions should be randomised.
The “experimental unit” (see Experimental Unit) must be correctly identified and be appropriately randomised to a treatment group. Such randomisation should continue throughout the experiment (see Randomisation).
Where possible investigators should be blind with respect to treatment group, particularly if there is any subjective element in assessing the outcomes, such as in the reading of histological slides (see Blinding).
An experiment should be powerful. It should have a good chance of detecting the effect of a treatment, if there is such an effect. High power is achieved by having a high “signal/noise” ratio. The experimental material needs to be sensitive to the treatment (high signal). Sensitive species and strains should be used or prior treatment may increase sensitivity. Where the level of sensitivity is unknown, preliminary experiments might be used to identify appropriate strains. Multi-strain experiments using a factorial experimental design might be used (see Experimental designs/Factorial experiments). These will at least reduce the likelihood that a resistant strain or stock is being used.
High power also depends on the “noise”, so uniform experimental material should be used. Animals need to be free of disease, and isogenic strains should be used wherever possible (see Uniformity). Where such material is not available, or the material has some sort of natural structure, such as coming in litter groups, then a design involving blocking may be appropriate (see Experimental designs/Blocking).
Power also depends on sample size. Other things being equal, the larger the experiment the greater the power but large experiments cost money, time and animals so increasing sensitivity and reducing noise should be considered first. Where possible sample size should be determined using an objective method such as power analysis or the resource equation method (see Sample size).
It is usually desirable to find out whether the results are applicable over a wide range of conditions such as both sexes, several strains, different ages, different environments and prior treatments. Responses which are only seen in old males of strains X fed diet A are much less likely to be of interest. While there is no way of assuring this, it is possible to discover what other variables affect the outcome of the experiment at little or no extra cost and without using more animals by using factorial experimental designs (see Experimental designs/factorials).
Mistakes need to be avoided at all costs, so the experiment should not be so complex that these are likely to occur. Pilot studies should be used wherever possible to ensure that the experiment is logistically sound (see Purpose of Study). Written protocols and standard operating procedures should always be used. The materials and methods section of any planned paper could be drafted at this time. An experimental diary should be kept and any unusual incidents recorded such as if a data point is suspect. The formal design of the experiment should be identified (i.e. between subjects, completely randomised, randomised block etc (see Experimental designs).
The type of data to be collected (e.g. measurement, qualitative, binary etc.) should be identified (see Outcomes) and the method of statistical analysis of the resulting data should be decided before the experiment is started. It may be necessary to modify the analysis, such as when a scale transformation is necessary.
Anyone starting an experiment should know in general terms how the resulting observations are to be treated (see Experimental designs/ANOVA).