Ch. 6— Alternatives to Animal Use in Research • 115 
Using animals maximally in a confined area is 
a mandatory part of experimental design in the 
research program of the National Aeronautics and 
Space Administration (NASA). Protocols typically 
call for investigators to combine projects and make 
efficient use of one small group of animals (125). 
Improved Experimental or 
Statistical Design 
“Every time a particle of statistical method is 
properly used, fewer animals are employed than 
would otherwise have been necessary,” wrote Rus- 
sell and Burch some 27years ago (174). Since then, 
progress has been made both in the number of 
statistical tools available and in the training of in- 
vestigators in the use of these tools. Yet training 
still lags behind the availability of tools. Insuffi- 
cient information for critical evaluation and inap- 
propriate statistical analyses appear frequently in 
the literature, particularly with investigators using 
the t-test in cases for which analysis of variance 
is the appropriate measure (82). 
An analysis of variance simultaneously tests two 
or more parameters of treatment groups for indi- 
cation of significant difference. When the test sta- 
tistic falls in the rejection region, the researcher 
can be reasonably sure that a real difference ex- 
ists between treatments. The t-test estimates the 
difference between the mean values of one param - 
eter of two treatments. It is a powerful measure 
of significance when the number of comparisons 
is small, but it is subject to an increasingly large 
potential for error as the number of parameters 
grows. Using multiple t -tests increases the risk of 
finding a significant difference between treatments 
where there is none. Such observations are not 
esoteric, since poor summarization and statistical 
usage may reflect poor experimental design, call- 
ing into question the results of an investigation 
and leading to otherwise unnecessary repetition. 
At least one group, the Harvard Study Group on 
Statistics in the Biomedical Sciences, is pursuing 
ways to improve statistical practice and report- 
ing (64). 
Serial sacrifice, crossover, and group sequen- 
tial testing are three experimental designs that can 
reduce animal use in laboratory research (82). In 
serial sacrifice, animals with induced effects are 
randomly selected for sacrifice and examination 
for the occurrence and progress of effects over 
time. Such studies, as in radiation oncology (22), 
have the dual advantage of cutting short the time 
some animals must spend in an affected state and 
providing information about changes within the 
animal other than those observed when it is al- 
lowed to die without further interference. The pri- 
mary disadvantage is that survival information is 
compromised; therefore, the resulting data can- 
not be compared with other studies in which sur- 
vival serves as an end point. 
A crossover design may be appropriate for stud- 
ies in which short-term effects are expected. Each 
animal serves as its own control by first receiving 
either a drug or a placebo, and then receiving the 
reverse. Such a design can be highly useful in lab- 
oratory and clinical testing, but crossovers must 
be used judiciously. Should there be any unex- 
pected long-term effects, the entire test is invali- 
dated and would need to be repeated as two sepa- 
rate tests. 
In the group sequential design, treatment groups 
are compared with each other in stages. For ex- 
ample, if two groups are given the same dosage 
of two different drugs, experimentation at higher 
dosages is undertaken only if there is no statisti- 
cally significant difference between the responses 
of the two groups. The sooner a difference be- 
tween groups is observed, the fewer the number 
of trials run. Both crossover and group sequential 
designs have potential applications in anesthesiol- 
ogy, endocrinology, nutrition, pharmacology, ra- 
diology, teratology, and toxicology. 
A commonly mentioned method of reducing the 
number of animals used is smaller treatment 
groups. Yet within the biomedical research com- 
munity a frequently heard complaint is that too 
few animals to yield useful estimates are likely to 
be included in each treatment group, particularly 
in fields such as radiology (95). Problems of this 
nature generally grow out of the extreme economic 
pressures being applied to investigators to con- 
trol animal costs. Well-established techniques such 
as saturation analyses, particularly radioimmuno- 
assays, have radically reduced the number of ani- 
mals used for any one procedure, but they may 
have resulted in little or no reduction in overall 
