How to calculate sample size in animal studies?

This is an open-access article distributed under the terms of the Creative Commons Attribution-Noncommercial-Share Alike 3.0 Unported, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.

Abstract

Calculation of sample size is one of the important component of design of any research including animal studies. If a researcher select less number of animals it may lead to missing of any significant difference even if it exist in population and if more number of animals selected then it may lead to unnecessary wastage of resources and may lead to ethical issues. In this article, on the basis of review of literature done by us we suggested few methods of sample size calculations for animal studies.

Keywords: Alpha error, animal studies, power, sample size

How many animals I should use for my study? This is one of the most confusing questions faced by a researcher. Too small sample size can miss the real effect in experiment and too large sample size will lead to unnecessary wasting of the resources and animals.[1] Issue of sample size has been highlighted adequately for the clinical trials and clinical studies, but not explored much in the case of animal studies in published literature. It is very important to teach young researchers and post-graduate students regarding importance and methods of sample size calculation. To clarify this issue of sample size in animal studies, we decided to search various articles available regarding the sample size in animal studies. We did PubMed search by using various MeSH terms such as “sample size,” “sample size calculations,” “animal studies” etc., and their combinations. We have also searched various articles through Google and Google Scholar. We have also searched various websites related to animal research (http://www. 3rs-reduction.co.uk/html/6__power_and_sample_size.html, http://www.acuc.berkeley.edu/, http://www.bu.edu/orccommittees/iacuc/policies-and-guidelines/sample-size-calculations/, http://www.ucd.ie/researchethics/etc.). First author read all available literature and an understanding about the concept is made in consultation with the second author. Here, we are explaining briefly about the method of sample size calculations in animal studies based on review of the literature carried out by us.

Basically, there are two methods of sample size calculation in animal studies. The most favored and most scientific method is calculation of sample size by power analysis.[2] Every effort should be carried out to calculate sample size by this method. This method is similar to the method used for calculation of sample size for clinical trials and clinical studies. Simple calculation can be carried out manually with the help of some formula [Appendix 1], but for complex calculations statistical software can be used or help from a statistician can be sought. To calculate the sample size by power analysis a researcher must have knowledge and information about these concepts:

Effect size: This is the difference between the mean of two groups (quantitative data) or proportions of events in two groups (qualitative data). A researcher should decide before the start of the study that how much minimum difference between two groups can be considered as clinically significant. The idea about clinically significant difference between the groups should be taken preferably from previously published studies[2,3,4,5]

Standard deviation: Standard deviation measures variability within the sample. Information about standard deviation is needed only in the case of quantitative variables. Information about the standard deviation of a particular variable can be taken from previously published studies. If no such study is available then author should conduct a pilot study first and standard deviation can be calculated from the pilot study[2,3,4,5]

Type 1 error: This is measured by significance level, which is usually fixed at the level of 5% (P = 0.05). This is an arbitrary value and can be decreased or increased according to the research question[2,3,4,5]

Power: Power of a study is probability of finding an effect, which the study is aimed to find. This may be kept between 80% to even 99% depending on research question, but usually, it is kept at 80%[2,3,4,5]

Direction of effect (one tailed or two tailed): When a researcher wants to explore the effect of some intervention, the actual effect observed in sample may be in same direction as researcher thought or it may be just opposite to that. If researcher feels that effect may be in both directions then he should use two tailed test and if he has strong reason to believe for the effect to lie in one direction then he can use one tailed test. In animal research, two tailed tests are usually used[2]

Statistical tests: For sample size calculation, it is important to have an idea about statistical test, which is to be applied on data. For simple statistical tests such as Students t-test or Chi-square test, manual calculation based on formula can be carried out [Appendix], but for complex tests like ANOVA or non-parametric tests help of statistician or use of software is needed[2,4]

Expected attrition or death of animals: Final sample size should be adjusted for expected attrition. Suppose a researcher is expecting 10% attrition then the sample size calculated by formula or software should be divided by 0.9 to get actual sample size. Suppose sample size calculated by software is 10 animals per group and researcher is expecting 10% attrition then his final sample size will be 11 animals per group (10/0.9 = 11.11). Similarly, for 20% attrition sample size should be divided by 0.8.[5] This can be explained in the form of structured formula i.e.,

Corrected sample size = Sample size/ (1− [% attrition/100])

We suggest use of freely downloadable software G Power (Faul, Erdfelder, Lang and Buchner, 2007) for sample size calculation. This software is equally good for sample size calculation for clinical trials also. This software can be used for simple as well as complex sample size calculations.[6] G Power can calculate sample size based on pre-designed effect size at small, medium, and large difference between the groups based on Cohen's principles.[7] Information about other freely available software and calculators for sample size calculation is given in Appendix 2. More complex sample size will need more sophisticated software such as “nQuery advisor” or “MINITAB.”

Second method of calculation is a crude method based on law of diminishing return. This method is called “resource equation” method.[2,8,9] This method is used when it is not possible to assume about effect size, to get an idea about standard deviation as no previous findings are available or when multiple endpoints are measured or complex statistical procedure is used for analysis. This method can also be used in some exploratory studies where testing of hypothesis is not the primary aim, but researcher is interested only in finding any level of difference between groups.

According to this method a value “E” is measured, which is nothing but the degree of freedom of analysis of variance (ANOVA). The value of E should lie between 10 and 20. If E is less than 10 then adding more animals will increase the chance of getting more significant result, but if it is more than 20 then adding more animals will not increase the chance of getting significant results. Though, this method is based on ANOVA, it is applicable to all animal experiments. Any sample size, which keeps E between 10 and 20 should be considered as an adequate. E can be measured by following formula:

E = Total number of animals − Total number of groups

Suppose a researcher wants to see the effect of a drug and he made five groups (one group as control and four groups of different doses of that drug) with 10 rats each. In this case E will be

E = 50 − 5 = 45, which is more than 20 hence sample size in this experiment is more than necessary. However, if sample size is five per group then E will be 20, which is the acceptable limit and hence can be considered as adequate sample size.

This method is easy, but it cannot be considered as robust as power analysis method.

We want to suggest researchers to include a statement about method of calculation of sample size and justification of sample size in the manuscript they want to publish. Animals in research: Reporting in vivo experiments guideline recommends inclusion of a statement mentioning justification of the sample size used in research and detail of method of calculation of sample size.[10] All components of sample size calculation such as effect size, type 1 and type 2 error, one tailed/two tailed test, standard deviation etc., should be reported in manuscript sent for publication the way it is suggested for the clinical research.[11] Shortage of resources (budget, manpower), time constrain etc., cannot be considered as valid justification regarding decision of sample size. Many researchers consider six animals per group as adequate sample size, but after reviewing available literature on this issue we came to a conclusion that this notion of six animals per group has little scientific and statistical basis. This is a brief description and readers are requested to read more resources available for better understanding of various concepts related to the sample size calculation in animal studies.

ACKNOWLEDGMENT

We want to acknowledge unknown reviewer for constructive comments and guidance for improvisation of this manuscript.