Increasing the statistical power of animal experiments with historical control data

• Published in: Nature neuroscience Vol. 24; no. 4; pp. 470 - 477
• Main Authors: Bonapersona, V., Hoijtink, H., Sarabdjitsingh, R. A., Joëls, M.
• Format: Journal Article
• Language: English
• Published: New York Nature Publishing Group US 01.04.2021; Nature Publishing Group
• Subjects: 631/114/2415; 631/378/1595/1554; 631/378/1831; 692/308/1426; 706/648/697; Animal Experimentation; Animal Genetics and Genomics; Animals; Bayesian analysis; Behavioral Sciences; Biological Techniques; Biomedical and Life Sciences; Biomedicine; Cognitive ability; Computer Simulation; Control data (computers); Ethical standards; Experiments; Methods; Mice; Models, Statistical; Neurobiology; Neurosciences; Reliability; Repair; Research Design; Sample Size; Source code; Statistical methods; Statistical power; Statistical software; Statistics; Netherlands
• ISSN: 1097-6256; 1546-1726
• DOI: 10.1038/s41593-020-00792-3

Low statistical power reduces the reliability of animal research; yet, increasing sample sizes to increase statistical power is problematic for both ethical and practical reasons. We present an alternative solution using Bayesian priors based on historical control data, which capitalizes on the observation that control groups in general are expected to be similar to each other. In a simulation study, we show that including data from control groups of previous studies could halve the minimum sample size required to reach the canonical 80% power or increase power when using the same number of animals. We validated the approach on a dataset based on seven independent rodent studies on the cognitive effects of early-life adversity. We present an open-source tool, RePAIR, that can be widely used to apply this approach and increase statistical power, thereby improving the reliability of animal experiments. Bonapersona and colleagues describe how historical control data can be used to improve statistical power while reducing the number of animals required in experiments. They present an open-source tool, RePAIR, that can be used to apply this approach.