Sample size and statistical power in the small-animal analgesia literature

OBJECTIVE: To document the power and required sample sizes to achieve certain treatment objectives in the veterinary analgesia literature. METHODS: Pubmed's MEDLINE database and selected journals were searched. Only publications produced between 1994 and 2004 that reported 'no difference' between experimental groups in the abstract, results or conclusion sections and those that were randomised, prospective and blinded were reviewed. The data reported in the publications were then subjected to power analyses to determine the power and necessary sample size (to achieve a power of 0.8) to allow detection of 20 per cent, 50 per cent and 80 per cent treatment effects. RESULTS: Twenty-two studies provided sufficient data for analysis. Five out of 22 (23 per cent) had sufficient power to detect a 20 per cent treatment effect, 12 of 22 (54 per cent) had sufficient power to detect a 50 per cent treatment effect and 18 of 22 (82 per cent) had sufficient power to detect an 80 per cent treatment effect. The mean number of animals required per group to document a 20 per cent, 50 per cent and 80 per cent treatment effect were 90, 15 and 7, respectively. CLINICAL SIGNIFICANCE: Publications that report no significant difference between analgesic regimens may have committed a Type II error. The reader may inappropriately conclude that there is no difference between treatments when there may, in fact, be a superior analgesic regimen. Clinical practice based on the principles of evidence-based medicine could therefore result in suboptimal care for patients.