Statistical approach for avoiding pseudoreplication and increasing power in wound-healing studies.

ABSTRACT

Many animal wound-healing models measure the progression of healing over time, resulting in counts of fully healed wounds from different treatments at several time points. Data from these models are usually analyzed using contingency table methods. However, pooling data from multiple animals without appropriate correction for animal-to-animal variability results in pseudoreplication. Kaiser and colleagues, overcame pseudoreplication by adjusting the estimate of healing variation to account for the interanimal covariance. This solution nevertheless is limited by the ability to accurately estimate the adjustment factor due to the small number of animals used. An improved method is described that both overcomes pseudoreplication and increases power. It involves estimating the time for half of the wounds within each animal to be completely healed (TCH(50)), rather than a pooled estimate for all animals (HT(50)). Subsequent ANOVA testing of the individual TCH(50) values, using a model with fixed treatments and random animals, generates unbiased estimates of treatment means and differences between means, and accurate p-values for these differences and for the overall model. This method has sufficient power to detect treatment differences with fewer animals. Furthermore, it is fully applicable to analyses of results from human trials having similar data organization.