New GRACE publication: Improved statistical approaches for better interpretation of feeding trial data

Display a printer-friendly version of this page

New GRACE publication: Improved statistical approaches for better interpretation of feeding trial data

In this paper*, the scientists applied a new approach to statistically analyse data from a toxicity study. Compared with traditional methods, the methods used here offer a double benefit: they handle and graph the large number of characteristics in a more comprehensive and consolidated way, making it easier to assess whether any significant differences can be considered biologically relevant or not.

There are several guidelines and publications dealing with the statistical treatment of toxicity study data. Most of the guidelines favour a traditional approach, which simply asks ‘Is there an effect?’, while other more recently published papers promote the reporting of effect sizes and confidence intervals and ask ‘How much of an effect is there?’ The latter approach can help to avoid the yes/no decision trap of statistical tests: classifying something as statistically significant does not mean it is biologically relevant. In any evidence-based decision-making situation, biological relevance should always be given preference over statistical significance.

The new approaches applied by GRACE can facilitate the interpretation of toxicity study data to determine whether an observed effect should be considered biologically relevant or not. Standardized effect sizes are calculated for all endpoints – weight, relative organ weights, haematology and clinical biochemistry – and displayed simultaneously in a graph, providing an overall picture of group differences and their relevance at a glance.

Furthermore, a linear mixed model (LLM) approach is used to analyse weight or feed consumption data. Unlike traditional ANOVA approaches, which provide multiple test results per week, the LMM results in only one statement on differences in weight development between groups. The LMM therefore allows these data to be analysed in a more comprehensive and consolidated way and facilitates interpretation.

This new approach enhances transparency and could support consensus between all actors involved in a decision-making process, namely toxicologists, statisticians and regulators. Furthermore, it helps communicate study results to the public in a way that is more easily understood.

 

* Enhancing the interpretation of statistical P values in toxicology studies: implementation of linear mixed models (LMMs) and standardized effect sizes (SESs). K. Schmidt et al., Archives of Toxicology, 2015, 10.1007/s00204-015-1487-8