Statistical methods and models for standardized toxicological and safety pharmacological studies.

Project no.: 1214
Ph.D. project: ATV - Novo Nordisk A/S
Ph.D. student: Helle Andersen
During the last couple of years, a group of scientists at Novo Nordisk have developed a decision tree for the statistical analysis of toxicity data from studies with animals.
Unfortunately the decision tree has been found to lead to inconsistencies as variables are often analyzed differently from study to study. Furthermore, it does not contain a test battery for the analysis of the dose-response relationship, or recommendations for the analysis of repeated measurements. An other shortcoming of the decision tree is that it does not contain recommendations for the statistical analysis of safety pharmacology data.
My project has been initiated to overcome these problems. The purpose of the statistical work in the project is to evaluate existing basic experimental designs and matching statistical models in toxicological studies where statistical methodology has already been applied.
The purpose of the statistical work is to establish a "knowledge data base" where experimental designs and empirical knowledge about biological variables determine the statistical model, and hence the statistical analysis. There is some empirical knowledge in the following areas (among others):
- Transformation of data
- Distribution of variables
- Statistical tests for outliers
- Statistical tests for homogeneity of variance
- Statistical analysis of single and correlated variables
- Statistical considerations of repeated measurements on individual animals
In this way, variables will be analyzed identically from study to study, i.e. the statistical method will be identical for the same variable independently of study. But at the same time, statistical methods will be established to spot abnormalities (outliers) which could indicate some (important) adverse biological response.