Robustness of a multivariate composite score when evaluating distress of animal models for gastrointestinal diseases.

The fundament of an evidence-based severity assessment in laboratory animal science is reliable distress parameters. Many readouts are used to evaluate and determine animal distress and the severity of experimental procedures. Therefore, we analyzed four distinct parameters like the body weight, burrowing behavior, nesting, and distress score in the four gastrointestinal animal models (pancreatic ductal adenocarcinoma (PDA), pancreatitis, CCl intoxication, and bile duct ligation (BDL)). Further, we determined the parameters' robustness in various experimental subgroups due to slight variations like drug treatment or telemeter implantations. We used non-parametric bootstrapping to get robust estimates and 95% confidence intervals for the experimental groups. It was found that the performance of the readout parameters is model-dependent and that the distress score is prone to experimental variation. On the other hand, we also found that burrowing and nesting can be more robust than, e.g., the body weight when evaluating PDA. However, the body weight still was highly robust in BDL, pancreatitis, and CCl intoxication. To address the complex nature of the multi-dimensional severity space, we used the Relative Severity Assessment (RELSA) procedure to combine multiple distress parameters into a score and mapped the subgroups and models against a defined reference set obtained by telemeter implantation. This approach allowed us to compare the severity of individual animals in the experimental subgroups using the maximum achieved severity (RELSA). With this, the following order of severity was found for the animal models: CCl < PDA ≈ Pancreatitis < BDL. Furthermore, the robustness of the RELSA procedure and outcome was externally validated with a reference set from another laboratory also obtained from telemeter implantation. Since the RELSA procedure reflects the multi-dimensional severity information and is highly robust in estimating the quantitative severity within and between models, it can be deemed a valuable tool for laboratory animal severity assessment.