Refining humane endpoint detection by time-series forecasting and threshold definition using a multivariate severity score

Le résumé fourni par la source

Severity assessment in animal-based research is not only mandatory but also requires the identification of robust, objective, and model-specific parameters. The RELative Severity Assessment (RELSA) procedure enabled multivariate severity assessment, allowing interpretation of outcomes in the context of a predefined severity grading. This study expands the application of the RELSA score by combining it with an Autoregressive Integrated Moving Average (ARIMA)-based prediction model, which we call foRcast. To facilitate interpretation of severity, a kernel density estimate (KDE) was applied to define exploratory thresholds on the RELSA scale. Data of mice reaching humane endpoints across models representing a broad range of severity profiles were reanalyzed using transmitter-derived parameters (e.g., heart rate) and behavioral parameters (e.g., voluntary wheel running), depending on the model. The primary focus was predicting RELSA scores at the humane endpoint using foRcast. With an overall root mean square error of 0.069 and 96% of actual RELSA scores falling within the 95% prediction interval of the forecasts, this tool showed promising preliminary performance. KDE was performed on sepsis data, yielding candidate thresholds at RELSA = 0.337 and RELSA = 0.647, which could be used to define severity zones for classifying the animals' burden, although these should not be confused with regulatory severity gradings. This proof-of-concept study showed the potential of the foRcast tool to accurately predict RELSA scores across various animal models and interventions, demonstrating the possibility to identify individual animals at risk of reaching the humane endpoint. The data also emphasize the intrinsic limitation of ARIMA models: drastic changes in the course cannot be reliably predicted. This underscores the need for higher-resolution measurements to detect early changes in animal well-being. Overall, the definition of candidate severity thresholds on the RELSA scale and the foRcast model represent valuable additions to the range of severity tools for refining severity assessment in animal-based research.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.