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

Area of Science:

• Animal research ethics
• Biomedical engineering
• Statistical modeling

Background:

• Robust, objective, and model-specific parameters are crucial for mandatory severity assessment in animal research.
• The RELative Severity Assessment (RELSA) procedure offers multivariate analysis for severity grading.
• Existing methods require enhancement for proactive identification of animal welfare risks.

Purpose of the Study:

• To integrate an Autoregressive Integrated Moving Average (ARIMA)-based prediction model, termed foRcast, with the RELSA procedure.
• To utilize foRcast for predicting RELSA scores at humane endpoints across diverse animal models.
• To establish exploratory severity thresholds using kernel density estimates (KDE) on the RELSA scale.

Main Methods:

• Reanalyzed data from mice reaching humane endpoints using transmitter-derived (e.g., heart rate) and behavioral (e.g., wheel running) parameters.
• Applied the foRcast model (ARIMA-based) to predict RELSA scores.
• Employed kernel density estimate (KDE) on RELSA scores, particularly for sepsis models, to define potential severity thresholds.

Main Results:

• The foRcast tool demonstrated promising performance with a root mean square error of 0.069.
• 96% of actual RELSA scores fell within the 95% prediction interval of the foRcast model.
• KDE analysis identified candidate thresholds (RELSA = 0.337 and 0.647) for defining severity zones in sepsis models.

Conclusions:

• The foRcast tool shows potential for accurately predicting RELSA scores and identifying individual animals at risk of reaching humane endpoints.
• The study highlights the need for higher-resolution measurements to capture rapid changes in animal well-being, a limitation of current ARIMA models.
• The foRcast model and candidate severity thresholds represent valuable advancements for refining severity assessment in animal-based research.