A generalized additive model using kinematic and demographic data provides an accurate and objective anesthesia recovery score in horses.

Overview

• This study developed and validated a generalized additive model (GAM) that uses kinematic data from inertial measuring units (IMUs) along with demographic data to objectively assess the quality of anesthesia recovery in horses.
• The GAM outperformed other regression models in predicting recovery scores, providing an accurate and objective measure based on quantifiable physical movements and horse-specific factors.

Introduction and Objectives

• Assessing recovery quality in horses after general anesthesia is important for ensuring safety and guiding clinical decisions.
• Traditional scoring methods are subjective, relying on observers assessing behavior and movement during recovery.
• The study aimed to create an objective recovery score using kinematic data extracted from an IMU, which records acceleration, angular velocity, and body orientation.
• By integrating kinematic features from recorded movement attempts to stand with demographic variables, the study sought to quantitatively model recovery quality.

Methodology

• Study population: 102 mixed-breed horses undergoing general anesthesia between February 2024 and March 2026 were included.
• Data collection: IMUs attached to horses collected:
  • 3-axis acceleration
  • Angular velocity
  • Orientation data
• Recovery scoring: Recovery quality was rated by three separate observers using a modified composite grading scale.
• Dependent variable: The mean recovery score from the three raters was used as the outcome to predict.
• Data processing pipeline: Developed in Python to:
  • Detect attempts to stand based on the first derivative of acceleration (identifying changes in movement)
  • Extract various kinematic features from these attempts (e.g., impact intensity, timing)
• Independent variables: Extracted kinematic features plus demographic data like duration of anesthesia.
• Model comparison: Evaluated four regression frameworks:
  • Multiple linear regression
  • Polynomial regression
  • Generalized additive model (GAM)
  • Random forest regression
• Model selection criteria:
  • Akaike information criterion (AIC)
  • Bayesian information criterion (BIC)
  • Root mean squared error (RMSE)
  • Coefficient of determination (R2)

Results

• The generalized additive model was the best performing model with:
  • Lowest AIC: 658.24
  • Lowest BIC: 737.96
  • Lowest RMSE: 5.21
  • Highest R2: 0.8 (indicating strong explanatory power)
• Key independent variables identified by the GAM included:
  • Cumulative fall intensity during attempts to stand
  • Time to stand after anesthesia
  • Duration of anesthesia (with a linear effect)
  • Mean impact severity showing a nonlinear relationship with recovery quality

Discussion and Conclusion

• The study demonstrates that recovery quality in horses post-anesthesia can be objectively quantified through combined kinematic and demographic data.
• The GAM’s flexibility in modeling both linear and nonlinear relationships provided superior fit and interpretability compared to linear, polynomial, or random forest models.
• Objective recovery scoring could:
  • Improve clinical assessment accuracy
  • Reduce observer bias
  • Enable automated monitoring and early detection of recovery complications
• The approach relies on wearable sensor technology, making it practical for real-world veterinary applications.
• Future work could involve real-time implementation and validation in diverse clinical scenarios.

Summary

• This research successfully developed a data-driven model to predict anesthesia recovery quality in horses by analyzing IMU-recorded movement patterns and horse-specific factors.
• The generalized additive model offers an objective, accurate, and interpretable framework that advances veterinary anesthesia monitoring and potentially improves animal welfare post-operation.