Predictive modeling toward refinement of behavior-based pain assessment in horses

After 25 years of studies on methodologies for behavioral assessment of equine pain, the Unesp-Botucatu Horse Acute Pain Scale (UHAPS) and the Orthopedic Composite Pain Scale (CPS) were recently considered suboptimal instruments to assess pain in hospitalized horses. However, the combination of the two instruments has never been examined. The objective was to investigate whether the merging, mining, and weighting of UHAPS and CPS behavioral items in a single instrument using a predictive model could improve the capacity to diagnose pain in horses. A previously video-collected behavioral database of 42 horses admitted to three different hospitals for orthopedic or soft tissue surgery was used. Multilevel binomial logistic regression models were used to merge, mine, and weight the behaviors of both instruments. The classification quality between the model and the instruments was compared by the area under the curve (AUC) and its 95% confidence interval. The short model containing 25% of the behaviors of the two instruments showed a higher AUC (98.64 [98.16 – 99.12]; p < 0.0001) than the UHAPS (84.63 [82.08 – 87.18]) and CPS (88.62 [86.56 – 90.66]), independently. We conclude that merging, mining, and weighting the UHAPS and CPS behavior items into a single predictive model appears to be a promising strategy to improve pain diagnostic skill and promote equine welfare.