Measuring validity and reliability of facial expressions of pain in horses as an innovative welfare indicator

This study aimed to develop and validate a standardized pain scale based on facial expressions of horses (Horse Grimace Pain Scale, HGPS). Forty stallions of different breeds aged 1÷5 years were assigned to one of two groups. A (n= 19) underwent routine surgery castration in general anaesthesia and received one injection of Flunixin-Meglumine immediately before anaesthesia. B (n= 21) underwent the same surgical procedure but received Flunixin-Meglumine before anesthesia and six hours after the surgery. Six horses with no pain but needing noninvasive, indolent diagnostic procedures received the same treatment than A but not surgery and were used as controls (C). The horses were hospitalized at the Havelland Clinic (see abstract “Pain assessment in horses after routine castration”) for 5 days. Two digital video-cameras were positioned on the top opposite sides of the box. Thirty-minutes videos were recorded on the day prior to surgery (baseline) and at 8 hours following castration. From each video images of the horses’ faces were extracted, enabling generation of a number of clear and high quality pictures. Images were compared to identify changes in facial expressions associated with castration pain by a treatment-blind observer experienced in assessing facial expressions in other species. Six facial action units were defined. Baseline and 8-hours post procedure images of each horse were selected at random. 126 images were then scored in a randomized sequence using the action units by five treatment-blind participants. In order to explore the effect of time and treatment, the horse grimace pain scale (HGPS) for each image was determined by summarizing the score provided for each action unit. 73,3% images were correctly assessed as true post-surgery or true basal. Inter-class correlation coefficient among observers was 0,91 showing high inter-observer reliability. GLM for repeated measures showed a significant effect of time and treatment (P<0,001), A and B significantly differed from C (P<0,01). A promising welfare indicator based on a standardized scale of facial expressions showed validity and reliability. It has the advantages of being not time consuming, easily trainable and applicable on farm.