Effects of assessment method (real-time versus video-recorded) on a validated pain-altered behavior scale used in castrated piglets

Abstract

We aimed to compare two assessment methodologies (real-time vs. video-recorded) using the Unesp-Botucatu Pig Composite Acute Pain Scale (UPAPS) in piglets before and after castration. Twenty-nine male piglets were castrated. Four observers scored the UPAPS over three perioperative timepoints of castration following two assessment methodologies. In real-time assessments, the observers were in-person observing the piglets in front of the pen. After two weeks, the observers did video-recorded assessments randomizing piglets and timepoints. Modeling was conducted to compare the UPAPS and each pain-altered behavior between methodologies. Intraclass correlation coefficient (ICC), Bland–Altman, and Lin’s concordance correlation coefficient (CCC) were conducted to investigate agreement between methodologies. UPAPS was statistically equivalent between methodologies (P = 0.4371). The ICC for each method was very good (0.85 to 0.91). The agreement of the UPAPS assessed between methodologies had minimal bias (− 0.04), no proportion bias, and 53% of the assessments presented a perfect agreement. However, CCC of the UPAPS was moderate (0.65), and only one pain-altered behavior (“presents difficulty in overcoming obstacles or other animals”) occurred more in real-time assessments (P = 0.0444). In conclusion, piglet pain assessment by UPAPS can be conducted in real-time based on a suitable agreement between the real-time and video-recorded assessment methods.

Real-time and video-recorded pain assessment in beef cattle: clinical application and reliability in young, adult bulls undergoing surgical castration

“…In addition, significant differences were found in M1, M3, and M4 when comparing the UCAPS total score between both methods. These results were different from a recent study in piglets undergoing castration, where real-time and video-recorded behavioral assessment methodologies did not significantly affect total pain scores over time 37 . The real-time assessment presented overall lower total pain scores than the video-recorded method, and at M1, a time-point before the surgical procedure, considered non-painful.…”

“…Therefore, this study aimed to investigate the reliability and the agreement of real-time and video-recorded pain assessment in beef cattle using the UCAPS on young, adult bulls undergoing surgical castration. Similar studies have been conducted in pigs, rats, and mice 35 – 37 . Our hypothesis proposes that there is no difference in reliability, and there is concordance in agreement between real-time and video-recorded assessment.…”

“…Similarly, to studies in cats, rats, and mice 35 , 36 , 40 , the Bland–Altman for repeated measures methods demonstrated a narrow limit of agreement and a minor bias between both methods of pain assessment, reinforcing the sufficient agreement between assessment methods. Additionally, a recent study in pigs, showed no significant impact on total pain scores over time in relation to castration whether using real-time or video-recorded method 37 . The low perfect agreement between methods probably resulted from the fact that in real-time assessment the evaluators were aware of the time-points, and for the video recorded they were masked.…”

“…S1 ) and Cameron and Trivedi’s test (overdisp::overdisp) (Lambda t-test score = 4.687 and p < 0.00001) proved an overdispersion (excess of zeros), requiring a zero-inflated model 49 . Zero-inflated models combine Poisson and Bernoulli distributions in the fixed effects of the same model for a better fit of the data 37 . Therefore, a multilevel zero-inflated poisson model (glmmTMB::glmmTMB) was identified as the best fit compared with other types of models (Linear, Poisson and Negative Binomial models) according to the histogram (Fig.…”

“…S1 ) and the Bayesian information criterion (stats::BIC), as proposed previously 34 . The UCAPS total was used as the response variable, while the fixed effects were composed of two components (Poisson and Bernoulli distributions), which is a special characteristic of the zero-inflated models 37 . Therefore, evaluators, breeds, and interaction between time-points and assessment methods were used as explanatory variables in the model fixed effects related to (Poisson distribution).…”