Trough or bowl? Observers need training for assessing resource as well as clinical parameters

Summary in the original language of the document

While the need for training on-farm assessors in clinical animal assessment has been widely recognised, assessment of husbandry resources is still often regarded as self-explanatory.
Within the scope of the international project ProPIG, 7 observers from seven countries were trained by an experienced observer (gold standard) to assess 15 clinical and 11 resource parameters in organic pigs in eight countries. The initial plan was to train and test observers before farm visit 1 and then again after one year before farm visit 2. Both trainings were repeated with all observers due to unsatisfactory agreement, resulting in T1a+b and one year later T2a+b. Agreement with the gold standard was calculated as exact agreement for categorical parameters (e.g. drinker type; mean n=11 pens per test and parameter, range 1 - 34) and Spearman rank correlation for numerical parameters (e.g. number of animals; mean n=9 pens, range 4 - 28).
Median (IQR) pairwise agreements [%] were T1a=83 (40), T1b=90 (29), T2a=92 (43), T2b=100 (11) for clinical parameters, and T1a=100 (25), T1b=100 (40), T2a=100 (23), T2b=90 (33) for resource parameters. Mean Spearman r for clinical parameters were T1a=0.52, T1b=0.76, T2a=0.42 and T2b=0.84 with ranges of -0.69, -0.33, -0.79 and 0.34, respectively, to 1.00. Mean Spearman r for resource parameters were T1a=0.59 (range 0 to 1), T1b=0.71 (-1 to 1), T2a=0.40 (0.30 to 0.49) and T2b=0.25 (-1 to 1). Initial training discussions showed that naïve observers differed in their assessment of resource as well as clinical parameters, and real life assessment together with training materials were needed to successfully train on both sets of parameters.
We therefore recommend the inclusion of resource parameters in observer trainings for on-farm assessment in order to assure sufficient observer agreement.