Evaluating the accuracy of models using routinely collected herd data for prediction of on-farm lameness prevalence

Lameness is a major welfare and production concern for the dairy sector worldwide. Quantifying the lameness challenge is essential for herd health management. The Claw Health Indicator (CHI), developed by a Norwegian dairy company, is calculated on a monthly basis using data routinely collected in the Norwegian Dairy Herd Recording System. The CHI was hypothesized to reflect the lameness prevalence on farm. Our cross-sectional study evaluated the accuracy of predictions made about the lameness prevalence on farm by the CHI. We also developed an alternative model for predicting the lameness prevalence using additional variables of routinely collected herd data (RHD) and evaluated its accuracy. We used data from 149 Norwegian freestall dairy herds. A univariable β regression model was built with the total proportion of lame and severely lame cows as the dependent variable and the CHI score for each herd as the independent variable. A second model, a multivariable β regression model with the same dependent variable, was built using selected variables of RHD. The accuracy of both models was assessed in terms of their concordance r and prediction errors. A higher CHI score was indicative, as hypothesized, of a lower lameness prevalence and vice versa. Moving from the lowest CHI score in our sample to the highest was equivalent to a reduction of ∼6% in the prevalence of lameness. There was, however, substantial variation in the total proportion of lame cows seen in herds with the same CHI score. The CHI model was highly inaccurate when predicting the lameness prevalence on farm, with a concordance r of 0.05. The alternative RHD model, which included 6 independent variables of RHD, was more accurate than the CHI model but still inaccurate with a concordance correlation coefficient of 0.31. Both the CHI and the alternative RHD models were unsuitable for accurately predicting the prevalence of lameness on farm. Whereas their constituent parts were reflective of claw health at the herd level, they failed to reflect the complex, multifactorial nature of lameness in dairy herds. The CHI may still be useful for claw health management, as a tool for identifying potential issues on farm, and motivating producer engagement but further investigation is required. The RHD variables in the alternative model are simpler and more widely available than the CHI, facilitating replication and further development in other countries or groups of dairy herds outside of Norway.