Prediction of metabolic status of dairy cows in early lactation using milk fatty acids and test-day variables

Early lactation metabolic imbalance is an important physiological change affecting the health, production, and reproduction of dairy cows. The aims of this study were (1) to evaluate the potential of test-day (TD) variables with or without milk fatty acids (FA) content to classify metabolically imbalanced cows and (2) to evaluate the robustness of the metabolic classification with external data. A data set was compiled from 3 experiments containing plasma β-hydroxybutyrate, nonesterified FA, glucose, insulin-like growth factor-I, FA proportions in milk fat, and TD variables collected from 244 lactations in wk 2 after calving. Based on the plasma metabolites, 3 metabolic clusters were identified using fuzzy c-means clustering and the probabilistic membership value of each cow to the 3 clusters was determined. Comparing the mean concentration of the plasma metabolites, the clusters were differentiated into metabolically imbalanced, moderately impacted, and balanced. Following this, the 2 metabolic status groups identified were imbalanced cows (n = 42), which were separated from what we refer to as “others” (n = 202) based on the membership value of each cow for the imbalanced cluster using a threshold of 0.5. The following 2 FA data sets were composed: (1) FA (groups) having high prediction accuracy by Fourier-transform infrared spectroscopy and, thus, have practical significance, and (2) FA (groups) formerly identified as associated with metabolic changes in early lactation. Metabolic status prediction models were built using FA alone or combined with TD variables as predictors of metabolic groups. Comparison was made among models and external evaluations were performed using an independent data set of 115 lactations. The area under the receiver operating characteristics curve of the models was between 75 and 91%, indicating their moderate to high accuracy as a diagnostic test for metabolic imbalance. The addition of FA groups to the TD models enhanced the accuracy of the models. Models with FA and TD variables showed high sensitivities (80–88%). Specificities of these models (73–79%) were also moderate and acceptable. The accuracy of the FA models on the external data set was high (area under the receiver operating characteristics curve between 76 and 84). The persistently good performance of models with Fourier-transform infrared spectroscopy-quantifiable FA on the external data set showed their robustness and potential for routine screening of metabolica