Predictive models for early lactation diseases in transition dairy cattle at dry-off

• Published in: Preventive veterinary medicine 2019-02, Vol.163, p.68-78
• Main Authors: Wisnieski, L., Norby, B., Pierce, S.J., Becker, T., Gandy, J.C., Sordillo, L.M.
• Format: Article
• Language: English
• Subjects: Animals; Biomarkers; Biomarkers - blood; Cattle; Cattle Diseases - blood; Cattle Diseases - diagnosis; Cohort Studies; Dairy; Dairying; Disease; Female; Lactation; Lactation - blood; Modeling; Models, Biological; Prediction; Predictive Value of Tests; Prospective Studies; Risk Assessment; Stress, Physiological; Time Factors
• ISSN: 0167-5877; 1873-1716
• DOI: 10.1016/j.prevetmed.2018.12.014

During the transition period, dairy cattle undergo tremendous metabolic and physiological changes to prepare for milk synthesis and secretion. Failure to sufficiently regulate these changes may lead to metabolic stress, which increases risk of transition diseases. Metabolic stress is defined as a physiological state consisting of 3 components: aberrant nutrient metabolism, oxidative stress, and inflammation. Current monitoring methods to detect cows experiencing metabolic stress involve measuring biomarkers for nutrient metabolism. However, these biomarkers, including non-esterified fatty acids, beta-hydroxybutyrate, and calcium are typically measured a few weeks before to a few days after calving. This is a retroactive approach, because there is little time to integrate interventions that remediate metabolic stress in the current cohort. Our objective was to determine if biomarkers of metabolic stress measured at dry-off are predictive of transition disease risk. We designed a prospective cohort study carried out on 5 Michigan dairy farms (N = 277 cows). We followed cows from dry-off to 30 days post-calving. Diseases and adverse outcomes were grouped in an aggregate outcome that included mastitis, metritis, retained placenta, ketosis, lameness, pneumonia, milk fever, displaced abomasum, abortion, and death of the calf or the cow. We used best subsets selection to select candidate models for four different sets of models: one set for each component of metabolic stress (nutrient metabolism, oxidative stress, and inflammation), and a combined model that included all 3 components. We used model averaging to obtain averaged predicted probabilities across each model set. We hypothesized that the averaged predictions from the combined model set with all 3 components of metabolic stress would be more effective at predicting disease than each individual component model set. The area under the curve estimated using receiver operator characteristic curves for the combined model set (0.93; 95% confidence interval [CI] = 0.90–0.96) was significantly higher compared with averaged predictions from the inflammation (0.87; 95% CI = 0.83–0.91), oxidative stress (0.78; 95% CI = 0.72–0.84), and nutrient metabolism (0.73; 95% CI = 0.67–0.79) model sets (p