Milk component ratios for monitoring of health during early lactation of Holstein cows

Objective: The objective was to evaluate the potential of milk fat-to-protein ratio (F:P), milk fat-to-lactose ratio (F:L), and milk protein-to-lactose ratio (P:L) as tools to assist with the preliminary screening of cows that need extra attention and are at risk of developing disease. Sub- sequently, the performance of 2 proposed indices, based on changes in the ratio of components, was evaluated.
Materials and Methods: Daily milk component data from 198 Holstein cows were collected until 60 DIM, and the occurrence of health disorders (mastitis [MAS], me- tritis [MET], digestive disorders [DIG], lameness [LAM], and ketosis [KET]) was assessed by veterinarians and farm personnel. Two indices were developed to explore the association between F:P, F:L, and P:L and disease: (1) cow index (CIx), based on changes in F:P, F:L, and P:L in the affected cow during the days previous to the diagnosis of health disorders; and (2) mates index (MIx), based on deviations in F:P, F:L, and P:L relative to previous days and healthy pen mate cohorts. Subsequently, alert models were proposed for each index (ACIx and AMIx) considering reference alert cutoff values based on the differences between index values in the cow itself (ACIx) and among healthy versus sick cows (AMIx) for each specific disease in analysis. In a final step, the proposed alert models were statistically tested by evaluating their sensitivity, specificity, and predictive values.
Results and Discussion: The odds of disease occurrence increased by 1.20 to 1.52, 1.29 to 1.30, and 1.16 to 1.67 for each decimal unit increment in F:P, F:L, and P:L, respectively. Sensitivity (Se) of alerts based on F:P ranged from 58% (LAM) to 80% (MET) with 58% to 62% specificity (Sp), respectively. The Se of alerts based on F:L ranged from 60% (LAM) to 71% (MAS) with 58% to 63% Sp, respectively. The Se of alerts based on P:L ranged from 45% (LAM) to 81% (MET) with 68% to 72% Sp.
Implications and Applications: Although the pro- posed system had overall low to moderate Se and Sp, resulting in poor accuracy, the results suggested that changes in F:P, F:L, and P:L have potential to be used as early indicators of disease. Mastitis, LAM, and KET were better detected using F:L, MET was better detected using P:L, and DIG was better detected using F:P. Overall, ACIx was more effective than AMIx for disease detection. The accuracy, Se, and Sp were below the acceptable range for reported disease conditions (area under the curve 0.59–0.77) and should be considered before considering for on-farm adoption. The overall moderate Se and low accuracy suggest that the systems should be further evaluated using a larger data set with more disease events to be confidently used as a decision-support aid.
© 2023 Elsevier B.V.