Skip to main content
SLU publication database (SLUpub)

Research article2021Peer reviewedOpen access

Diagnostic properties of milk diversion and farmer-reported mastitis to indicate clinical mastitis status in dairy cows using Bayesian latent class analysis

Bonestroo, John; Fall, Nils; van der Voort, Mariska; Klaas, Ilka Christine; Hogeveen, Henk; Emanuelson, Ulf

Abstract

The development of digital farming gives bovine mastitis research and management tools access to large datasets. However, the quality of registered data on clinical mastitis cases or treatments may be inadequate (e.g. due to missing records). In automatic milking systems, the decision to divert milk from the bulk milk tank during milking is registered (i.e. milk diversion indicator) for every milking and could potentially indicate a clinical mastitis case. This study accordingly estimated the diagnostic performance of a milk diversion indicator in relation to farmer-recorded clinical mastitis cases in the absence of a "gold standard". Data on milk diversion and farmer-reported clinical mastitis from 3,443 lactations in 13 herds were analyzed. Each cow lactation was split into 30-DIM periods in which it was registered whether milk was diverted and whether clinical mastitis was reported. One 30-DIM period was randomly sampled for each lactation and this was the unit of analysis, this procedure was repeated 300 times, resulting in 300 datasets to create autocorrelation-robust results during analysis. We used Bayesian latent class analysis to assess the diagnostic properties of milk diversion and farmerreported clinical status. We analyzed different episode lengths of milk diversion of 1 or more milk diversion days until 10 or more milk diversion days for two scenarios: farmers with poor-quality (51% sensitivity, 99% specificity) and high-quality (90% sensitivity, 99% specificity) mastitis registrations. The analysis was done for all 300 datasets. The results showed that for the scenario where the quality of clinical mastitis reporting was high, the sensitivity was similar for milk-diversion threshold durations of 1-4 days (0.843 to 0.793 versus 0.893). Specificity increased when the number of days of milk diversion increased and was >= 98% at a milk-diversion threshold durations of 8 or more consecutive milk diversion days. In the scenario where the quality of clinical mastitis reporting was low, the sensitivity of milk diversion and reported clinical mastitis cases was similar at milk-diversion threshold durations of 1-7 days (0.687 to 0.448 versus 0.503 to 0.504) while specificity exceeded the 98% at milk-diversion threshold durations of 7 or more consecutive milk diversion days. In both scenarios, a milk diversion threshold duration of 4-7 days achieved the most desirable combined sensitivity and specificity. This study concluded that milk diversion can be a valid alternative to farmer-reported clinical mastitis as it performs similarly in indicating actual clinical mastitis.

Keywords

antibiotic treatment; proxy:automatic milking system; milk withdrawal; latent class analysis

Published in

Livestock Science
2021, Volume: 253, article number: 104698