Use of differential somatic cell count, somatic cell score, and milk mid-infrared spectral analysis for monitoring mastitis in dairy cows during routine milk recording

Abstract

Routine mastitis screening is crucial on dairy farms, and it usually involves analysis of somatic cell count (SCC), which serves as an indicator of mastitis. Milk mid-infrared (MIR) spectral analysis, the standard method to determine milk components, was evaluated as an additional tool for the prediction of mastitis, since health condition of a cow is affecting milk composition. Differential somatic cell count (DSCC), the percentage of SCC due to polymorphonuclear leukocytes (PMN) and lymphocytes, is another new indicator to detect cows having an infected udder mammary gland. In a previous study we developed mastitis prediction models based on somatic cell score (SCS) and MIR spectral data from routine milk recording. The present study compared various models to predict clinical mastitis based on the parameters DSCC, SCS and MIR spectral data, alone or in combination. Data on DSCC, SCC, MIR spectral data and mastitis diagnosis by veterinarians were extracted for 40,332 dairy cows of the breeds Fleckvieh, Brown Swiss and Holstein Friesian from the Austrian milk recording and health monitoring systems. Three data subsets were created based on whether test-day records were classified as “mastitis” when they occurred 21 days (subset 1), 14 days (subset 2), or 7 days (subset 3) before or after diagnosis of mastitis; all other records were classified as “healthy”. The performance of seven models (DSCC alone, SCS alone, MIR alone, DSCC+SCS, DSCC+MIR, MIR+SCS and DSCC+SCS+MIR) was assessed in these subsets using partial least squares discriminant analysis. In all subsets, the DSCC model showed the highest sensitivities (0.710 to 0.792) but lowest specificities (0.564 to 0.562) during external validation. The SCS model showed lower sensitivities (0.624 to 0.662) but higher specificities (0.725 to 0.769) than DSCC. The MIR model showed sensitivities from 0.598 to 0.617 and specificities from 0.668 to 0.681. The models combining multiple predictors showed the highest balanced accuracies: the MIR+SCS model in subset 1 (0.697), DSCC+SCS+MIR in subset 2 (0.712) and MIR+SCS in subset 3 (0.744). Taken together, our analyses suggest that DSCC, SCS and MIR spectral data are useful for routine mastitis screening, with SCS performing as the best single predictor and the MIR+SCS model performing best overall.