Near infrared spectroscopy analysis as a screening tool to classify milk from bovine subclinical mastitis and promote pathogen-based therapy

Milk is one of the most important foods in the human diet. Its composition and quality may be compromised by bovine mastitis, which is currently the most serious disease affecting dairy cows due to its economic and productive impact. Subclinical bovine mastitis (SBM) is particularly problematic due to no obvious visible symptoms, which hamper the current diagnosis and lead to an underestimation of its impact. Since mastitis is the primary reason for antibiotic use in dairy production, there is a growing focus on a selective therapy, based on the management of the disease by treating only those cases that will benefit from antibiotics. However, the discrimination of these cases remains a major challenge. The objective of this study was to evaluate near infrared spectroscopy (NIRS) analysis of milk samples as a screening method to discriminate SBM aetiology and promote pathogen-based therapy using predictive models. Individual milk samples from 101 Holstein-Friesian cows across 29 herds were collected, subjected to somatic cell count, microbiological culture, and biochemical tests to identify the bacterial species involved. Subsequently, these samples were analysed by NIRS, and the resulting spectral information was used to design different predictive models. Notable differences were noticed among the spectra of the groups compared. The model classification or prediction success ranged from 85.71 % to 95.24 %, with high sensitivity (88.89–100 %) and specificity (81.82–90.91 %). These preliminary data highlight the usefulness of spectral information obtained from NIRS analysis of milk. The proposed approach holds potential as a screening methodology to provide a fast diagnostic result of SBM and promote a pathogen-based therapy to prevent milk quality losses and public health issues. Increasing sample variability in future studies is considered of paramount importance to achieve reliable performance of predictive models.