Diagnostic tools and innovative approaches for promoting the rational use of antimicrobials in mastitis control

• Ano de defesa: 2025
• Autor(a) principal: Garcia, Breno Luis Nery
• Tipo de documento: Tese
• Tipo de acesso: Acesso aberto
• Idioma: eng
• Instituição de defesa: Biblioteca Digitais de Teses e Dissertações da USP
• Palavras-chave em Português: Antimicrobial resistance; Aprendizado de máquina; Bovine mastites; Cultura na fazenda; Machine learning; Mastite bovina; On-farm culture; Resistencia aos antimicrobianos; Selective dry cow therapy; Terapia seletiva de vaca seca
• Link de acesso: https://www.teses.usp.br/teses/disponiveis/10/10135/tde-24102025-173056/

Antimicrobial resistance (AMR) is a global public health concern, partly driven by antimicrobial use (AMU) in livestock. In dairy cattle, clinical mastitis (CM) treatment and dry cow therapy (DCT) are the main drivers of AMU. This thesis addresses four objectives across four chapters aimed at improving antimicrobial stewardship in mastitis control. Chapter 1 focused on developing and validating a machine learning (ML)-based application (Rumi) for interpreting on-farm culture (OFC) results to guide selective CM treatment. Rumi′s diagnostic accuracy was compared to a microbiology expert using MALDI-TOF MS as the gold standard, and to farm personnel users (FPUs) through Bayesian latent class models. Rumi demonstrated similar sensitivity and specificity to the expert, and no significant difference compared to FPUs. These findings support the use of Rumi as a reliable tool for on-farm CM treatment decisions. Chapter 2 evaluated the use of ML models to predict clinical cure in mild and moderate CM cases. A dataset comprising 319,341 CM cases from 3,770 herds, from a herd management mobile app focussed on mastitis control, was used to train nine ML algorithms. The models achieved only moderate predictive accuracy, ranging from 0.580 to 0.589, and no model significantly outperformed the others. While current predictive capacity is limited, further improvements in model inputs could enhance their utility in supporting AMU decisions. Chapter 3 assessed the diagnostic agreement between a rapid direct antimicrobial susceptibility test (RDAST) and the standard agar disk diffusion method in major mastitis pathogens. Agreement ranged from slight to moderate (Cohen′s Kappa: 0.09 to 0.47), with a mean inhibition zone bias of 2.81 mm. To evaluate the association between in vitro susceptibility and bacteriological cure (BC), 340 mild/moderate CM cases were analyzed using a generalized linear mixed model (GLMM) including cow-level variables. No significant association was found between susceptibility test results and BC. These findings suggest that while in vitro susceptibility testing is valuable for AMR surveillance, its utility for guiding CM treatment decisions remains limited. Chapter 4 evaluated selective dry cow therapy (SDCT) using OFC combined with the California Mastitis Test (CMT) as selection criteria. A total of 1,386 quarters from 355 cows were randomly assigned to: a) blanket DCT (control), b) SDCT at the quarter level (Quarter-level SDCT), or c) SDCT at the cow level (Cow-level SDCT). Compared to the control, AMU was reduced by 53% in Quarter-level SDCT and 30.5% in Cow-level SDCT. A GLMM including cow-level variables was used to assess new intramammary infection (NIMI) risk. Although somatic cell count and days in milk were associated with NIMI risk, no significant differences were found among treatment groups. These findings support the use of OFC+CMT for SDCT at both the quarter and cow levels. In conclusion, the results of this thesis support the implementation of on-farm diagnostic tools and data-driven strategies to optimize AMU in mastitis control and advance antimicrobial stewardship in the dairy industry.