Quantifying the causal impact of diseases on lactation features in Holstein cattle based on causal inference...

Disease represents a key constraint to farm sustainability by greatly compromising productivity and animal welfare in the dairy industry. Previous research has focused on short-term milk losses during the disease periods. However, in the longer term, cows often fail to fully recover milk yield even after clinical cure. Therefore, the main objective of this study was to quantify the short-term (during the disease period) and long-term (after clinical cure) milk losses and elucidate the causal relationships between diseases and lactation features using large-scale on-farm records and causal inference approaches, thereby enhancing the understanding of disease challenges and helping to reduce their burden. We collected high-throughput session milk yields and disease records (udder health; reproductive, metabolic, and digestive disorders; and hoof health) of 37,246 Holstein cattle from 2020 to 2024. Two causal inference approaches were applied to assess causal effects on milk yield and the variability of session milk yields, including propensity score matching and overlap weighting. Overall, cows of later parities, lower overall resilience, difficult calving, greater number of artificial inseminations, and stillbirth faced higher disease risks, with hazard ratios ranging from 1.05 to 2.40. Genetic analyses revealed that higher milk yield and greater variability in session milk yields were positively genetically correlated with increased disease prevalence. Causal inference revealed that clinical diseases significantly reduced 305-d milk yield by 680.81 ± 0.01 kg per lactation. This causal impact intensified with disease frequency within a lactation, escalating from 646.18 ± 0.01 kg for a single disease onset to 920.83 ± 0.04 kg for multiple onsets, suggesting a nonlinear cumulative burden. For each type of disease, the estimated causal effects on 305-d milk yield ranged from 345.76 ± 0.81 kg for metabolic disorders to 511.72 ± 0.04 kg for reproductive disorders. On average, 7.20% of milk yield was completely unrecoverable after disease cure, ranging from 4.06% for metabolic disorders to 8.42% for digestive disorders. Notably, phenotypic trends, causal estimates, and genetic correlations consistently identified increased variability in session milk yields as a concomitant feature of disease onset. The coefficient of variation increased sharply approximately 5 d before the appearance of clinical signs, rising by 30% for metabolic disorders and up to 58% for digestive disorders, highlighting its potential as an early indicator of diseases in dairy cattle. In summary, this study provides new insights into the causal relationships between diseases and lactation features. Furthermore, it demonstrates the potential of causal inference approaches to advance precision livestock farming.