Benchmarking dairy herd health status using routinely recorded herd summary data

Area of Science:

• Animal Science
• Data Science
• Veterinary Medicine

Background:

• Genetic improvement for dairy cattle health using producer data is feasible but slow due to low heritability.
• Environmental and managerial factors significantly influence lowly heritable health traits.
• Existing herd data offers potential for advanced health status benchmarking.

Purpose of the Study:

• To assess the feasibility of using routinely collected herd data to benchmark dairy cattle health status.
• To compare the predictive accuracy of different modeling techniques for health events.
• To identify key factors influencing dairy cattle health.

Main Methods:

• Combined over 1,100 herd characteristics with producer-recorded health event data.
• Utilized parametric and nonparametric models, including logistic regression, support vector machines, and random forests.
• Analyzed health events categorized into mastitis, reproductive, and metabolic, using herd and individual incidence as dependent variables.

Main Results:

• Random forest models demonstrated the highest accuracy in predicting herd and individual cow health status across all categories.
• Prediction accuracy at the herd level ranged from 0.61 to 0.63; at the individual level, it ranged from 0.87 to 0.93.
• Key predictive factors identified across models included herd size, movement out of the herd, and weather variables.

Conclusions:

• Benchmarking dairy cattle health status using routinely collected herd data is feasible and effective.
• Nonparametric models, particularly random forests, are well-suited for analyzing complex, high-dimensional herd data.
• Data mining techniques can accurately predict health status, supporting evidence-based herd management decisions.