Predictive models to determine best strategy for metaphylaxis application in cattle at arrival to a feedyard

Metaphylaxis, the application of an antimicrobial to a cohort of cattle at arrival to a feedyard, is an important bovine respiratory disease (BRD) control strategy for certain cattle populations. Predictive modeling techniques could be used to assist in determining which cohorts should receive metaphylaxis based on a desired economic outcome instead of subjectively. The study objective was to evaluate predictive models trained with cattle demographic variables to determine which cohorts should receive metaphylaxis based on an economic evaluation of highest net returns and to elucidate the benefit to model performance with the addition of origin and external economic variables. Data from 16,368 cattle cohorts were used to build four predictive models: boosted decision tree, logistic regression, neural network, and random forest. Area under the Receiver Operating Characteristics curve (AUC-ROC) was used to evaluate model performance. The same algorithms were used to compare adding origin and external economic data to the baseline models. Overall, model performance was high with AUC-ROC values ranging from 0.80 to 0.93 in the baseline models. Adding external economic variables such as commodity futures prices increased performance (AUC-ROC=0.92–0.94). Adding origin data, such as city and state, resulted in poorer performance (AUC-ROC=0.79–0.89). The combination of external economic data and origin resulted in intermediate AUC-ROC values (AUC-ROC=0.87–0.91). The study demonstrated that predictive models can be used successfully to select an optimal metaphylaxis strategy as determined by economic evaluation for cattle arriving at the feedyard.