Development of Benchmark Curves to Early Detect Health-Related Productivity Deviations Using Production Indicators in Swine Nursery and Finishing Lots.

The nursery and finishing phases are critical for profitability and sustainability in swine production, but effective methods for early disease detection in these stages remain underdeveloped. This study used routinely collected production indicators from nursery (42-56 days postfarrowing) and finishing lots (115-120 days postfarrowing) to create production benchmark curves for anomaly detection. These curves were developed for farms without diagnosed health challenges and compared to those with diagnosed health issues, based on tissue submissions and diagnostic codes (Dx codes). Statistical methods such as resampling techniques, Bayesian statistics, and standard deviation (SD) thresholds were employed to build the benchmark curves. The main objective was to test and compare the benchmarks using detection, early detection rate (EDR), time-to-detect (TTD), and false positive rate (FPR). Results showed that bootstrapping (BOOT), jackknife (JK), and Markov chain Monte Carlo (MCMC) methods provided the highest EDRs, although they were prone to false positives. For nursery lots, it was observed that using cumulative average with one SD for feed disappearance (EDR 49.2% and FPR 9.8%) and estimated weight (EDR 47.2% and FPR 8.8%) showed the best balance between EDR and FPR, and using MCMC for mortality showed the best balance between EDR and FPR (EDR 38.8% and FPR 13.8%). For finishing lots, using cumulative average with one SD showed a more balanced performance with FPR below 14.0% and EDR of 21.3% for feed disappearance, 67.4% weight, and 59.7% for mortality. These findings demonstrate the potential of using production indicators for early health challenge detection in swine operations.