66 A novel analytical method for identifying periods of increased mortality in individual wean-to-finish pig barns

Abstract Mortality rates in all phases of pork production have increased over the past several years. Decreasing mortality rate is of utmost economic importance to producers, especially in weaning-to-finishing pigs where additional investment in feed, medications, and other inputs are expended. A novel analytical method was used to identify sequences of days of increased mortality, defined as mortality episodes, within cohorts of weaning-to-finishing pigs. Cubic smoothing spline regression equations were estimated between days post placement (DPP) and mortality (total daily number of dead and euthanized pigs) for 42 groups of pigs placed in 12 rooms within 6 barns (2 rooms per barn) in Illinois and Iowa from December 2020 to July 2022 (n = 5,676 observation days). For each group, the instantaneous slope between DPP and mortality from spline regression was calculated and used to identify changepoints i.e., peaks (negative instantaneous slope at DPPt and positive instantaneous slope at DPPt-1) and valleys (positive instantaneous slope at DPPt and negative instantaneous slope at DPPt-1). Within each mortality sequence, the maximum and minimum instantaneous slope were considered the start and end of the mortality episode, respectively. Mortality episodes were only considered high mortality episodes if the following metrics were satisfied: 1) greater than or equal to 1 mortality per day, or 2) greater than or equal to 1/3 of the days in the mortality sequence had at least 1 mortality. Days were classified as “start days” if they were within 1 d from the maximum instantaneous slope; likewise, we classified a day as “end days” if it was 1 d from the minimum instantaneous slope. Any day in between the start and end of a mortality episode were defined as “peak days”, and a “normal day” was all other days that did not fall within the start, peak, or end of a high mortality episode. Least-squares means for predicted mortality from a negative binomial generalized linear model showed that mortality was highest and lowest for peak days and normal days, respectively (3.475 ± 0.1729 and 0.591 ± 0.0156 pigs, respectively; P 0.05; Table 1). Mortality during start or end days of an episode did not differ from each other (2.185 ± 0.1973 and 2.069 ± 0.1881 pigs, respectively; P 0.05; Table 1) but were significantly greater and lower than a normal or peak day, respectively (P 0.05; Table 1). The proposed analytical method provides more precise classification of mortality episodes in pig cohorts than previously proposed methods. Results from this method can be used in models to predict the start of mortality episodes, providing ample time for intervention to reduce the overall number of dead pigs in wean-to-finish pig production.