Key Farm-to-Fork Factors InfluencingLevels in Pastured Poultry Production.

Abstract

Pastured poultry farms offer a unique model for investigating microbial ecology in less controlled environments, presenting challenges and opportunities for food safety management. This study aims to identify the key factors that influencelevels with two complementary modeling approaches: a linear mixed-effect model (LMM) and a random forest (RF) model. Data were collected from 11 pastured poultry farms in the southeastern United States from 2014 to 2017. Five sample types were analyzed: soil (= 812), feces (= 817), ceca (= 206), postprocessing whole carcass rinse (WCR-P;= 235), and final product whole carcass rinse (WCR-F;= 230). Two different sets of predictor variables were used separately: 1) 32 farming practices and 26 physicochemical properties and 2) 80 meteorological factors. The model performance was compared with the randomized mean squared error (RMSE) with a test dataset. LMM was not used for meteorological factors because of the multicollinearity. Significant differences (α = 0.05) inlevels were observed between all sample types, with feces samples showing the highest level. Compared to LMMs, RF models generally showed higher predictive accuracy (lower RMSE) on the test dataset. For soil samples, higher pH and sodium levels were linked to higherlevels. The same trend with pH was observed in fecal samples. WCR-P samples showed that the organic acid treatment in the rinse water led to lowerlevels than other treatments. In WCR-F samples, longer storage time led to lowerlevels. Meteorological factors showed a weaker relationship withlevels compared to farming practices and physicochemical properties, but in soil samples, mild and stable temperature played an important role insurvival. This study can help stakeholders develop data-driven management strategies targeting key factors to aid in the reduction of food safety and animal health risk.