T3-08 Logistic Regression Models: A Resource-focusing Tool to Identify Retail Delis with Increased Likelihood of High Prevalence Listeria monocytogenes Contamination

Monday, August 4, 2014: 3:45 PM
Room 111-112 (Indiana Convention Center)
Susan Hammons, Purdue University, West Lafayette, IN
Brittany DiPietro, Purdue University, West Lafayette, IN
Jingjin Wang, Purdue University, West Lafayette, IN
Haley Oliver, Purdue University, West Lafayette, IN
Introduction: Mitigation strategies to eliminate or reduce Listeria monocytogenes (LM) contamination in retail delis often require a large investment (e.g., labor, resources). Cost-effective screening tools are needed to identify and focus intervention resources in delis with the most contamination which therefore pose the greatest public health risk.

Purpose: This study developed and tested the efficacy of a model to identify retail delis with high levels of environmental LM contamination from a limited number of samples.

Methods: Forward-stepwise selection constructed a Firth’s bias-corrected logistic regression model using sites from a longitudinal retail deli study (n = 30 stores, 6 months, 4508 samples). Since not all delis have floor drains yet, floor drain and drain-associated sites (e.g., floor adjacent drain) were the largest predictors (OR = 243; CI95= 32, >999) in the model, a second model was constructed excluding drain-associated sites. Models were externally validated using data from the second 6-month sampling period (n = 30 stores, 4495 samples); then used to screen 50 additional delis for evidence of high LM prevalence. 

Results: External validation of model A–floor drain included—successfully predicted high prevalence LM contamination in 166/180 events (α = 0.056, β = 0.022).  External validation of model B–drain-associated sites excluded—successfully predicted high prevalence LM contamination in 165/180 events (α = 0.050, β = 0.033). We tested the models in 50 delis among 6 states.  Collectively the models identified 13 of 50 delis with previously unknown LM contamination history to have increased risk for high LM prevalence; 7 stores were confirmed highly prevalent with evidence of persistence through LM environmental testing (≥2 of 20 food and non-food contact surfaces LM positive ≥2 months).

Significance: The developed screening models are conservative and potentially useful for retailer resource focusing efforts, but not recommend for use in a regulatory capacity due to the considerable false-positive identification rate.