A company that provides products and services to the animal health industry was looking to help meat processors reduce the risk of E.coli contaminations. Among the stated goals of their initiative were:
- Provide a unique forum to add value & enhance food safety systems through informatics;
- Identify potential predictive metrics of high pathogen levels.
The company called on First Analytics, with whom a retainer relationships exists to extend their analytics capabilities, to develop this service.
- We took inventory of available data and built a database of several dozen potential predictors of high pathogen levels. These predictors fell into the areas of (1) suppliers of the beef (producers, lot characteristics), (2) operations (downtime, employees, chain speed, carcass characteristics), and environment (weather, time of year).
- In a traditional test of hypotheses sense, the team, in conjunction with subject matter experts, developed a set of hypothesis, to be supported or rejected by the data. Other tools, such as tree modeling and association rules analysis were used to provide ancillary insights.
The models supported some of the hypotheses, while rejecting others. For example, these were among some of the factors supported by the data, and their impact on risk of E.coli positives quantified:
- Unscheduled down time allows micro-organisms to grow;
- The impact of carcass age, in hours, can be quantified;
- Temperature, humidity, and pressure can often be one of the largest risk factors;
- Employee absenteeism, as a proxy for appropriate staffing levels, contributes to risk.
Some factors, such as chain speed, which were assumed to have positive associations, were not statistically significant.
The model was able to quantify the increase in the chance of positives by an amount ranging from 16% to 30%, depending on the factor.