1512 Mokhtari and Frey
Key methodological questions addressed in- clude: How does uncertainty in model inputs af-
fect the ranking of key sources of variability? Does comingling variability and uncertainty into a single
dimension provide different insight regarding key model inputs than two-dimensional simulation? How
can one discriminate the importance between closely ranked inputs? What are the limitations of commonly
used sensitivity analysis methods when applied to MFSPR models? What are the implications of sen-
sitivity analysis for refinement of the analysis and for risk management?
Food safety risk assessment modeling is dis- cussed. Motivations for sensitivity analysis of MFSPR
models are explained. The rationale for focusing on a selected sensitivity analysis method and an MFSPR
model as a test bed is given.
1.1. Food Safety Risk Assessment Modeling
In the field of food safety, quantitative risk assess- ment is widely used to provide insight with respect
to risk management strategies.
4,5
There has been increasing development and use of MFSPR models
for risk assessment of food-borne diseases.
6
Exam- ples include Salmonella in eggs,
7
E. coli O157:H7 in
ground beef,
8
Vibrio parahaemolytics in raw mollus-
can shellfish,
9
and Listeria monocytogenes in ready- to-eat RTE foods.
10
MFSPR models generally have probabilistic frameworks, are able to quantify uncertainties in the
population risk, include many parameters and inputs that have uncertainty andor variability, have complex
or highly nonlinear interactions between inputs, have threshold or saturation points in the model response,
and use different types of inputs e.g., continuous ver- sus categorical. A threshold is a value in an input do-
main below which a model output does not respond to changes in the input. Similarly, a saturation point is an
input value above which the model output does not re- spond. Parameters and inputs of MFSPR models are
generally quantified by using probability distributions that represent variability, uncertainty, or both. Vari-
ability represents true heterogeneity in the population that cannot be further reduced. Uncertainty repre-
sents lack of perfect knowledge and can be reduced by further measurements.
11–20
1.2. Need for Sensitivity Analysis
Sensitivity analysis has long been recognized as a useful adjunct for model building
1,21
Sensitivity analysis highlights the inputs that have the great-
est influence on the results of a model; therefore, it provides useful insights for model builders and
users.
22
Sensitivity analysis provides insight regard- ing which model input contributes the most to un-
certainty, variability, or both, for a particular model output. Knowledge of key sources of uncertainty is
useful in prioritizing additional data collection or research.
Knowledge of key sources of variability is useful in identifying control measures. Sensitivity analysis
is an aid in developing priorities for risk mitigation and management strategies. Sensitivity analysis can
be used to prioritize potential critical control points CCP in MFSPR models and identify corresponding
critical limits. CCPs are points, steps, or procedures in the process of bringing food from farm to the table at
which direct control can be applied, and a food safety hazard can be prevented, eliminated, or reduced to
an acceptable level.
23,24
A critical limit CL is a cri- terion that must be met for each preventive measure
associated with a CCP.
1.3. Selection of the Sensitivity Analysis Method and the MFSPR Model for the Case Studies