FSIS FSIS Listeria Listeria Risk Risk Assessment Assessment Daniel Gallagher Daniel Gallagher Dept. of Dept. of Civil & & Environmental Engineering Environmental Engineering Virginia Tech Virginia Tech Eric Ebel and Janell Kause FSIS Risk Assessment Division USDA Listeria Public Meeting February 26, 2003
62
Embed
FSIS Listeria Risk Assessment Daniel Gallagher Dept. of & Environmental Engineering Dept. of Civil & Environmental Engineering Virginia Tech Eric Ebel.
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
FSIS FSIS Listeria Listeria Risk Assessment ModelRisk Assessment Model
Data Inputs and AssumptionsData Inputs and Assumptions
Model ImplementationModel Implementation
Risk Assessment OutputsRisk Assessment Outputs
Summary of FindingsSummary of Findings
FDA Risk Ranking of RTE ProductsFDA Risk Ranking of RTE Products
Food Category
Deli M
eats
Deli S
alads
Past.
Milk
Frank
furte
rs
Misc
Dair
y
Smok
ed S
eafo
od
Soft C
hees
e
Mea
t Spr
eads
CRTEC
HTNC
Med
ian
Num
ber
of L
iste
rioso
s C
ases
0
200
400
600
800
1000
1200
1400
1600
Approach: Relative Risk Approach: Relative Risk Ranking of Food Ranking of Food CategoriesCategories
Purpose: Identify foods Purpose: Identify foods that pose the greatest that pose the greatest public health risk and public health risk and focus resources focus resources accordinglyaccordingly
Evaluated Lm from retail Evaluated Lm from retail to public healthto public health
Examine the effectiveness of testing and Examine the effectiveness of testing and sanitation of food contact surfaces on mitigating sanitation of food contact surfaces on mitigating product contamination, and reducing the product contamination, and reducing the subsequent risk of illness subsequent risk of illness
Evaluate the effectiveness of other interventions Evaluate the effectiveness of other interventions (e.g., pre- and post-packaging interventions)(e.g., pre- and post-packaging interventions)
Provide guidance on how frequently to test and Provide guidance on how frequently to test and sanitize food contact surfaces for sanitize food contact surfaces for ListeriaListeria spp. spp.
Possible Sources of Lm in RTE Possible Sources of Lm in RTE Products at RetailProducts at Retail
Inadequate lethality during processingInadequate lethality during processing
Direct deposition from non-food contact Direct deposition from non-food contact surfacesurface
Transfer from food contact surfaceTransfer from food contact surface
Focus on food contact surfaces as the source of Lm
Model DescriptionModel Description
Dynamic “in-plant” Dynamic “in-plant” Monte CarloMonte Carlo model model predicts Lm concentrations at retailpredicts Lm concentrations at retail
Coupled with an updated version of the FDA Coupled with an updated version of the FDA ListeriaListeria risk assessment model to predict risk assessment model to predict human health impactshuman health impacts
Mass balance approach –track bacteria as Mass balance approach –track bacteria as move from one media to anothermove from one media to another
The risk of illness or death on a per annum basis The risk of illness or death on a per annum basis from Lm in deli meat as a function of:from Lm in deli meat as a function of:
– Testing (Testing (Listeria Listeria spp.) and sanitation frequency spp.) and sanitation frequency (based on plant size) of food contact surfaces(based on plant size) of food contact surfaces
– Testing (Lm) and disposition of RTE productTesting (Lm) and disposition of RTE product
– Pre- and post-packaging interventionsPre- and post-packaging interventions
The likelihood of detecting Lm in a product lot if The likelihood of detecting Lm in a product lot if a FCS tests positive for a FCS tests positive for Listeria Listeria spp.spp.
Data source: FSIS IDV data Data source: FSIS IDV data
Date description: Lspp Date description: Lspp prevalence over time for prevalence over time for various food contact surfacesvarious food contact surfaces
Method: Fit with survival Method: Fit with survival analysisanalysis
Results: Log normal Results: Log normal distributiondistribution– Mean time between Mean time between
contamination events: 23.1 dayscontamination events: 23.1 days– Standard deviation: 38 daysStandard deviation: 38 days
Data Source: Tompkin Data Source: Tompkin 20022002
Data description: Data description: Sequential weekly Lspp Sequential weekly Lspp positives for food contact positives for food contact surfacessurfaces
Method: Fit with survival Method: Fit with survival analysisanalysis
Results: log normal Results: log normal distributiondistribution– Mean: 8.8 daysMean: 8.8 days– Standard deviation: 2.1 daysStandard deviation: 2.1 days
1.80
2.20
2.60
3.00
3.40
0.40 0.70 1.00 1.30 1.60
Lognormal Probability Plot
Lognormal Quantile
Ln(T
ime)
Transfer CoefficientsTransfer Coefficients
Data Source: published literatureData Source: published literatureResults:Results:– Montville et al. (2001) and Chen et al. (2001) found Montville et al. (2001) and Chen et al. (2001) found
that transfer coefficients of bacteria were log normally that transfer coefficients of bacteria were log normally distributed. Range: 0.01% to 10% Standard distributed. Range: 0.01% to 10% Standard deviation ~ 1 logdeviation ~ 1 log
– Midelet & Carpentier (2002) found that after 12 Midelet & Carpentier (2002) found that after 12 contacts, 60%-100% of initial Lm transferred, ~100% contacts, 60%-100% of initial Lm transferred, ~100% of other bacteria.of other bacteria.
LMRA uses a log mean of –0.14 with a log SD of LMRA uses a log mean of –0.14 with a log SD of 1. Truncate to 100% at max.1. Truncate to 100% at max.
Lspp to Lm RatioLspp to Lm Ratio
Data Source: prevalence data provided in Tompkin Data Source: prevalence data provided in Tompkin (2002) and blinded industry data (no available (2002) and blinded industry data (no available ListeriaListeria level data) level data)
Assumption: prevalence ratio used for level ratioAssumption: prevalence ratio used for level ratio
Results: ratios fit (truncated) normal distributionResults: ratios fit (truncated) normal distribution– Mean: 52.6%Mean: 52.6%– Standard deviation: 26.7% Standard deviation: 26.7%
Production Levels/ Lot Volumes by Plant Production Levels/ Lot Volumes by Plant SizeSize
Data source: FSIS RTE SurveyData source: FSIS RTE Survey
Lot size per line per shift varies by plant Lot size per line per shift varies by plant size:size:– Large: 48%, 19000 lb Large: 48%, 19000 lb ± 14000± 14000– Small: 48%, 7100 lb Small: 48%, 7100 lb ±± 10600 10600– Very Small: 4%, 2800 lb Very Small: 4%, 2800 lb ± 9500± 9500
Assume that FCS area varies Assume that FCS area varies proportionately.proportionately.
Contamination EventContamination EventAdded Lspp to Food Contact Added Lspp to Food Contact
SurfaceSurfaceSource: No available data (calibrated Source: No available data (calibrated data input)data input)
Assumption: Added levels are log Assumption: Added levels are log normally distributednormally distributed
Method: Calibrate the added levels so Method: Calibrate the added levels so that the predicted Lm distribution at that the predicted Lm distribution at retail matches the FDA retail distributionretail matches the FDA retail distribution
Model CalibrationModel CalibrationFDA Lm concentration distribution at retailFDA Lm concentration distribution at retail
Prevalence of Lm in productPrevalence of Lm in product– Levine et al. (2001) found prevalence from 0.52% - Levine et al. (2001) found prevalence from 0.52% -
5.16% in RTE meat and poultry. Generally 5.16% in RTE meat and poultry. Generally decreasing with time. Generally 1-3% in 1999.decreasing with time. Generally 1-3% in 1999.
– Luchansky (in press) found 1.6% prevalence in Luchansky (in press) found 1.6% prevalence in frankfurter packagesfrankfurter packages
– NFPA (2002) found 0.9% prevalence in RTE meatNFPA (2002) found 0.9% prevalence in RTE meat
23 Lm positives out of 997 samples23 Lm positives out of 997 samples
prevalence of 2.3%prevalence of 2.3%
Model Implementation and Model Implementation and Baseline DataBaseline Data
Data Entry Screens and Baseline DataData Entry Screens and Baseline Data
Data Entry Screens and Baseline DataData Entry Screens and Baseline Data
Data Entry Screens and Baseline DataData Entry Screens and Baseline Data
Data Entry Screens and Baseline DataData Entry Screens and Baseline Data
Data Entry Screens and Baseline DataData Entry Screens and Baseline Data
Data Entry Screens and Baseline DataData Entry Screens and Baseline Data
Output ScreensOutput Screens
Output ScreensOutput Screens
Results: Lot TimelineResults: Lot Timeline
Post processing Growth ?Post processing Growth ?FDA model assumes about 1.9 logs of growth FDA model assumes about 1.9 logs of growth on average between processing and retail.on average between processing and retail.
SourceSource % at Processing% at Processing % at Retail% at Retail
FSISFSIS 1% - 3%1% - 3%
measuredmeasured
??
NFPANFPA ?? 0.9%0.9%
measuredmeasured
LMRA uses same approach as FDA but with less LMRA uses same approach as FDA but with less growth (1 log vs 1.9 logs). Lack of variability may growth (1 log vs 1.9 logs). Lack of variability may impact growth-inhibiting-packaging conclusions.impact growth-inhibiting-packaging conclusions.
Analysis of growthAnalysis of growth
ProductionLog(Lm per gram)
GrowthLog(Growth)
=RetailLog(Lm per gram)
+
Normal(µ1,σ1) Normal(µ2,σ2) Normal(µ1+µ2, σ21 +σ2
2)+ =
Given a retail distribution,solve for production distributionfor different assumed growth distributions.
Then examine implied sample prevalence levelsassuming a test positive threshold of 1 Lm organismin 25 gram samples.
Comparison of FSIS Model Outputs to Comparison of FSIS Model Outputs to FDA Risk Assessment InputsFDA Risk Assessment Inputs
Percentile
75 80 85 90 95 100 105
Lm in
RT
E p
rodu
ct a
t ret
ail
(cfu
/g)
1e-7
1e-6
1e-5
1e-4
1e-3
1e-2
1e-1
1e+0
1e+1
1e+2
1e+3
1e+4
1e+5
1e+6FDA
LMRA Calibration
Baseline predictionsBaseline predictions
Calibration to “average” Calibration to “average” updated FDA/FSIS updated FDA/FSIS model Lm model Lm concentration in retail concentration in retail deli meat distributiondeli meat distribution
Lack of uncertainty Lack of uncertainty about concentration about concentration has small effect on has small effect on uncertainty about public uncertainty about public health predictionshealth predictions
Median, 5Median, 5thth and 95 and 95thth percentiles percentilesin predicted elderly deathsin predicted elderly deaths
Variability of 20 runs of 4-2-1 scenario(1,000,000 lots per run)
Percentiles
Lm C
onc
entr
atio
n a
t Ret
ail
1e-7
1e-6
1e-5
1e-4
1e-3
1e-2
1e-1
1e+0
1e+1
1e+2
1e+3
1e+4
1e+5
1e+6
1e+7
Q80 Q99 Q99.99
Model StabilityModel Stability
Model ResultsModel Results
Scenarios TestedScenarios TestedBaseline calibration: no testing, no interventions, no post-processing, no Baseline calibration: no testing, no interventions, no post-processing, no GIPGIPFCS Testing Levels, test and hold yes, dispose product yes, test lot yes, FCS Testing Levels, test and hold yes, dispose product yes, test lot yes, enhanced cleaning yes (No. tests per line per month for large, small, very enhanced cleaning yes (No. tests per line per month for large, small, very small plants)small plants)– 4-2-14-2-1– 8-4-28-4-2– 10-10-1010-10-10– 16-8-416-8-4– 32-16-832-16-8– 60-60-6060-60-60
60-60-60 Lot testing, dispose product yes60-60-60 Lot testing, dispose product yes100% post-processing treatment (90% - 95% effective) for all three plant 100% post-processing treatment (90% - 95% effective) for all three plant sizes, no testingsizes, no testing100% growth-inhibiting packaging (90% 95% effective) for all three plant 100% growth-inhibiting packaging (90% 95% effective) for all three plant sizes, no testingsizes, no testing
All scenarios tested for production of 1,000,000 lots.
Overall FCS prevalence approximately constant at Overall FCS prevalence approximately constant at ~13-14 % regardless of test and hold.~13-14 % regardless of test and hold.
Overall lot prevalence 15-16% if test and hold, 4-5% Overall lot prevalence 15-16% if test and hold, 4-5% if not test and holdif not test and hold
Recall overall lot prevalence ~2.2%Recall overall lot prevalence ~2.2%
For retail Lm, test and hold only significant at more For retail Lm, test and hold only significant at more frequent FCS testingfrequent FCS testing
Evaluation of prevalence for Evaluation of prevalence for different Lm/Lspp ratiosdifferent Lm/Lspp ratios
Parameter Low Ratio Baseline High Ratio Mean Lm/Lspp ratio 0.052 0.52 0.95 Std dev Lm/Lspp ratio 0.026 0.26 0.026 Mean Lspp/cm2 added during contamination event (log scale)
-5 -6 -6.4
Std dev Lspp/cm2 added 3.5 3.5 3.5 overall lot prevalence (%) 2.2 2.2 2.0 overall FCS prevalence (%) 18.7 13.8 12.0 contingent lot prevalence when FCS is positive (%)
11.7 15.7 17.0
Improvement 5.3 7.1 8.5
Each new ratio requires a recalibration to match the observed Lm distribution at retail. These results are preliminary
RTE sampled mass, RTE sampled mass, retail Lm retail Lm– Mass should be limited only by lab Mass should be limited only by lab
considerations.considerations. FCS area sampled, FCS area sampled, retail Lm retail Lm– Caution: assumes Lm evenly distributedCaution: assumes Lm evenly distributedFCS testing is effective for a wide range of FCS testing is effective for a wide range of Lm/Lspp ratios. The effectiveness is higher at Lm/Lspp ratios. The effectiveness is higher at higher ratios.higher ratios. Post processing & industry participation, Post processing & industry participation, retail Lmretail Lm
Public Health ImpactsPublic Health Impacts
0
50
100
150
200
250
300B
asel
ine
4-2-
1
8-4-
210
-10-
1016
-8-4
32-1
6-8
40-2
0-10
60-6
0-60
60-6
0-60
RTE
PP
-95%
PP
-99% GIP
PP
-95%
& G
IP
Scenarios
Eld
erl
y d
ea
ths
pe
r a
nn
um
; M
ed
ian
Pre
dic
tio
ns
Predicted elderly deaths from deli meats
0
10
20
30
40
50
60
Baseli
ne4-
2-1
40-2
0-10
60-6
0-60
60-6
0-60
RTE
PPGIP
PP & G
IP
Scenarios
Inte
rmed
iate
ag
e d
eath
s p
er a
nn
um
; M
edia
n P
red
icti
on
s
Predicted intermediate age deaths from Predicted intermediate age deaths from
deli meatsdeli meats
Predicted neonatal deaths from Predicted neonatal deaths from deli meats deli meats
Test and control combinationsTest and control combinations
050
100150200250300
Bas
elin
e4-
2-1
4-2-
1 P
P4-
2-1G
IP4-
2-1
PP
&G
IP32
-16-
832
-16-
8 P
P32
-16-
8 G
IP32
-16-
8 P
P&
GIP
60-6
0-60
60-6
0-60
PP
60-6
0-60
GIP
60-6
0-60
PP
&G
IP
Scenarios
Eld
erly
dea
ths
per
an
nu
m;
Med
ian
Pre
dic
tio
ns
Note: Testing is non-additive with post-processing treatment.
Model VariablesModel Variables
Only considers food contact surface as Only considers food contact surface as source of Lspp/Lm in productsource of Lspp/Lm in product
Only a “generic” food contact surfaceOnly a “generic” food contact surface
Assumes Lspp evenly distributed across Assumes Lspp evenly distributed across food contact surface, and Lm evenly food contact surface, and Lm evenly distributed within productdistributed within product
Operates on a product lot basisOperates on a product lot basis
Summary FindingsSummary FindingsFood contact surfaces found to be positive for Food contact surfaces found to be positive for Listeria Listeria species species greatly increased the likelihood of finding RTE product lots greatly increased the likelihood of finding RTE product lots positive for Lm (x7 if test and hold, x2 if not) ).positive for Lm (x7 if test and hold, x2 if not) ).
Frequency of contamination of FCS with Frequency of contamination of FCS with Listeria Listeria species appears species appears to encompass a broad timeframe, and the duration of to encompass a broad timeframe, and the duration of contamination lasts about a week.contamination lasts about a week.
The proposed minimal frequency of FCS testing/sanitation, as The proposed minimal frequency of FCS testing/sanitation, as presented in the proposed rule (66 FR 12589, Feb. 27, 2001) presented in the proposed rule (66 FR 12589, Feb. 27, 2001) results in a small reduction in the levels of Lm in deli meats at results in a small reduction in the levels of Lm in deli meats at retail.retail.
Increased frequency of testing/sanitation leads to proportionally Increased frequency of testing/sanitation leads to proportionally lower risk of listeriosis. lower risk of listeriosis.
Combinations of interventions appear to be much more effective Combinations of interventions appear to be much more effective than any single intervention in mitigating the potential than any single intervention in mitigating the potential contamination of RTE product with Lm and reducing the contamination of RTE product with Lm and reducing the subsequent risk of illness or death.subsequent risk of illness or death.