Campylobacter Risk Assessment in Poultry
Helle Sommer,
Bjarke Christensen,
Hanne Rosenquist,
Niels Nielsen and
Birgit Nørrung
P r e v a l e n s
C o n c e n t r a t i o n
SLAUGHTERHOUSE RETAIL CONSUMER RISK
Pfarmh.
Ca.bleeding Probability of Infection
Probability of Exposure
• Data examinations – distributions
• Process model building – explicit equations
• Explicit equations/ simulations
• Cross contamination
• What-if-simulations
Slaughter house modules
Data examinations
• Data for 3 different purposes
- prevalence distribution -> slaughterhouse program
- concentration distribution
- model building, before and after a process
• From mean values to a distribution
• Lognormal/ normal –> illustrations
• Same or different distributions –> variance analysis
From mean values to a distribution
Histogram of the 'after bleeding' data
2 2.2 2.4 2.6 2.8 3 3.2 3.4 3.6 3.8 4 4.2 4.4
Concentration of Campylobacter [log10 cfu/g skin]
17 log mean values from different flocks and from 2 different studies
From mean values to a distribution
-1 0 1 2 3 4 5 6 7 8
Concentration of Campylobacter (log10 cfu/g skin)
Oosterom et al.
Mead et al.
Sum (Oo+Me)
17 distributions -> one common distribution
Log-normal or normal distribution ?”True” data structure = simulated data (sim.=)
Assumed distribution (dist.=)Published data = means of 4 samples,6 means from one study
sim.= lognormal(6.9,2.3) dist.= normal or lognormalSamples 1 2 3 4 5 6
1 1.293 2.454 2.742 2.751 2.278 1.4822 3.603 5.548 4.238 3.074 2.485 2.1973 3.283 2.866 2.546 2.351 2.793 2.4244 4.505 4.694 2.311 3.311 3.039 2.745
Mean 3.171 3.890 2.959 2.872 2.649 2.212SD 1.355 1.473 0.871 0.416 0.335 0.536
sim.= lognormal, dist.= lognorm
0 1 2 3 4 5 6 7 8 9 10
Concentration, log scale
sum distribution
6 mean values
sim.= lognormal, dist.= normal
0 2000 4000 6000 8000 10000 12000 14000
Concentration, normal scale
sum distribution
0 20000 40000 60000 80000 100000 120000
3 data points 6 mean values
1 2 3 4 5
sim.= normal, dist.= normal
0 2000 4000 6000 8000 10000 12000
Concentration, normal scale
sum distribution
6 mean values
sim.= normal, dist.= lognormal
2.5 2.7 2.9 3.1 3.3 3.5 3.7 3.9
Concentration, log scale
sum distribution
6 mean values
2000 2500 3000 3500 4000
3.3 3.35 3.4 3.45 3.5 3.55 3.6
sim.= lognormal, dist.= normal
0 2000 4000 6000 8000 10000 12000 14000
Concentration, normal scale
sum distribution
0 20000 40000 60000 80000 100000 120000
3 data points 6 mean values
Mean values calculated back from mean of log values
Samples Log obs.1 Obs.1 Log obs.2 Obs.2 Second best 1 Second best 21 1.29 19.63 2.454 284.282 3.60 4009.82 5.548 352988.873 3.28 1919.22 2.866 734.274 4.50 31974.93 4.694 49441.95
Mean 3.17 9480.90 3.890 100862.34 1479.11 7762.47SD 1.35 15084.31 1.473 169659.87
Reference # samples mean log obs. "mean" obs.Mead et al . (1995) 10 3.7 5011.87
Mead et al . (1995) 10 4 10000.00Mead et al . (1995) 15 3.9 7943.28Mead et al . (1995) 15 3.8 6309.57Mead et al . (1995) 15 3.4 2511.89Mead et al . (1995) 15 3.9 7943.28Mead et al . (1995) 15 3.6 3981.07Mead et al . (1995) 15 3.5 3162.28Mead et al . (1995) 15 4.3 19952.62Mead et al . (1995) 15 3.9 7943.28Mead et al . (1995) 15 3.7 5011.87
Real data set
9 "mean" values
Normal scale
New Danish data
Data after wash
0 50 100 150 200 250 300 350 400 450
Concentration, normal scale
Histogram
0 50 100 150 200 250 300 350 400 450 500
Concntration, normal scale
Data after wash
0 0.5 1 1.5 2 2.5 3
Concentration, log scale
Histogram
02468
10121416
0 0.5 1 1.5 2 2.5 3
Concentration, log scale
Slaughterhouse process
Concentration log10 cfu/g skin Concentration log10 cfu/g skin
Building mathematical models
Concentration level through the slaughterhouse processes
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
after bleeding after scalding afterdefeathering
afterevisceration
afterwach+chill
Co
nce
ntr
atio
n [l
og
cfu
/g]
Modelled
Observed
Old methode
95% confidence limit
95% confidence limit
95% confidence limit
Why new mathematical process models ?
A given proces, neutral
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5
Before a process [log cfu/g]
Aft
er
a p
roc
es
s [
log
cfu
/g]
1 : 1
Explicit mathematical process model
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
5
5,5
0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 5,5
Before a process [log cfu/g]
Aft
er
a p
roc
es
s [
log
cfu
/g]
1 : 1
A given proces, multiplicativ
Explicit mathematical process model
In normal scale
μy = μx / Δμ
100 = 10000/100
In log scale
μlogy = μlogx – Δμ
2 = 4 - 2
A given proces, multiplicativ
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
5
5,5
0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 5,5
Before a process [log cfu/g]
Aft
er
a p
roc
es
s [
log
cfu
/g]
1 : 1
Explicit mathematical process model
In normal scale
μy = μx / Δμ
100 = 10000/100
In log scale
μy = μx – Δμ
2 = 4 - 2
σy2 = β2 · σx
2 Transformation line
y = + β·x
A given proces, multiplicativ
0
12
34
5
67
89
10
0 1 2 3 4 5 6 7 8 9 10
Before a process [log cfu/g]
Aft
er a
pro
cess
[lo
g c
fu/g
]
1 : 1
Δμ
Explicit mathematical process model
Overall model
μy = μx - Δμ
σy2 = β2· σx
2
Local model
Y = + β·xCalculation of
= (1-β)· μx- Δμ
0
0.5
1
1.5
2
2.5
3
3.5
4
4.5
5
5.5
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5 5.5
Before scalding [log cfu/g]
Aft
er s
cald
ing
[lo
g c
fu/g
]
1: 1
Explicit mathematical process model
In normal scale
μy / μx = 158
In log scale
μy = μx - 2.2
A given proces, additive process
0
0,5
1
1,5
2
2,5
3
3,5
4
4,5
5
5,5
0 0,5 1 1,5 2 2,5 3 3,5 4 4,5 5 5,5
Before a process [cfu/g]
Aft
er a
pro
cess
[cfu
/g]
1 : 1
Explicit mathematical process model
In normal scale
y = x + z
z Є N (μ, σ)
Summing up
• Explicit equations for modelling slaughterhouse processes + Monte Carlo simulations, modelling each chicken with a given status of infection, concentration level, order in slaughtering, etc.
• New data of concentration (input distribution) -> different or same distribution ? (mean and shape)
• Data + knowledge/logical assumptions of the process -> multiplicativ or additive process
Advantage with explicit equations
• Accounts for homogenization within flocks
• More information along the slaughter line does not give rise to more uncertainty on the output distribution.
• Faster than simulations/Bootstrap/Jackknifing