A Simulation System for HPAI Diseases Control
ABSTRACT
As Highly Pathogenic Avian Influenza (HPAI) had become an issue in Korea,
Korean Government has decided to make a complete defense system of HPAI
using system dynamics and agent based modeling techniques after a preliminary
research. This paper summarizes the simulation part of the Korean Government
efforts on HPAI.
KAHIS (Korea Animal Health Integration System) is designed to help government
agencies to perform their HPAI (and other animal diseases) related tasks on their
computers. The simulation part of KAHIS(KAHIS-Sim in short) is designed to
support tasks such as HPAI possibility evaluation, retrospective analysis, diffusion
analysis, defense line analysis, and social cost evaluation.
KAHIS-Sim consists of a user interface, and two models (AIR Model; and AID
Model), which are programmed using Vensim DSS. AIR Model forecasts the
possibility of HPAI occurrence for 4967 regions and more than 20,000 farms in
Korea. AID Model simulates the diffusion processes of HPAI and estimates the
economic impacts. The user interface summarizes and shows the simulation
results using geometry information system.
After test trials of the system, a few issues have been raised, including the
simulation time, lacking data, and linkage with the weather forecast. Nevertheless,
they have decided to use the system as references for their tasks. They have
planned to upgrade the system for another next 4 years.
INTRODUCTION
Highly Pathogenic Avian Influenza (HPAI) has become a hot issue recently. This
nationwide concern stems from three HPAI cases; year 2003-2004 winter, year
2006-2007 winter, and 2008 spring as summarized in the Table 1. There have
been huge social concerns not to speak of economic impacts since chickens are
major food materials.
<Table 11> —_ History of HPAI Events in Korea
Year Number of Infected Farms Remarks
2003/12 — 2004/3 19
2006/11 - 2007/3 7
2008/4 — 2008/5 33 11 Cities and Provinces
Sum 59
Sources: [1], [2], [3]
When a HPAI case occurs, many government agencies from local government to
central government with diverse expertise become involved in the case.
Sometimes it is not easy to maintain consistencies between them especially if the
viruses are spread over many local governments.
The spread of HPAI viruses follows the general rule of diffusion, which is a typical
feedback problem. In other words, the spread of HPAI diseases cannot be simply
guessed or assessed. Furthermore, every decision has to be made quickly since
the timeliness is a key factor for controlling the diseases [4].
Korean Government has decided to make a complete defense system using
system dynamics technique and agent based modeling technique after 6 month
preliminary research.[4] This paper summarizes the simulation part of the Korean
Government efforts on HPAI.
OVERVIEW OF THE SYSTEM
KAHIS (Korea Animal Health integration System) is designed to help government
agencies to perform their HPAI related tasks on their computers. Therefore the
simulation is one of the major functions of the system KAHIS. However, every
feature is designed to make the simulation function workable since all the
information is required to perform simulations before or after HPAI occurrences.
The simulation part of KAHIS (from now on, KAHIS-Sim in short) is designed to
perform tasks as shown in Table 1(defined from the previous research [4]). Most
of the time they are interested in the HPAI possibility since HPAI has been
occurred once in couple of years.
<Table 2> Tasks and Models
Tasks Usages Models Inputs
Seasons,
HPAI Possibility Evaluations | Before Occurrence AIR Surveillances
Evidences, etc.
Retrospective Analyses AID
Diffusion Analyses AID
- After Occurrence Reported Cases
Defense Line Analyses AID
Social Cost Evaluations AID
OVERVIEW OF THE MODELS
Although two models are involved with KAHIS-Sim, the two models can be
merged into one big model. Since simulation time has become an issue, the
model is divided into two models (AIR Model; Aviation Influenza Risk Model and
AID; Aviation Influenza Diffusion Model) to reduce the memory used.
Figure 1 shows the overall causal loop diagram of KAHIS-Sim. As shown in the
figure, conceptually AID Model covers AIR Model. AIR Model is applied only for
forecasting of HPAI occurrence possibilities before any real HPAI situation occurs.
On the other hand, AID Model is for the situation where any real HPAI case
occurs. Since we do not know if the original HPAI virus sources are still active, all
logics in AIR Model should be included in the model as shown in the figure.
However, since there is no infected case before HPAI occurrences, AIR Model
does not need diffusion logics.
.
GIS information fp Neighbor Risk 4 _ F arm information. -¢ Farm information
ton 7 Factor rf
\
® ossibility of HPAI Se a a | \,
/ Introduction via = 3
f Migration Bird Individual Farm Conditional Probabilty
/ o
B cabins - re
Migration Birds Regional --,
—_srabnt cans
Background Risk Farm Pith eae
Possibility of HPAL____— =ao-=-= eg
Virus in Foreign € d Defense Logic SEI SOE
Employee ‘
Possibility of HPAL wee rrafsterfna velicles
Virus in Feedstuff
4
'\Controled Farms Economic impaét__, Compensation to
= Comp
XQ zone Ax to farms x, Farms x
Possibility of HPAL
Virus in S muggled
Birds
q EcOnamic Impact
. | |
Possibility of HPAL
Vinsintoveles i ee vee Na
oo Economic impact to generation stage ¢ ompensation to Farms
¢ Weather information —_c Oversea Information Economic impacto logistic stage Economic impacto final stage
<Figure 1> Overall Causal Loop Diagram
HPAI Occurrence Possibilities (AIR Model)
The main purpose of AIR Model is to forecast the possibility of HPAI occurrence.
The model assumes 6 pathways for HPAI to be introduced into Korea as shown
the Figure 2(summarized from previous researches; [1], [2], [3], [5], [6], [7]). The
model predicts risk factors for both 4967 regions and 1500 poultry farms (7000
farms are empty for future uses) using subscript variables.
1
1
Background }if Neighbor }
1
Loa.
Farm }
Characteristics
Migration
Birds
Foreign —
\__Employee _)
(ForeignFeed )
Matrial
Imported
Cattles
Smuggled _
Annimals
Tourists
Evidences
<Figure 2> HPAI Occurrence Possibility Structure (Pathways of Introduction)
Once calculated via models, the simulation results are stored in the database.
Then users may access the data in various forms using user interfaces. For
example, users can play a video for one year period changing colors as the risk
level does on the map of Korea with zoom in and out functions.
Diffusion Model (AID)
Diffusion model (AID Model) is prepared to analyze the diffusion processes of
HPAI after the first reporting of HPAI case. The model assumes that there must be
at least one influenza source among the 6 pathways of introduction, which no
one knows when the first infected farm is reported. For example, the HPAI
source could be a wetland near the first reported farm. It means that diffusions
could be made through that wetland without noticing if we neglect the fact.
Therefore the usage of the AID Model should be extended to the retrospective
analyses. The assumed methodology of retrospective analyses is Monte Carlo
simulation technique starting with random sampling of initial conditions with
maximum utilization of the information of the first reported farm as well as
surveillance information gathered so far. For example, the closer wetland from the
farm may have more chance of existence of HPAI than the further one.
As for the diffusion processes between farms, 8 diffusion modes (see Table 3) are
taken into account. Parameters characterizing the diffusion processes may vary
depending upon the types of farms. AID Model can describe up to 15,000 poultry
farms, and the types of farms described in the model is summarized in Table 4.
<Table 3> Diffusion Modes
Modes Examples Remarks
Workers .
Transfer via
Human Non-Workers
. — clothes, shoes, etc.
Professional Visitors Vet., etc.
. Material Vehicles
Equipments ,
Equipment Share
Direct Trade
Only for traditional
Trades Market .
chickens
Origin
<Table 4> — Farm Types
Species Groups Numbers Remarks
Chicken Ordinary | Egg Chickens
Farms Meat Chickens
Traditional Chickens
Origin Egg Chicken Origin
Meat Chicken Origin
Traditional Chicken Origin
Duck Ordinary | Meat Ducks
Farms Origin Egg Duck Origin
Meat Duck Origin
Pig Farm ~20,000
Social Cost Model
HPAI diseases results in catastrophic situation for the infected farms because of
the strong contagiousness. If we consider the side effects, it becomes more
serious. All together, the social costs of HPAI diseases can be summarized as
shown in Figure 3.
Retailer Damage
Consumption
Stage Cost
Restaurant sales decrease
Duck damage distribution phase
Logistic Stage Damage caused by egg consumption
Social Cost
Cost
Chicken meat damage distribution phase
Control Costs
Production Stage Direct
Cost Compensation
farm Economic Impact
Subsidies
<Figure 3> Social Cost Structure
As soon as the Government concludes “a new HPAI case,” it declares three
defense circles centering the reported farm. Within the first defense circle, all
poultries will be killed with some compensation. Between the first and second
lines, local government can decide whether to kill the poultries or not. Around
the third largest circle, Government will post guards and limit the all the
transportation of poultries. Therefore the production stage costs directly depend
on the defense activities. Since the third circle covers huge area, the logistic stage
cost is also very dependent on the Government decisions. Besides the direct
effects on the all restaurants in the defense area, there will be psychiatric impact
across the whole country for a while. In short, as the defense area becomes large,
the direct costs become larger, and the potential risk decreases. The model takes
the sizes and shapes as decision variables so that users can simulate the impact
of those decision variables.
SIMULATION RESULTS
HPAI Occurrence Possibilities
Risks involved with HPAI diseases are strongly seasonal mainly because of
migration birds as well as the temperature. Figure 4 shows risks (o farm risk) for
two farms in different locations. The figure shows that farm “CEF1” varies a lot,
while farm “CMF3780" stays in low values.
Farm
0.4
0.2666
Percents
0.1333
A A , , , i , , , , ,
0
0 36 72 108 144 180 216 252 288 324 360
Time (Day)
o farm risk x[CEF1] : Test836 —+— +—+—_+—+
o farm risk x [CMF3780] : Test36
<Figure 4> A Simulation Example of Farm HPAI Occurrence Possibilities
Figure 5 shows risk factors for two regions (respectively related farm CEF1 and
CMF3780). The variables with “cat” in their names give, through user interfaces,
one of the four different colors as shown in Figure 6.
0.8 Percents 2 | lal
3 Dmnl i a4
0.5333 Percents f ApS
2 Dmnl / / \
4
0.2666 Percents \
1 Dmnl
0 Percents
Omni
0) 54 108 162, 216 = 270 324
Time (Day)
© region background risk x[R1] : Testa6. ——+—_+__+__+___ Percents
o region background risk x[R2] : Test36_ ———2——_2__2—_2— _ Percents
o region risk catx[R1] : Test36 8 Dmnl
o region risk catx[R2] : Test36 Omni
<Figure 5> A Simulation Example of Regional HPAI Occurrence Possibilities
Location Date for the values
O HPAL watts Al Be0l <4 —
© ABalols $523] 2a] |x [a
1
14199506573 1
.0007408924 1
ooses499989 1
Values for risk factors.
— lal
<Figure 6> Regional HPAI Occurrence Possibilities in User Interface
If a user selects a specific region or farm, as well as time of the year, the values
will shows up on the left hand side. The right hand side shows a Korean map
which is composed of 4963 local areas. Each local area has its own color
matching with the simulation results.
The major contributors for the variations of the values are the locations of
migration birds. Hundred kinds of birds exist in Korea. However, 31 kinds are
considered important for HPAI diseases, and most of them migrate along the
wetlands. Most wetlands are located either in the west coast or east coast. Figure
7 shows two types of bird migrations described in the model. The migration
paths are designed as external variables, which means system managers can
upgrade the paths any time. At the time point when the system is developed, no
detailed researches have been made about bird migrations. Therefore we adopted
two major migration paths as shown in the figure with random variations.
Bird_Location_West Bird_Location_East
2M 2M
15M 15M
500,000 500,000
0 °
200000 300000 00000 ‘600000 200000 300000 "500000 ‘00000
400000 “200000
bird x location[81] bied x location [824]
bird y location{81] : Test36
bicd y location(824] : Test36
<Figure 7> A Simulation Example of Birds’ Migration (East Coast Birds)
Retrospective Analyses
Retrospective analysis is a process to figure it out how the reported farm was
infected. Mote Carlo simulation technique is adopted for retrospective analyses.
However, more intelligence has to be involved in order for the simulation
numbers to be controllable.
They believe that the compensation for HPAI infection is large enough that the
farmers are willing to report if they knew the fact. Therefore it is very likely that
the reported farm is the first infected farm. Furthermore, although we are
considering 6 pathways, introduction of HPAI virus through the migration birds is
the most likely pathway. So random sampling of initial values is made with higher
weighting factors for those variables related to bird migration and nearest
wetlands.
After 200 simulations, the user interface summarized the cases where the
reported farm is infected with optional additional conditions. Figure 8 is an
example of those cases. In this case the first infected farm and the reported farm
are different.
© ABA SS Fal
© ABA Bap
<Figure 8> A Simulation Example of Retrospective Analysis
Diffusion Analyses
After the potential pathway and the first infected farm are assumed via the
retrospective analysis, diffusion analysis starts from those assumptions with
evidences. Obvious evidence is that the reported farm is infected. As time goes
many facts will be informed, and evidences have to be reflected into diffusion
analyses. For example, as time goes on a farm may report the infection to the
authority, or the suspected wetland has been proved that it contains HPAI. These
facts have to be reflected in the model and reflection of them results in less
uncertainty.
Figure 9 shows the mechanism to see the simulation results. If users press the
video play button, the infected farm will turn into red color as time goes on.
Aue|s 4840001 yf ae BacsmeaBsee 2)
$
Re
A a
a K
Press the play r
button to see the i ;
simulation results g nan goy Bee
a Pas 8 ~
a i
O° aaanes aba
= mi a in Tae 3 suai G00, Sa7eh 000
H 1
: y
<Figure 9> A Simulation Example of Diffusion Analysis
Defense Line Analyses
The model also includes defense mechanism. Centering the reported farm, the
authority will announce 3 defense lines, which limits transportations across the
lines, and all the poultries and pigs inside the smallest defense line will be killed.
Figure 10 shows two of the three defense lines. As shown in the figure, users can
identify the farms inside defense lines as well as the details of the farms including
heads of poultries.
Osa sea MBAS OS
sao ce | ea Uo 2a
” ea o> j
oog ay +
seus ag
ecco}
eens H
sens
Bane a
S01 Sa Neve
sauee 30
weenie seins
sia ge)_s5)(_43
aon
<Figure 10> A Simulation Example of Diffusion Lines
Social Cost Evaluations
As mentioned earlier, the social costs are divided into three categories;
production stage, logistic stage and consumption stage. The calculation of the
social costs in the production stage is straightforward since all the infected farms
and numbers of infected poultries are in the simulation results. Social costs in the
logistic stage and consumption stage need many assumptions. All the
assumptions are designed to be modeled as constants, which are listed in the
database table. Users may test those assumed constants with their own values.
Figure 11 shows examples of the social cost evaluations.
a) Summary
7,761,608,539,136
2,518,184,361,984
SU A5 0a ABS 3,573,499,904
Namo BATS 5,173,229,387,776
Be RSs 0
A 7,694, 987,249,664
5,529,285,632)
61,092,003,840
(66,621,289,472)
BAS eS | BARS as es
b) Direct costs
[ae] samme /wopsus | s7ienesea | aEaa
aaseuse | saseu anes
Ae i gHee ens Aeoray
1938 8 = = = ane 288 BAUaS SORE!
Qari 0 253,635,248 0) ) 2,422,216,589,312 95,967,698,944
HBAS ° 0) ) 0 C) 3,573,499,904
BATS 8,050 0
a 8,050 253,635,248. 0 ) 2,422,216,589,312 99,541,198, 848 0
Basse se | SAS ee BSE
c) Logistic costs
(2o| summa | wonsus| a7eaesem [aeoA
| RRECATS 4,111,428,608
22 S824 WH 1,027,857,152
AL SEOA TS 390,000,000
BFE TAS 5,529,285,760,
aaseuse | aaseu anes
d) Consumption stage costs
29) aun | wapstis | eSsesEA| SECA
{48 Boe Ss seq 599,999,971,328)
1S 228 )
ASH OS 22H 18,000,001,024)
BS 82 0S > 24,000,000,
OS 228 0
BS HS 225 0
ABR 2 35,100,004,352)
S55 08% 53,100,005,376
<Figure 11> A Simulation Example of Social Cost Evaluation
CONCLUSIONS
After test trials of the system, following issues have been raised;
© Simulation time is too long to perform any meaningful Monte Carlo
simulations.
© Some data are missing including farms and migration birds
© The connection with weather system needs to be improved
Nevertheless, they have decided to use the system as references for their tasks.
The current KAHIS-Sim is not the end of the Korean Government efforts. They
have planned to upgrade the system for the next 4 years.
REFERENCES
1. SYFTHAAS, 2001 DAAISRAMA YSAAR TM
2. HULITSYAAS, 2003 DAYAR SAA TAY
3. SFLU AAA, 2006~2007 RAMS HUA ASA, A
4. ASF YI, 2009 SFIS Say), Haw |Sye ope
5. Jinping Li, Qianlu Ren and Jianqun Yin, et al., 2004 Study on Transmission
Model of Avian Influenza, IEEE, 2004,
6. World Health Organization WHO. Avian influenza (“bird flu”) fact sheet
[online]. WHO; 2006 Feb. Available = at: — http://www.who.int/
mediacentre/factsheets/avian_influenza/en/index.html# humans. Accessed 1
Aug 2007.
7. BLY , 2007, HPALAAIAD FQ AHSApr|o] Ay SAS Ye Ae
SE] BABA, SHBSUMSAAA BA eal AstApap 2007.12.
8. J. Otte et al., 2008 Impacts of avian influenza virus on animal production in
developing countries, CAB Reviews: Perspective in Agriculture, Veterinary
Science, Nutrition and Natural Resources 2008, 3, No, 080.
9. Michael. PR. Ward, 2007, Geographic information system-based avian influenza
surveillance systems for village poultry in Romania, Veterinaria Italiana, 43(3),
483-489.
10. CRSPH(The Center for food security & Public Health) & OIE, 2009, High
Pathogenicity Avian Influenza: Fowl Plague, Grippe Aviaire, January 8, 2009
11. Arni S.R. Srinivasa Rao, 2008, MODELING THE RAPID SPREAD OF AVIAN
INFLUENZA (H5N1) IN INDIA, MATHEMATICAL BIOSCIENCES AND
12.
13.
14.
ENGINEERING, http://www.mbejournal.org/ Volume 5, Number 3, July 2008 pp.
523-537
Mirzet Sabirovic et al., 2006 HPAI HSN1 SITUATION IN EUROPE AND
POTENTIAL RISK FACTORS FOR THE INTRODUCTION OF THE VIRUS TO THE
UNITED KINGDOM, VITT1200/HPAI developments in Europe,6 July 2006
Rick Quax 2008 Modeling and Simulating The Propagation of Infectious
Diseases Using Complex Networks —A Thesis Presented to The Academic
Faculty, College of Computing Georgia Institute of Technology, August 1. 2008
S. Ayyalasomyajula et al. 2008, A Network Model of H5N1 Avian Influenza
Transmission Dynamics in Domestic Cats, Zoonoese and Public Health
55(2008)497-506.
