Adapting existing models of highly contagious diseases to countries other than their country of origin.

Abstract

Many countries do not have the resources to develop epidemiological models of animal diseases. As a result, it is tempting to use models developed in other countries. However, an existing model may need to be adapted in order for it to be appropriately applied in a country, region, or situation other than that for which it was originally developed. The process of adapting a model has a number of benefits for both model builders and model users. For model builders, it provides insight into the applicability of their model and potentially the opportunity to obtain data for operational validation of components of their model. For users, it is a chance to think about the infection transmission process in detail, to review the data available for modelling, and to learn the principles of epidemiological modelling. Various issues must be addressed when considering adapting a model. Most critically, the assumptions and purpose behind the model must be thoroughly understood, so that new users can determine its suitability for their situation. The process of adapting a model might simply involve changing existing model parameter values (for example, to better represent livestock demographics in a country or region), or might require more substantial (and more labour-intensive) changes to the model code and conceptual model. Adapting a model is easier if the model has a user-friendly interface and easy-to-read user documentation. In addition, models built as frameworks within which disease processes and livestock demographics and contacts are flexible are good candidates for technology transfer projects, which lead to long-term collaborations.

Full text

Page 1

Rev. sci. tech. Off. int. Epiz., 2011, 30 (2), 581-589
Adapting existing models of highly
contagious diseases to countries other
than their country of origin
C. Dubé(1), J. Sanchez(2)& A. Reeves(3)
(1) Animal Health and Production Division, Canadian Food Inspection Agency, 59 Camelot, Ottawa, Ontario,
K1A 0Y9, Canada
(2) Department of Health Management, Atlantic Veterinary College, University of Prince Edward Island,
550 University Avenue, Charlottetown, PEI, C1A 4P3, Canada
(3) Animal Population Health Institute, Department of Clinical Sciences, College of Veterinary Medicine and
Biomedical Sciences, Colorado State University, Fort Collins, CO 80523, United States of America
Summary
Many countries do not have the resources to develop epidemiological models of
animal diseases. As a result, it is tempting to use models developed in other
countries. However, an existing model may need to be adapted in order for it to
be appropriately applied in a country, region, or situation other than that for
which it was originally developed. The process of adapting a model has a
number of benefits for both model builders and model users. For model builders,
it provides insight into the applicability of their model and potentially the
opportunity to obtain data for operational validation of components of their
model. For users, it is a chance to think about the infection transmission process
in detail, to review the data available for modelling, and to learn the principles of
epidemiological modelling. Various issues must be addressed when considering
adapting a model. Most critically, the assumptions and purpose behind the model
must be thoroughly understood, so that new users can determine its suitability
for their situation. The process of adapting a model might simply involve
changing existing model parameter values (for example, to better represent
livestock demographics in a country or region), or might require more substantial
(and more labour-intensive) changes to the model code and conceptual model.
Adapting a model is easier if the model has a user-friendly interface and easy-
to-read user documentation. In addition, models built as frameworks within
which disease processes and livestock demographics and contacts are flexible
are good candidates for technology transfer projects, which lead to long-term
collaborations.
Keywords
Epidemiological modelling – Model adaptation – Technology transfer.
Introduction
Models of highly contagious diseases are increasingly being
used in many countries of the world to support
contingency planning efforts (1, 4, 5, 16, 20) and to study
ongoing (3, 11) or past epidemics (7, 13). Some models
are developed for quite specific circumstances: they might
be developed for use in a specific region or country, for a
specific disease, or based on a specific epidemic. Other
models are more generic, in that they allow the simulation
of highly contagious diseases in different populations,
regardless of their location in the world.
The process of building models for infectious diseases is
dependent on various factors, such as the objective or
scope of the model, the disease status of the country
developing the model, the experience of the country in
responding to such diseases, the availability of data and
knowledge to build and validate the model, and the
experience of the modellers themselves (18). The choice of
the approach used for modelling (for example,
deterministic or stochastic, mathematical or simulation,
spatial or non-spatial) will be affected by all these factors.
As a result, models developed may range from very specific
models targeting a specific disease, or even a specific

Page 2

epidemic in a defined population, to general model
frameworks that can be used to model a variety of highly
contagious diseases and populations.
The model-building process also requires resources,
including:
– individuals with specialised skills in computer
programming, epidemiology, economics, sociology, and
mathematics
– funds to support model development and evaluation
– funds to support data collection and analysis if required
– training opportunities for users of the tools developed
(these kinds of resources are not available in all regions of
the world).
As model building is a resource-intensive process, models
developed in one country or region might be considered
and used by other countries or regions. This type of
collaboration has been facilitated by the ease of global
communications and the development of virtual work
collaborations.
In some situations, it may be helpful to ‘adapt’ a model to
represent the characteristics of disease transmission in a
country other than in its country of origin. The process of
adaptation might include a number of different processes
that fall into two general categories:
a) adapting the parameter values used to inform the
model, or
b) more fundamentally altering the conceptual model so
that it better fits the new situation in the other country.
How these changes are implemented is highly dependent
on how the model was built. This paper discusses the
benefits of using existing models in situations they were
not originally designed for and examines the ways in which
they may need to be adapted for these types of
circumstances. The authors use different examples from
the literature and from their own experiences in using the
North American Animal Disease Spread Model (NAADSM)
(5).
Benefits of using already
existing models in countries
other than the country of origin
There can be great benefits for both the model builders and
for the users of these models. For the model builders, using
existing models might allow them to have access to data
not available in their own country to support
improvements to, or testing of, their own models. For
example, data on animal movements might be available in
the recipient country which can be used to evaluate how
contacts are represented in the model. This type of work
can also provide modellers and epidemiologists who
usually work in disease-free countries with experience of
working in countries where disease is endemic. Moreover,
by having the model reviewed by experts in the field, this
process can also be used as a validation of the conceptual
model on which the model was built, which will help
increase the level of confidence in model outputs. Finally,
when an existing model is used in another country as a tool
to assist eradication and control measures, the global risk
of transmission of an infection is further diminished.
For the recipient country, this is a great opportunity to
build technical expertise in epidemiological modelling and
to think about the disease transmission process and control
of diseases in a structured way. The process of developing
parameters for epidemiological models pushes one to
think about the epidemiology of the disease and also about
the response mechanisms in place. Using a model will not
only assist countries to respond in outbreak situations, but
can be a valuable tool to guide future response strategies.
Collaborations among developers and end-users of disease
spread models lead to increased inter-disciplinary access to
specialists who can provide guidance and advice by
analysing outbreak data, livestock movement databases
and other data required to develop parameters for
epidemiological models. Finally, using an existing model is
a low-cost, quickly available alternative to developing a
brand-new model.
The process of model building
and potential for adaptation
A modelling exercise typically starts with a clearly stated
question or purpose. This purpose might be research-
oriented: for example, a modelling study might be
undertaken to develop an understanding of the
epidemiology of a disease, or to conduct a retrospective
analysis of an outbreak to understand how the disease
spread and what could have been done differently to
control the outbreak. The question might also be
motivated primarily by policy: for example, what are the
benefits and costs associated with using vaccination against
foot and mouth disease (FMD) during an outbreak in a
previously FMD-free country? Depending on the question,
the level of knowledge of the epidemiology of the disease,
and the data available to inform a model, very different
models can be built, ranging from simple deterministic
models to complex spatial simulation models. Taylor (18),
in his review of the use of models in informing disease
control policy, provided a table (based on the work of
Holling) (6) showing how the level of epidemiological
Rev. sci. tech. Off. int. Epiz., 30 (2)
582

Page 3

583
Rev. sci. tech. Off. int. Epiz., 30 (2)
knowledge and the quantity/quality of data will define the
types of questions that can be answered with a model
(Table I). If the level of knowledge is high and good quality
data are available in sufficient quantities, reasonably
reliable predictions are possible.
Table I
Representation of the questions which can be explored using
models based on the data and knowledge available
Table reproduced with permission of the author (Taylor) (18)
Epidemiological
knowledge
Quantity / Quality of data
PoorGood
Poor
(a) Exploratory –
hypothesis
development
(c) Empirical/analytical –
hypothesis testing
Good
(b) Simplified
representation of past
events with data
assumptions
(d) Good representation
(‘simulation’) of past events
Guarded predictive use
(‘what if?’) BUT with
uncertainty limits
Can be used predictively
(what if?) IF the future
is predictable
When considering the use of an already existing model in
another region or country, the first task is to carefully
evaluate the assumptions behind the model, and to
determine whether the model was built for a purpose that
is compatible with the needs of the recipient country. For
example, a model built for contingency planning might not
be suitable for making tactical decisions during an
outbreak. Then, it is important to determine the level of
adaptation that will be required. A distinction should be
made between simply changing the value of parameters to
represent the local situation and making modifications to
the structure and computer code of the model, which can
be labour intensive and require specialised programming
skills. The process of changing parameter values may also
be labour intensive, depending on how they are included
in a model. ‘Hard-coded’ parameters (i.e. those included
directly in computer code) will be more difficult to change
than parameters passed to the model in the form of data
files (the existence of a user-friendly model interface can
simplify this process). In addition, the availability and
quality of data must be evaluated in the context of the
model to be used, to determine if these data are applicable
in the receiving country. Although situations where data
are lacking are more typical, it is possible to encounter the
opposite problem: the recipient country might have a level
of detail in their data that cannot be accommodated by an
existing model, and the implications of the loss of this
detail should be considered before using the model.
It is also important to consider the geographical area of
application of a model. Do we want to use the model to
represent only a specific area in a country that is
experiencing a disease outbreak? Or in a region that is very
important economically? Or do we want to use this model
anywhere in the country, or in the world? A modelling
framework built to be used in a specific area of a country
might include specific parameters for that area, but might
only require slight changes in its computer code to be able
to represent other areas with different parameters. An
example of this is the model developed by Bates et al. (1),
which included parameters collected in a three-county
region of California. The rates used in their study are
therefore not necessarily applicable to other regions of the
United States, but the modelling concepts and framework
could be used with other parameters.
Examples of adapting
epidemiological models
in other countries
Three examples will be discussed in this paper. First, the
authors describe a study that used a model built
specifically for the 2001 epidemic of FMD in the United
Kingdom (UK) which was later applied in Denmark.
Secondly, they present some examples of the use of
InterSpread Plus, a model framework used in various
countries of the world. Lastly, they present how the
NAADSM has been used in different countries in South
America.
A number of models were built in 2001 in the UK to
support response efforts to the FMD outbreak (3, 8, 9, 11).
The model by Keeling et al. (11) was subsequently
modified and used to support contingency planning for
future outbreaks in the UK (10, 20). The model uses a
disease transmission kernel which accounts for all
transmission mechanisms simultaneously and provides a
function of risk of transmission versus distance to an
infected farm. In 2008, Tildesley and Keeling (19)
evaluated the utility of the model developed in the UK
when used for situations in different countries. They chose
to apply the model in Denmark as a contingency planning
effort. This work was part of an International EpiLab
model comparison project to evaluate the potential
consequences of FMD introductions into Denmark (21).
The purpose of the project was to identify the most
appropriate control measures (culling, vaccination, or
some combination) to use in various outbreak scenarios.
The main issue for adapting the model was related to the
differences in livestock demographics in the two countries.
Although pigs did not play a significant role in the UK
outbreak of 2001 and the pig density is low in most areas
of the UK, it is very high in Denmark. In addition, the

Page 4

sheep density is very low in Denmark compared to the
density in the UK. Cattle density is equivalent. Denmark is
a much smaller country in terms of area, and the
transmission kernel used in the model to represent the UK
situation was not applicable to the Danish situation. In
addition, due to a lack of precise knowledge, assumptions
had to be made about the susceptibility and infectivity of
swine, and a number of values were tested. Values used for
the susceptibility and infectivity of sheep were similar to
those developed for the conditions of the UK outbreak.
Because of the type of model used – a mathematical model
without a user interface – and because of the significant
amount of modification required to the core of the model,
the modellers were heavily involved in adapting the UK
model to the Danish situation. The main finding of this
study was that the livestock demographics greatly
influenced the control measures recommended by the
model. In Denmark, a uniform national control policy
would not be the optimal strategy due to the varying
livestock demographics in the country.
The second example involves the use of InterSpread Plus
(IS+) in the Republic of Korea to re-create the 2002 FMD
epidemic (Yoon et al., 22). InterSpread Plus is highly
flexible in the sense that users specify the parameter values
used to represent the disease-spread process at the farm
level and also the characteristics of the population at risk
(17). Originally built in New Zealand (InterSpread, 16), it
has since been used in various countries, including
Switzerland, the Netherlands and the UK. Most
memorably, InterSpread was one of several models applied
during the 2001 FMD epidemic in the UK (12).
InterSpread was originally built as a decision-support tool
to provide epidemic predictions when facing outbreaks of
FMD. This type of model therefore requires a lot of data in
order to produce reliable outcomes.
The objective of the IS+ project for the modellers was
to determine if the model could be used to re-create the
2002 FMD epidemic in the Republic of Korea.
The epidemiologists in Korea also wanted to use the model
to determine what the potential impact (in terms of the
predicted number of infected farms) would have been, had
the control measures been implemented either earlier
or later during the epidemic. They were also interested in
knowing what would have been the effect of using
3-km or 5-km vaccination rings in combination with either
a standard, limited depopulation strategy or a more
extensive depopulation strategy. While all the modelling
work was done from New Zealand, local epidemiologists
worked on the parameters for the population at risk,
movement frequencies, distances of movements and
meteorological data and they created a baseline scenario to
which to apply control measures. Following an initial
meeting of three to four days, all remaining collaboration
was done through e-mail communications. The conceptual
model provided by IS+ was not altered for this project;
rather, the primary challenge was producing population
data that could be used in the IS+ framework. InterSpread
Plus was successfully applied in the Republic of Korea and
this was facilitated by its highly flexible design. Its
successful application in another country also confirmed
the model’s operational validity, as it appeared to be able to
represent a real epidemic.
The third example is of the technology and capability
transfer of modelling tools and expertise. Canada’s
Department of Foreign Affairs and International Trade
funded a technology transfer project as a capacity-building
effort for FMD preparedness and response in South
America. The model used was NAADSM (5), developed in
North America, where there is limited experience in
responding to highly contagious diseases, and limited
information about the location of livestock holdings and
the contact structure (in terms of the movements of
animals and indirect contacts such as people, equipment
and other fomites) among holdings. The principal
objective for building NAADSM was to support
contingency planning in North America for foreign animal
diseases, such as FMD, highly pathogenic avian influenza
and classical swine fever. As a result, it was built as a
software framework that allows the user to develop
simulation models for various diseases and in different
populations. It has a user-friendly interface and it is
available on the internet in English and Spanish. A detailed
user’s guide is available, and training opportunities are
routinely provided. The source code is available for
download and a description of the model has also been
published (5).
From the model builders’ perspective, the objectives of the
project were to:
– obtain experience in understanding the epidemiology of
FMD in South America
– test NAADSM’s ability to represent the demographics and
the contact structure among livestock holdings in the
region
– further assess the conceptual validity of the model
through expert review
– obtain data to support improvements to the model.
The main partner in the project was the Pan American
Foot-and-Mouth Disease Center (PANAFTOSA) and its
objectives were to:
– provide a tool for contingency planning for countries of
South America
– provide training on epidemiological modelling to
epidemiologists in the region and create a pool of experts
that could provide ongoing training
Rev. sci. tech. Off. int. Epiz., 30 (2)
584

Page 5

– increase the level of collaboration with North American
epidemiologists
– potentially tailor the model to the needs of the region in
order to have a sense of ownership of the tool.
The project was initiated with a meeting of subject-matter
experts in March 2008 in Brazil (2). The objectives of the
meeting were to:
– train local epidemiologists with expertise in FMD to use
NAADSM
– gain an understanding of the characteristics of the
livestock population and its contacts, and of the
epidemiology and control of FMD in the region
– identify potential modifications necessary to simulate
FMD epidemics in the region
– have NAADSM reviewed in detail by FMD experts.
The model was provided with a Spanish user-interface and
various issues were discussed, with five code modifications
and enhancements recommended for NAADSM to be used
in South America and to properly represent the different
epidemiological conditions of FMD in the region. The
main issue discussed was the importance of appropriately
representing the different production systems found in
South America in terms of their contact parameters, their
husbandry and management attributes. These factors are
known to be very important in the transmission of FMD in
the region.
The two-year project led to the completion of two pilot
studies using NAADSM, one in Chile (15) and one in
Brazil, each of which evaluated FMD spread in a specific
region. As a result of this project, the model was improved,
or is in the process of being improved, so that it will be
able to represent within-herd spread of infection and take
account of variable vaccine efficacy and coverage. This
collaboration has resulted in an increase in confidence in
the capacity of NAADSM to model FMD and it is currently
being used as a preparedness tool in different countries in
South America, with ongoing collaborations with the
NAADSM development team.
Requirements for adapting
models to other countries
Several factors are important to consider when deciding
whether to adapt a model and which model to adapt. The
recommendations provided in this paper are based on the
authors’ experience of building and using models.
First and foremost, the model-building process must be
transparent in order for the model to be easily transferred
or used by local experts. This means that all assumptions
must be clearly documented and refined as the model is
being adapted. These assumptions must be available for
review by the experts and decision-makers who use the
outputs of the modelling studies. The models to be
adapted must have gone through a series of evaluations in
order to be verified and, as much as is possible, validated
for the purpose for which they were built. The reader is
referred to Reeves et al. in the current issue for a discussion
of model verification and validation (14).
For a model to be easy to use in another country it must
have an easy-to-read and easy-to-understand model
description that can be used in training initiatives in the
other country and as a guide for future use. Ideally, the
model should have an easy-to-use interface in which users
may change parameters to represent different outbreaks or
diseases or populations. Having such an interface will
enable technology transfer projects which are interesting in
the long term. Users should be aware, however, that an
easy-to-use interface may convey a false sense of ease or
simplicity regarding the modelling process: the presence of
a simple user interface is not a substitute for the
development of a thorough understanding of the
assumptions of the conceptual model. Again, having a
proper and detailed user’s guide and supporting
documents can lead to very successful application.
The more flexible the model, the easier it will be to adapt
in other countries and to be part of technology transfer
projects. First, it must be able to represent the differences
in livestock demographics, management practices,
livestock movements and other contacts that are important
for infection spread. Spatially explicit models, for example,
can take into account different spatial distributions and
densities of livestock populations. Network-based models
can take into account the topology of livestock movement
networks in order to represent this level of heterogeneity, if
it is judged to be important. The model must also be able
to deal with different strains of the infectious disease agent
under study, if applicable, and also provide different
starting conditions for the start of outbreaks. This ensures
greater flexibility for the application of the model. The
model must also be able to take account of differences in
susceptibility, infectivity and clinical signs by species, if
applicable. Finally, it must include various possibilities for
control measures, and combinations of these measures,
that are appropriate for the country or region in which the
model will be used and for the purpose for which the
model was developed.
Ideally, the model should be easily accessible to the
receiving country, preferably offered in the country’s official
language and available on the internet for easy access.
Training must be provided to ensure maximum benefit for
local experts and users, and translation services are vital in
Rev. sci. tech. Off. int. Epiz., 30 (2)
585