Risk Analysis, Vol. 27, No. 5, 2007 DOI: 10.1111/j.1539-6924.2007.00946.

Special Issue: The Future of BSE Assessments

Analyzing BSE Transmission to Quantify Regional Risk

Aline A. de Koeijer∗

As a result of consumer fears and political concerns related to BSE as a risk to human health, a
need has arisen recently for more sensitive methods to detect BSE and more accurate methods
to determine BSE incidence. As a part of the development of such methods, it is important to
be able to identify groups of animals with above-average BSE risk. One of the well-known risk
factors for BSE is age, as very young animals do not develop the disease, and very old animals
are less likely to develop the disease. Here, we analyze which factors have a strong influence
on the age distribution of BSE in a population. Building on that, we develop a simple set of
calculation rules for classifying the BSE risk in a given cattle population. Required inputs are
data on imports and on the BSE control measures in place over the last 10 or 20 years.

KEY WORDS: BSE; discrete time; predictive model; risk

1. INTRODUCTION Especially in countries where few or no BSE cases

have been reported so far, active surveillance (testing
The presence of BSE in the European cattle pop-
of animals at slaughter) is far more effective than pas-
ulation led to decreased beef consumption in the late
sive surveillance (mandatory reporting of clinical sus-
1990s. Clearly, consumers worry about the possibility
pects) to assess the incidence. To establish and quan-
that through consumption of material from infected
tify a low BSE incidence, large numbers of animals
cattle they may acquire an infection leading to new
need to be tested. The number of tests could possi-
variant Creutzfeld Jacob disease, a fatal and incur-
bly be reduced if one could target specific risk groups
able disease in humans. For proper assessment of the
in the cattle population and, thus, design an optimal
human risk, it is important to estimate the regional
targeted surveillance program.When a proper risk as-
incidence of BSE in cattle. This is not straightforward
sessment indicates negligible risk, this can be a good
because the infection can spread among cattle without
reason to refrain from an intensive surveillance pro-
being noticed for several years: The disease develops
gram and apply a limited but targeted surveillance
very slowly, and early clinical symptoms are various
program.
and difficult to diagnose, especially by unexperienced
Doherr et al. (2001) showed that, in Switzerland,
people. This difficulty to diagnose can lead to major
BSE is found with a higher probability in fallen stock
underreporting of the disease, as was shown to have
(five times higher) and emergency slaughter (four
happened in the United Kingdom and Switzerland in
times higher), than it is in the normal slaughter line.
the beginning of their respective epidemics (Heim &
Targeted sampling from groups with higher risk can
Wilesmith, 2000). It has by now become clear that sev-
increase the sensitivity of a surveillance system. Thus,
eral European countries had BSE circulating in their
surveillance could be improved by concentrating on
cattle population while still claiming to be free from
fallen stock and emergency slaughtered animals. Rel-
the disease.
evant other information can be derived from routine
testing of all slaughter animals over 30 months, which
∗ ASG, WUR, Infectious Diseases, Lelystad, The Netherlands; most EU countries started in 2001. These results con-
aline.dekoeijer@wur.nl. firm the higher BSE incidence in fallen stock and
1095 0272-4332/07/0100-1095$22.00/1 
C 2007 Society for Risk Analysis
1096
emergency slaughter animals. However, large differ-
ences between countries appear in the relative risk
of these categories. Therefore, a careful approach
and good data are needed to extrapolate to other
countries.
General patterns in the age distribution of BSE
cases, as resulting from the local cattle population
structure and from the local measures in BSE con-
trol, are analyzed in de Koeijer et al. (2002). Here,
we will show in more detail how to develop a model
to analyze such patterns and the effect of changing
BSE control conditions over time. Using the available
information on development of the infection, taking
local conditions into account, and including informa-
tion on import of risk animals and infectious mate-
rial in the past, it is possible to obtain a well-founded
assessment of the infection risk in a given country.
However, to do so, several difficult steps in modeling
and calculation need to be made. In this article, we in-
troduce a deterministic BSE transmission model and
transform it into a rather simple calculation system,
based on discrete time steps. This calculation system
can be used to make a quantitative regional BSE risk
assessment, that requires rather simple and straight-
forward input information. It also offers the option to
analyze the age distribution of detected BSE in a cat-
tle population and subsequently evaluate the quality
of historical information on import or control condi-
tions. The results can be used to support discussions
on trade safety, and can be applied to further target
the BSE surveillance to a smaller group of cattle with
relatively high risk of BSE.
2. AGE-STRUCTURED MODEL
According to basic theory, the number of new in-
fections at some point in time depends on the infection
pressure at that moment in time. The number of in-
dividuals suffering from the disease depends on the
number of infected individuals some time ago (i.e.,
the incubation period ago); the precise dependence
has to take account of processes like survival of the
infected individual; see Diekmann and Heesterbeek
(2000). For very slowly developing diseases like BSE,
the delay introduced by the incubation period has a
major impact on the incidence of the infection in spe-
cific age groups. This is especially clear for BSE, where
most infections start at a very young age.
In de Koeijer et al. (2002), it is shown that the
age distribution of BSE cases is influenced by life his-
tory parameters of the population; for instance, a high
culling rate of cattle leads to a lower average age of
de Koeijer
BSE cases. The control history also has a high impact;
a region with little BSE control will find BSE cases
with a lower age (on average) than a region with good
BSE control measures. This is because with little con-
trol we expect the numbers of infections to grow over
time. The BSE cases with a young age got infected
at a later stage than the older ones, and are there-
fore born in a cohort with higher infection levels and
therefore more cases. Thus, for a growing epidemic
the cases are younger than in a declining epidemic.
This all follows from straightforward age-structured
modeling of the dynamics of the infection. However,
a few important features are ignored in this publi-
cation: (1) the fact that conditions can change over
time—most EU countries have introduced many new
control measures over the last 15 years, (2) the lack of
suitable input data, (3) the fact that cattle can become
infected at an advanced age (Arnold & Wilesmith,
2004). Most of these features can be incorporated in
a more careful mathematical formulation. We aim at
a model that describes the number of BSE cases over
time and age, based on parameters concerning the
transmission of BSE among cattle, survival of cattle,
and impact of BSE control measures.
We model the infection deterministically and as-
sume that the fraction of susceptibles in the popu-
lation remains close to 1. We define a BSE case as
an animal that would give a positive result if tested
for prions, and denote the number of BSE cases at
time t by c(t). New infections induced at time t will be
referred to as n(t), the infection cohort of time t. As-
sume a simple model with a fixed incubation period
τ ∗ , where the animals will die as a result of the disease
after the incubation period. In this case, the normal
survival of cattle (not influenced by the disease), de-
noted by F(α), determines which part of this infection
cohort will live sufficiently long to become cases later
on (for more details, see Diekmann & Heesterbeek,
2000), so:
c(t) = n(t − τ ∗ )F(τ ∗ ). (1)
Animals in early stages of BSE infection hardly
contribute to the transmission of BSE, so for prac-
tical purposes we neglect their contribution and as-
sume that the infectiousness is concentrated in a very
narrow time window centered around the moment
when disease symptoms show. Therefore, n(t) is pro-
portional to c(t), say n(t) = θ c(t). Combining this with
Equation (1) we obtain:
c(t + τ ∗ ) = R0 c(t), (2)
Analyzing BSE Transmission to Quantify Regional Risk
where R0 is the basic reproduction ratio of the infec-
tion, given by R0 = θ F(τ ∗ ). In the context of BSE, we
find that θ is actually time dependent, since behavior
and control measures are changing in the course of
time. Here, it is also relevant to realize that in gen-
eral θ is mostly influenced by the major transmission
route, i.e., transmission via cattle derived meat and
bone meal (MBM) in cattle feed. Other factors also
influence R0 but are assumed to be small compared
to the above even when basic control measures are
implemented. This is based on the experience of the
United Kingdom, where additional, extreme control
measures on the feed route since 1997 led to a further
substantial reduction of R0 , bringing it close to (but
not equal to) zero. When R0 is constant, the model in
Equation (2) is straightforward. When R0 varies over
time, strictly speaking we cannot interpret R0 (t) as
a reproduction ratio since the very concept does not
make sense when environmental conditions change
with time while generations overlap. In this article,
we will make the more specific assumption that R(t)
denotes a quantity that relates the number of cases
at time t to the number of cases in animals that got
infected at time t and will become cases at time t + τ ∗ .
More specifically this means that:
c(t + τ ∗ ) = R(t)c(t). (3)
The BSE case data from the United Kingdom
(Wilesmith et al., 1998) indicate that infection mostly
occurs at very young age but cases display a large
variation in the incubation period. Therefore, we ex-
tend the basic model into a very simple age-structured
model by adding a variable incubation period while
assuming that BSE infections start at the birth of an
animal. As a consequence, the length of the incuba-
tion period coincides exactly with the age of the case.
Let g(τ ) denote the probability density function of the
variable incubation period τ . Note that the distribu-
tion of the incubation period of cases detected in the
cattle population will generally differ from g(τ ), due
to a lower probability for animals with a long incuba-
tion period to survive until the onset of the disease.
We will now distinguish cases according to the vari-
ables time and age and define the relation between
cases in the past and future by:
 ∞  ∞
c(t + σ, σ ) dσ = R(t) c(t, τ ) dτ. (4)
0 0
∞
Here,  0 c(t, τ ) dτ is the number of cases at time t,
∞
while 0 c(t + σ, σ ) dσ corresponds to all cases that
became infected at time t. Previously, we saw that
R0 consisted of a time-dependent factor θ(t) and sur-
1097
vival F(τ ∗ ). With a variable incubation period, the
survival influence
 ∞ is given by the average survival of
a BSE case 0 g(τ )F(τ ) dτ . When quantifying the
value R(t) from information on feeding patterns, pop-
ulation data, and BSE control, the moment that these
ingredients have their impact is important. For θ (t)
this is simple; the effect will be timed around the mo-
ment of transmission, i.e., between death of a BSE
case and infection of a susceptible young cow through
MBM. When the survival function changes over time,
it changes during incubation. Therefore, such changes
in R(t) should be analyzed using the F(τ ) that is valid
for the birth cohort of time t. In general, the survival
function does not depend on time, but major policy
changes like the British OTMS (Over Thirty Months
Scheme: cattle over 30 months old are not accepted for
human consumption) may lead to substantial changes.
In such cases, specification of the correct survival func-
tion at a given time becomes important.
The number of BSE cases at time t is distributed
with respect to age. This distribution is influenced by
three factors: (1) the distribution of the BSE incuba-
tion period as described by g, (2) the survival proba-
bility F, which may vary between countries and over
time, and (3) the way new infections in the past varied
with time, which depends on the local epidemic his-
tory. Let k denote the normalization constant of this
∞
distribution (so k−1 = 0 g(ξ )F(ξ ) dξ ). We can now
formulate the number of cases in a cohort by:
 ∞
c(t + τ, τ ) = k g(τ )F(τ )R(t) c(t, ξ ) dξ, (5)
0
and find that the age distribution of cases at time t,
denoted by h(t, a), is given by:
c(t, a)
h(t, a) =  ∞
c(t, ζ ) dζ
0
 ∞
c(t − a, ξ ) dξ
= k g(a)F(a)R(t − a) 0 . (6)
∞
c(t, ζ ) dζ
0
When R(t) is constant over time (R(t) = R0 ), a stable
age distribution will be reached (see, e.g., Diekmann,
1999), where the total number of cases grows (or de-
clines) exponentially over time, with rate r:
 
∞ ∞
c(t0 + t, ξ ) dξ = e rt
c(t0 , ζ ) dζ. (7)
0 0
1098
For the method to calculate r, we refer to de Koeijer
et al. (2004). Substituting Equation (7) in Equation
(6) we find that h is independent of time and is given
by:
h(t, a) = k g(a)F(a)R0 e−ra .
Note that the age distribution of the BSE cases
displays the growth rate of the infection. An epidemic
in a stage of fast growth will display a lower average
age of the BSE cases than an epidemic in a stage of
slow or negative growth. Real data and predictions
derived from these can be found in de Koeijer et al.
(2002). Obviously, information on the age distribution
of the standing cattle population is needed to analyze
the growth rate of the epidemic from the age distribu-
tion of the cases. Such results can be applied to moni-
tor the efficacy of BSE control measures in a country
that recently started testing and has since detected
enough BSE cases to determine the age distribution.
An abrupt change in R0 at time t∗ induces an
abrupt change in the growth rate. Let r1 denote the
growth rate of the epidemic before change. Then, soon
after the change, the age distribution of cases will
show a running wave over time, and slowly settle into
the new stable age distribution. As before, the age dis-
tribution of the older cases displays the growth rate
of the infection before change:
h(t ∗ + t, a) = k g(a)F(a)e−r1 a ∀a > t. (8)
For younger cases, the age distribution can be calcu-
lated, but it will not immediately settle into the new
stable distribution because the infection pressure is
changed abruptly due to the new R(t), but the num-
ber of cases after change will still grow with the old
growth rate for awhile because they are already in-
cubating. It may take many years before the new age
distribution will be visible in the case data. When R(t)
is constantly changing, the model (Formula (6) has to
be applied to predict or analyze the age distributions
that will appear in the data.
It has become clear that older cattle can also be-
come infected with BSE, although at a much lower
rate than calves (Arnold & Wilesmith, 2004). Neglect-
ing the susceptibility of older cattle may affect the
answer to some specific questions that involve the in-
fection age. Therefore, we also include an extended
version of the model that uses a variable age at in-
fection. The technical details are explained in the Ap-
pendix. This model can be applied when the impact of
the exposure of adult cattle appears to be important
for the question addressed.
de Koeijer
3. PRACTICAL VERSIONS OF
THESE MODELS
Users of this model need to specify all the rele-
vant functions in quantitative terms. To facilitate the
use of the model by people with little background in
mathematics, the specifications and the implementa-
tion should be as straightforward as possible. Given
the sort of information that is available in most Eu-
ropean countries, the functions F(a), f (a), and g(τ )
may best be given as step functions with one-year
steps over time and age. The case data are often pre-
sented in statistics clustered by year of detection and
by age group. Thus it appears logical to use the fact
that time is usually applied in a discrete version of one-
year steps. Applying a model with discrete time steps
of a year has the advantage that seasonality patterns
in the data disappear. These can therefore be fully
neglected. It also allows us to develop the model into
a form that can be implemented into a spreadsheet.
This makes it more easily accessible to a wider audi-
ence. In the Appendix, we explain how we developed
a discrete time version from the above model.
Several versions of a case cohort model have been
developed and described in this article. The common
issue is that they all relate incidence and infection
pressure in the past to incidence and infection pres-
sure in the future. The continuous time models (Equa-
tion (5)) give exact answers, given the assumptions
on BSE and cattle behavior. However, it is very hard
to quantify the necessary functions accurately, unless
extra assumptions are included on the general shape
of these functions. The semi-discrete version of these
models overcome these practical issues. They require
a limited amount of input, which links better to the
type of information that can generally be obtained.
The method typically fits the sort of data that can
be found in statistics describing a cattle population
and its management. But most importantly, the semi-
discrete model allows for easy use in a spreadsheet to
calculate the behavior of a regional cattle BSE situa-
tion. Therefore, little is required in terms of computer
software to do the analysis, which will be very valu-
able if there are no mathematicians involved in the
risk assessment study.
This article focuses on deriving an easy-to-apply
method for quantitative BSE risk assessment, with a
secondary aim in predicting the age distribution of
BSE, which can be used to target age-specific surveil-
lance. However, once available, such a model can
be applied in many ways. Three typical ways to ap-
ply the model are explained here. One way uses the
Analyzing BSE Transmission to Quantify Regional Risk
model and case data over a long period of time to
estimate the previous infection pressure, and from
that calculate the historical reproduction ratio in the
country. Obviously, this method can only be applied
to countries with a sufficiently large number of BSE
cases over at least a decade and a good record on the
changes in disease notification. The model takes ac-
count of the known number of cases, but can only be
related to the real number of cases when a good esti-
mation of disease notification can be made, including
its development over time. This method of analysis is
very suitable for countries such as the United King-
dom, Switzerland, and Portugal. We will not address
this method any further because similar work has been
done with more sophisticated continuous time mod-
els, and is explained with much more detail by Don-
nelly and Ferguson (2000). This method can be used
to analyze the history of the BSE epidemic in a coun-
try and, within limitations, it also allows for forward
extrapolation to predict future development of the
epidemic. However, for those scientists who would
like to do such an analysis themselves, our method is
more accessible and easier to apply than those more
detailed models.
The other two ways to apply the model are fit
for analyzing BSE incidence in countries with few
BSE cases. In such countries, a case-data-based model
cannot be applied, and therefore other available data
sources must be analyzed to assess the local incidence.
The semi-discrete model can be applied to such coun-
tries for two main purposes: prediction of undetected
and future incidence and quantification of parame-
ters.
The model focuses on making a quantitative risk
assessment of the BSE incidence, but at the same time
it supplies information that can be used to target the
disease surveillance more efficiently and it offers fu-
ture predictions on incidence of the infection. The ef-
fectivity of present and intended control measures can
also be calculated and predicted. This may then be
used to support a cost-benefit analysis.
Quantification of parameters can be used to make
an analysis of the efficacy of the BSE control over
the last few decades. Recent results of active BSE
surveillance can be used to calculate the BSE inci-
dence earlier on. Thus, the efficacy of control mea-
sures can be quantified. Control measures can include
a feed ban, import ban on live cattle, etc. This method
can also be used to quantify disease specific charac-
teristics, like the distribution of the incubation pe-
riod, by applying it to a large data set of BSE cases.
(U.K. case data would be the best.) The extended
model is used to quantify f (a) by Wang and de Koeijer
1099
(in prep.). A disadvantage of the extended model is
that it is not easy to apply for assessing a past of
which we have insufficient information since the un-
certainty on age at infection leads to high uncertainty
in the results. For assessing the future, however, the
model works fine and is hardly more complicated to
apply. We found that for future extrapolation, pre-
dictions show little difference with the first model
that assumes infection at birth. Therefore, the sim-
plest version of the model will be preferred for most
questions.
A quantitative risk assessment for countries with
few or no BSE cases can best be built on an analysis
of the main risk factors and prevention of import and
transmission. The model we have just developed does
not take account of the effect of import or export
of infection into and from the population addressed.
However, without an initial import, very few countries
would have had a problem with BSE. Therefore, the
above model on transmission of BSE over time needs
to be combined with the aspects of the (continued)
import of risk material.
The EU (now EFSA) has developed a good
method to analyze the combination of these two as-
pects in its geographical BSE risk assessment, which
makes a qualitative analysis of BSE propagation (i.e.,
transmission), with results ranging from very unsta-
ble (R > 1) to a very stable system (little propaga-
tion, R < 1). Next, this is combined with the assessed
internal challenge (i.e., local incidence of infection)
and external challenge (i.e., import of risk material)
into the system each year. Obviously, due to the qual-
itative level of the assessment, the dynamics of the
interactions cannot be incorporated completely, but
for rather short-term analysis, the method works very
well.
We will explain how to extend the simple ver-
sion of the discrete time model to do just that, but
quantitatively. (Given the expected accuracy of the
input data, it seems that use of the extended model
with f (a) will not contribute much to the accuracy
of the result, but will complicate the assessment very
much.) We start by quantifying all the relevant func-
tions, g(τ ), F(a), and R(t), where the last may prove
to be quite a challenging task (the methodology to do
so is explained in de Koeijer et al., 2004). To quantify
g(τ ), we use the U.K. case data of a few cohorts as ex-
plained in de Koeijer et al. (2002) and correct for the
age-dependent survival of cattle in the United King-
dom, F(a) should be quantified based on the regional
statistics and when the extended model is chosen, we
need to quantify f (a), which has been done by Arnold
and Wilesmith (2004). They find that the probability
1100
for older cattle to get infected is about 10 times lower
than that of calves (0 to 11 months).
Subsequently, risk imports need to be analyzed
in terms of the expected number of cases they will
induce in the exposed cohort of cattle. Thus we derive
a starting distribution of BSE cases, due to the first
import of BSE. The expected incidence of BSE due to
imports must be added to the BSE incidence, resulting
from internal BSE transmission and we obtain the
following equation. Let I(y, z) denote the expected
number of BSE cases per year y, and per age group z,
directly resulting from import of live cattle and MBM
in previous years. Then we extend Equation (A.2) and
derive:  EU-GBR work. Furthermore, so far, this is the only
1  ∞
C(y, z) = k g(z)F(z) R(y − z) C(y − z, i)
2 suitable for low- and zero-incidence countries. Be-
i=0
 cause of the easy-to-apply methodology, it may even
∞
+ R(y − z − 1) C(y − z − 1, i)
i=0
+ I(y, z).
To determine I(y, z), lots of factors need to be
taken into account. For instance, MBM imports in
year y will lead to cases in the birth cohort of year
y, and they will distribute over I(y + z, z) for ran-
dom z. Imports of live cattle can only lead to direct
cases in those imported animals. The age of the im-
ported animals and the purpose of import will affect
the distribution of cases over I(y, z). Imports often
consist of calves for slaughter, which will lead to a
very small import risk, but cattle imported for breed-
ing are mostly one or two years old and will have a
longer life expectancy than local stock. Finally, to de-
termine I(y, z) one will need to assess the incidence
of infection in the exporting country, and bearing the
previous in mind, then determine the age distribution
for cases due to these imports. We refer to the EU-
GBR methodology (see website EFSA) for further
details on the many factors that can influence the age
distribution of the cases, as it is somewhat outside the
scope of this article.
This method for risk analysis has been applied in
a quantitative risk assessment of BSE in Norway. A
scientific publication of that work was recently fin-
ished (Hogasen et al., 2007). It was also applied to the
Netherlands as described by de Koeijer et al. (2006),
and publication of these results is expected soon. We
find that this model offers an easy-to-apply calcula-
tion, and find that most of the actual work is spent
on assessing the input parameters of the model, but
given their straightforward definition, they can be de-
termined within the limits set by the available data.
We conclude that a quantitative BSE risk assessment
de Koeijer
of a country is a major job to fulfill, and note that the
reports of the EU GBR are a good starting point for
each country that would be interested in applying this
method.
4. DISCUSSION AND CONCLUSIONS
We have derived a method of calculation that can
be applied to assess regional BSE risk in quantitative
terms. Given that it follows the general lines of the
European Geographical BSE Risk (EU-GBR) As-
sessment, it may be of interest to the EU and to those
countries that wish to expand on the results of the
systematic approach to BSE risk assessment that is
be of interest for countries with high BSE incidence,
as a tool to easily explore the effect of specific control
measures and the effect of changing the surveillance
program.
The model has been applied to the Netherlands
and Norway (Hogasen, 2007), and it becomes clear
that for optimal accuracy, one in fact needs to assess
all countries where the imports originate. Thus the
results of the assessment will always depend on the
quality of the assessed BSE risk in other countries.
Fortunately, the assessment of the exporting countries
need not always be a huge task, especially when com-
promises on accuracy level are acceptable. For exam-
ple, for assessing the Netherlands, this import analysis
is a rather small task because the majority of imports
originate from the United Kingdom. In the last 10
years, however, imports from Germany dominate the
statistics. Given that the German BSE incidence is
presently of the same order of magnitude, it might at
first seem that the impact of these imports is neutral.
However, whereas BSE control in the Netherlands
leans a bit more on the quality of the rendering sys-
tem, the German BSE control until 2000 was mainly
based on its feed ban. Therefore, MBM produced in
Germany used to carry a larger risk than Dutch MBM
in the period 1995 to 2001.Via cross-contamination,
this German MBM could lead to a higher risk of new
BSE infections in the Netherlands than it would have
caused in Germany. Thus, even imports from coun-
tries with similar BSE incidence may have a negative
effect on the BSE risk. Clearly, the recent total ban
on the use of MBM for all animal husbandry feeds
has ended the effect mentioned before. For Norway,
we found that the very limited imports came from
other Scandinavian countries, which all have low to
zero BSE incidence. The risk of BSE being imported
Analyzing BSE Transmission to Quantify Regional Risk 1101

in Norway therefore depends very strongly upon the Deriving the age distribution of the cases is some-
BSE incidence in those countries. For a good assess- what more complicated. First, new infections are dis-
ment of the BSE risk in any European country, it tributed over the age groups according to f (α)F(α)
would be best to assess most other EU countries too. and, similar to before, the incubation period is dis-
We suggest that a joint European study to quantify tributed according to g(τ ) F(α + τ)
F(α)
. Thus, the cases at
BSE incidence and risk will contribute to the present time t that got infected at age α and had incubation
discussion on the safety of cattle and cattle-derived period τ are given by:
products.
Finally, we note that a human BSE exposure as- F(α + τ )
c̃(t, α, τ ) = κ f (α)F(α)g(τ ) R(t − τ )
sessment must always be based upon a proper BSE F(α)
risk assessment of the cattle population. Therefore, we  ∞ ∞
expect that linking of this model with existing human × c̃(t − τ, ζ, ξ ) dζ dξ,
0 0
exposure models that use an estimated BSE incidence
in the slaughter population can improve the results of where κ is now the normalization constant for
those models in making them more accurate. the
 ∞ distribution of the cases over α and τ (κ −1 =
∞
0 0 f (α)g(τ )F(α + τ ) dα dτ ).
ACKNOWLEDGMENTS In the case data, the only available characteris-
This work was partly funded by EU FAIR 98- tics are the time and the age, whereas the age at
7021 and Neuroprion EU CT 2004-506579. The fund- infection and the incubation period will not be de-
ing source waived any right to review or approve the tectable. Therefore,
∞ we denote
 ∞cases by c(t, a), which,
∞
manuscript. using that 0 c(t · a) da = 0 0 c̃(t, σ, ξ ) dσ dξ , can
also be written as:
APPENDIX: AN EXTENDED MODEL THAT  a
TAKES ACCOUNT OF THE AGE c(t, a) = κF(a) f (a − ζ )g(ζ )R(t − ζ )
AT INFECTION 0
 ∞
This age at infection is indeed mainly at very × c(t − ζ, ξ ) dξ dζ. (A.1)
young age, but can occasionally be much older. Un- 0
der this extended model, the incubation period is no
The age distribution of cases is then given by:
longer the same as the age of a case. Formula (5) re-
mains valid, but it can be extended by including the h(t, a)
age of the animals separate from the incubation pe-  a  ∞
riod. κF(a) f (a − ζ )g(ζ )R(t − ζ ) c(t − ζ, ξ ) dξ dζ
Let f (α) denote the probability density function = 0
 ∞ 0
.
for the age at infection by BSE in a test group of cattle c(t, ξ ) dξ
with a uniform age distribution. In a production pop- 0
ulation of cattle, the probability density function of When R is constant over time (R0 ), the system
the age at infection is then given (modulo normaliza- reaches a stable age distribution given by:
tion) by f (α)F(α) because the lower survival of older
cattle influences the effective age at infection. We as- c(t, a)
h(t, a) =  ∞
sume that g(τ ) is independent of the age at which an
c(t, ξ ) dξ
animal is infected. 0
Let c̃(t, α, τ ) denote the BSE cases at time t that  a
were infected at age α and had incubation period τ . = κF(a) f (a − ζ )g(ζ )Re−r ζ dζ,
As before (see Formula (4)), R(t) is defined by the 0

balance of all the cases in an infection cohort and all which is independent of t.
the cases at the moment that cohort got infected. (An
infection cohort is a group of animals that got infected
A.1. Discretizing the Model
at the same time.)
 ∞ ∞ Let C(y, z) denote the number of detectable cases
c̃(t + τ, α, τ ) dα dτ in the full calendar year y and having an age of z, when
0 0
 age is expressed in full years. Below, we use δ to denote
∞ ∞
the part of the year that has passed since the first of
= R(t) c̃(t, ζ, ξ ) dζ dξ.
0 0 January and we use ε to denote the time elapsed since
1102 de Koeijer

the animal’s last birthday. So 0 ≤ {δ, ε} < 1 and {y, z} ∈ cases develop as a step function over time. Then we
N. The step function assumption amounts to F(z) = use Equation (A.1) and argue as follows:
F(z + ε), g(z) = g(z + ε), and R(y + ε) = R(y).
Given that  y+1 z+1  a
 y+1  z+1 C(y, z) = κF(a) f (a − ζ )g(ζ )R(t − ζ )
y z 0
C(y, z) = c(t, a) da dt,  ∞
y z
× c(t − ζ, ξ ) dξ dζ dt da,
using Formula (5) we find that: 0
 y+1  z+1 from which we derive,
C(y, z) = k g(a)F(a)R(t − a)
y z
 C(y, z)
∞
 1 1
× c(t − a, τ ) dτ da dt. z+ε
0 = κF(z) f (z + ε − ζ )g(ζ )
0 0 0
Now replace a and t by, respectively, z plus ε and y plus  ∞
δ and separate into parts where ε is bigger or smaller × R(y + δ − ζ ) c(y + δ − ζ, ξ ) dξ dζ dδ dε.
than δ. When we remove those functions from the 0
integral, that are not dependent on τ , ε, and δ, and Now we can split the right-hand side of the equa-
analyze the remaining integral further, we find: tion for various values of ζ , to obtain functions that
C(y, z) = k g(z)F(z)R(y − z) can be rewritten in a discrete version. To do so, we let
 1 δ ∞ ζ = x + φ, where x ∈ N and 0 ≤ φ < 1 and derive:
× c(y + δ − z − ε, τ ) dτ dε dδ
0 0 0

z
C(y, z) = κF(z) g(x) f (z − x)
+ k g(z)F(z)R(y − z − 1) x=0
 1 1 ∞  1  1  ε
× c(y + δ − z − ε, τ ) dτ dε dδ. × R(y + δ − x − φ)
0 δ 0 0 0 0
 ∞
The
∞ latter integrals ∞cannot be written in terms of × c(y + δ − x − φ, ξ ) dξ dφ dδ dε
i=0 C(y, i) and i=0 C(y − 1, i), unless we assume 0
that c(t, τ ) also behaves as a step function over t,

z−1
i.e., all cases found in year y and with age z are + κF(z) g(x) f (z − x − 1)
detected with equal probability throughout that year. x=0
In that case C(y, z) = c(y + δ, z + ε) and we see that   
1 ∞ 1 ∞ 1 1 1
0 (1 − δ)
1
c(y + δ, τ ) dτ dδ = 0 (1 − δ) i=0 C(y, × R(y + δ − x − φ)
0
∞ 1 ∞ 0 0 ε
i) dδ = 0 (1 − δ) dδ i=0 C(y, i) = 2 i=0 C(y, i).  ∞
1δ 11
Also using 0 0 1dε dδ = 0 δ 1dε dδ = 12 , we find × c(y + δ − x − φ, ξ ) dξ dφ dδ dε.
that: 0
 ∞ We now separate the case where δ < ε from the case
1
C(y, z) = k g(z)F(z) R(y − z) C(y − z, i) where δ > ε, and simplify R and c to their semi-
2 i=0 discrete versions. Then we can calculate the remaining
∞  integrals over δ, ε, and φ and find all six of them to be
+ R(y − z − 1) C(y − z − 1, i) . equal to 16 , so we arrive at:
i=0

(A.2) 1 z
C(y, z) = κF(z) g(x) f (z − x)R(y − x)
Thus, the cases in year y and having age z are partly 3 x=0
born in y − z, and partly born in year y − z − 1.

∞
Obviously, in some cases a discrete version of the × C(y − x, i)
extended model can be applied best, so we will also i=0
transform the model with a variable age at infection
1 z
into a semi-discrete model. We use all definitions as + κF(z) g(x) f (z − x)R(y − x − 1)
given above and already include the assumption that 6 x=0
Analyzing BSE Transmission to Quantify Regional Risk 1103


∞ de Koeijer, A., Heesterbeek, H., Schreuder, B., Oberthur, R., Wile-
× C(y − x − 1, i) smith, J., & de Jong, C. M. (2004). Quantifying BSE control
i=0 by calculating the basic reproduction ratio R0 for the infec-
tion among cattle. Journal of Mathematical Biology, 48, 1–
1 
z−1 22.
+ κF(z) g(x) f (z − x − 1)R(y − x) de Koeijer, A., Schreuder, B., & Bouma, A. (2002). Fac-
6 x=0 tors that influence the age distribution of BSE cases:
Potentials for targeting in surveillance. LPS, 76, 223–

∞ 233.
× C(y − x, i) de Koeijer, A. A., Heres, L., & van Roermund, H. (2006). Anal-
i=0 ysis of BSE prevalence in the Netherlands: Present situation
and effect of changes in surveillance and control. ASG report,
1 
z−1 Infectious Diseases; Animals Sciences Group, Wageningen
+ κF(z) g(x) f (z − x − 1)R(y − x − 1) UR.
3 x=0 Diekmann, O. (1999). Modeling and analysing physiologically
structured populations. In V. Capasso & O. Diekmann (Eds.),

∞ Mathematics Inspired by Biology (pp. 1–37). Berlin: Springer
× C(y − x − 1, i). LN.
i=0 Diekmann, O., & Heesterbeek, H. (2002). Mathematical
Epidemiology of Infectious Diseases: Model Building,
Thus the semi-discrete version of the extended Analysis and Interpretation. Chichester: John Wiley &
model is derived. Obviously, it is far more complicated Sons.
Doherr, M. G., Heim, D., Fatzer, R., Cohen, C. H., Vandevelde,
than the earlier model because the sum over all pos- M., & Zurbriggen, A. (2001). Targeted screening of high-risk
sible ages at infection continuously complicates the cattle populations for BSE to augment mandatory reporting
model. However, in applications this system proves of clinical suspects. Preventive Veterinary Medicine, 51(1/2), 3–
16.
to be practical. It can be fully calculated from a given Donnelly, C. A., & Ferguson, N. M. (2000). Statistical Aspects of
starting situation, and calculations forward in time are BSE and vCJD. Models for Epidemics. Monographs on Statis-
no problem. For calculations backward in time, this tics and Applied Probability 84. Boca Raton, FL: Chapman
and Hall CRC.
model is less suitable because of the far more compli- Heim, D., & Wilesmith, J. W. (2000). Surveillance of BSE. Archives
cated dependence of a new value of C on many others, of Virology, 16, 127–133.
earlier in time. Hogasen, H., & de Koeijer, A. A. (2007). Quantitative BSE risk
assessment for BSE in low or zero-prevalence countries: The
example of Norway. Risk Analysis, 27(5),.
Wang, C. F., & de Koeijer, A. A. (in prep.). Quantifying the age
REFERENCES dependent probability to get infected with BSE.
Wilesmith, J. W., Wells, G. A. H., Cranwell, M. P., &
Arnold, M. E., & Wilesmith, J. W. (2004). Estimation of the age- Ryan, J. B. M. (1998). Bovine spongiform encephalopa-
dependent risk of infection to BSE of dairy cattle in Great thy: Epidemiological studies. Veterinary Record, 123, 638–
Britain. PVM, 66, 35–47. 644.