md08 PDF

MD08 Modelling HIV and STIs
EPM302 Modelling and the Dynamics of Infectious Diseases
Home
Contents
Module Map
Glossary
Tools
Help
Section 1: Modelling sexually transmitted infections

page 1 of 37
next
OVERVIEW
To introduce models of the transmission and control of sexually transmitted infections
(STIs)
OBJECTIVES:
By the end of the session you should:
Understand the important characteristics of STIs for modelling, and how they differ
from the infections modelled so far
Be able to use compartmental models to explore the transmission dynamics and
control of short-duration curable STIs, such as gonorrhoea
Be able to use a compartmental model to explore the transmission dynamics of
HIV/AIDS
This session should take 4-5 hours to complete .

Home
Contents
Module Map
Glossary
Tools
Help
Section 2: Important characteristics of STIs

page 2 of 37
prev
next
STIs share a number of characteristics important for how they are

modelled:
Risk of infection depends on intimate contact and therefore the density-dependent
force of infection assumption is not appropriate (as we assumed for "respiratory"
infections)
There is great heterogeneity within and between populations in risk behaviour and
transmission
The risk of infection is affected by numbers of contacts - some people have many
contacts but most have relatively few
Those not sexually active are not at risk
Some STIs don't elicit an effective immune response in the host (us!)
It is difficult to collect reliable data on sexual behaviours
An STI in a subject/partner may act as a cofactor increasing risk of infection with a
different STI (eg on HIV transmission)
But there are also important differences between STIs:
Duration of infection
Stages of infection
Proportion of asymptomatic infections (particularly in women)
No recovery from some infections
Some STIs kill others do not
Sexual behaviours are also very important:
Partnership acquisition rates
Mixing patterns
Type of contraception used (eg hormone pills vs condoms)
Partnership concurrency ...

Home
Contents
Module Map
Glossary
Tools
Help
Section 3: The simplest STI model

page 3 of 37
prev
next
The simplest model structure used by Hethcote & Yorke is shown below. It is often used
for modelling non-immunising infections like N.gonorrhoea:
Figure 2. A simple model of STI transmission dynamics.
As can be seen, this structure assumes no incubation period so that a susceptible (S)
immediately becomes infectious when infected (I), and is immediately susceptible to reinfection upon recovery . As we learnt in session MD01 this model is known as an "SIS"
model.
Exercise:
Fill in the missing terms (provided in the drop down list) to complete the expressions
correctly for the rate of change for the number
,
. Each term can only be used
once.
First describe in words
dS(t)
=
dt
Choose...
Choose...
Choose...
Choose...
dl(t)
=
dt
Choose...
Choose...
Choose...
Choose...
Answer
Then describe the same model algebraically

dS(t)
=
dt
Choose...
Choose...
Choose...
Choose...
dI(t)
=
dt
Choose...
Choose...
Answer
Choose...
Choose...

Home
Contents
Module Map
Glossary
Tools
Help
3.1: Reformulating the force of infection equation for STIs

page 4 of 37
prev
next
Although data on contacts (e.g. number of new sexual partners per year) may not always
be reliable, in principle it is easier to measure than contacts for respiratory infections. So
STI modellers usually split the transmission parameter "" used for the modelling of
respiratory infections (see MD01 ), into a "behavioural" component: the effective rate of
partner change per year (c), and a "biological" component: the transmission probability per
partner p, ie:
cp
If we assume random mixing and "frequency dependence" (see MD01 ) the equation for
the force of infection (t) can be defined as the product of the effective partner change
rate, c, the transmission probability per partnership, p, and the prevalence of infection at
time t, I(t)/N:
(t) = c p I(t)
N
The R0 for this infection in this population is given by the expression: R0 = c pD where D is
the duration of infectiousness.

Home
Contents
Module Map
Glossary
Tools
Help
3.2: Simple model prediction

page 5 of 37
prev
next
However, if we assume plausible parameter values for the US at the time, i.e. an average
partner change rate, c, of two partners per year, a transmission probability per partnership,
p, of 0.75 2 , and a duration of infection with some treatment services in the population,
D, of two months (0.167 years), the model predicts the infections dies out:
Show
A Berkeley Madonna version of this model has been set up for you. Check for yourself that
the infection does indeed die out.
Click here
Click here
to open the Berkeley Madonna file with the flowchart.

to open the Berkeley Madonna file without the flowchart.
Q1.1 Why does this happen? (Think about the basic reproduction number).
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 4: Importance of heterogeneity in sexual activity

page 6 of 37
prev
next
The primary reason the simple model fails to show gonorrhoea spreading is that it does not
take into account the influence of heterogeneity in human sexual activity. Surveys show
that most people have relatively low numbers of partners, but some have much higher
numbers Schneeberger et al, 2004.
Figure 4: The distribution of reported annual number of partners for heterosexual and homosexual men and
women in Britain from the National Survey of Sexual Attitudes and Lifestyles, 2000. The plots show the relative
number of partners and excludes individuals who report zero partners.

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.1: Incorporating heterogeneity in sexual activity

page 7 of 38
prev
next
We can include heterogeneity in sexual activity in the model by splitting the population into
two groups, one with lower numbers of partners (subscript L), and one with higher
numbers (subscript H).
Figure 5. Model of gonorrhoea transmission incorporating heterogeneity in activity
What are the equations for rate of change of the number of susceptible and infectious
individuals per unit time in this model?
Model incorporating heterogeneity in sexual activity
Q1.2 Fill in the missing terms (provided in the drop down list) to complete the expressions
correctly for the rate of change for the number of susceptible and infected individuals
dS H (t), dS L(t), dlH (t), dlL(t)
dtdtdtdt
Each term can only be used once.
dS H (t)
dt
dS L(t)
dt
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
dlH (t)
dt
dlL(t)
dt
Answer
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.
Please
select
an item.

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.2: Incorporating heterogeneity in sexual activity

page 8 of 38
prev
next
Model incorporating heterogeneity in sexual activity (2)

Q1.3 Now put the verbal descriptions from the previous exercise into the correct
mathematical form using the terms provided in the drag and drop term banks below to
complete the expressions correctly for the rate of change for the number of susceptible
and infected individuals.
Reset
Reset
Reset
Reset
Feedback
And the equation for the force of infection on these two groups, L and H, is:
L (t) = c Lpp(t)
H (t) = c H pp(t)
where:
c H & c L are partner change rates per year for high and low activity population groups
p is the transmission probability per discordant partnership
p(t) is the probability a selected partner is infectious

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.3: Probability a partner will be infectious in an STI

model with heterogeneity in sexual activity
page 9 of 38
prev
next
In the model without heterogeneity in sexual activity the Force of infection was calculated
as:
(t) = c
I(t) = c p
p
N
The probability a partner is infectious, p, was the same as the overall prevalence.
Q1.4 Why was this true?
Answer
If there is heterogeneity in sexual activity the force of infection equation must now take into
account the numbers in each group, the partner change rate in each group, and how the
two groups mix with each other.
The simplest assumption is to assume proportionate mixing. This means that partners are
chosen in proportion to the numbers of partnerships "offered" by each of the two groups,
as opposed to choosing partners by (say) the number of people in each of the two groups.
Q1.5 NH & NL are the population sizes of high and low activity population groups,
respectively; c H & c L are the numbers of partners that each person in these groups has
each year, respectively. What is the probability, g H , of selecting at random a partnership
with someone from the high activity group?
Hint: Remember that the number of partnerships "offered" by the whole population will be:
c H NH + c L NL
Answer
A similar equation can be written down for the probability of selecting a partnership with
someone from the low activity group (g L) but as the partnership must be selected from one
of the two groups, we also know that g H + g L = 1 so g L =1- g H .

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.4: Force of infection for a STI model with

heterogeneity in sexual activity
page 10 of 38
prev
next
Combining these probabilities with the proportion of people in the two groups that are
infectious, I H (t)/N H =i H (t) and I L (t)/N L =i L (t), then the probability that a partner is infectious
is
p(t)= g H i H (t) + g L i L(t)
And the forces of infection are:
H (t) =c H p p = c H p [g H i H (t)+ g L i L(t)]
L (t) =c L p p = c L p [g H i H (t)+ g L i L(t)]

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.5: Practical Exercise: Modelling an STI in a

population with heterogeneous activity
page 11 of 38
prev
next
A model with heterogeneity in sexual activity has been set up for you. Use this model to
explore how the heterogeneity affects the spread of gonorrhoea.
In order to make a valid comparison with the simple (homogeneous activity) model used
earlier in this session, the mean rate of partner change, c mean , is kept constant at 2
partners per year by using the following formula:
cH =
c mean N - c LNL
NH
This means that as you change the sexual activity in the low activity group in this model,
the activity level of the high activity group will also vary to ensure that the overall mean
remains at 2 partners per year.
You will see this formula in the section "Initial conditions and parameters" in the model
where it looks like this:
c_H = ( c_mean * N - c_L * N_L ) / N_H ; Partner change rate of high-activity
group member
Open the model and find this formula. Then proceed to the next page to start the exercise.
Click here
to open the Berkeley Madonna file with the flowchart.
Click here
to open the Berkeley Madonna file without the flowchart.

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.6: Practical Exercise (2)

page 12 of 38
prev
next
The plot of prevalence over time is shown in the graph on the bottom-left. (Ignore the
graph bottom-right for now). When you first run the model you will find that you still do not
see sustained transmission. This is because both groups are still set to have 2 partners
per year.
Now use the slider to reduce the partner change rate, cL, for the low activity group.
Q1.6 What happens to the partner change rate in the high activity group as you lower the
rate in the low activity group? (hint, open the parameter window)
Answer
Q1.7 At what rate of partner change in the low activity group does gonorrhoea prevalence
start to increase?
Answer
Q1.8 What is the corresponding partner change rate in the high activity group when this
occurs? (look at graph for answer)
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.7: Practical Exercise (3)

page 13 of 38
prev
next
Now set the partner change rate in the low-activity group to 1.4 partners per year.
Q 1.9Using your model or Excel or a calculator, calculate the fractions of all partnerships
provided by the high and low activity groups in this situation?
(Remember that for the high activity group this fraction, g H , is found using the formula:
gH
=
c H NH
c H NH + c L NL
and similarly for g L. Note that, in the model, 2% of the population is in the high activity
group)
Answer
Q1.10 Using you model, what is the equilibrium prevalence of gonorrhoea in the lowactivity group, high-activity group and overall in this situation?
Answer
Q1.11Is this partner change rate in the low-activity group more plausible than the rate we
had to assume in order to get gonorrhoea to invade if we did not model heterogeneity?
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 4.8: Calculation of R0

page 14 of 38
prev
next
Importantly, this means that by introducing heterogeneity in sexual activity into our model,
while we keep the mean rate of partner change per year constant, gonorrhoea will invade
the population (unlike the situation without behavioural heterogeneity) and we will see
something like this:
Figure 6. Predicted prevalence of gonorrhoea in a model with heterogeneity in sexual activity. Parameter values
in text.
The formula for the calculation of R0 assuming proportionate mixing is analogous to the
form of the force of infection equation:
R0 =g H RH + g L RL
where RH =CH p D and RL= CL p D
Q1.12 If c H = 31.4 & c L = 1.4 partners per year, p = 0.75, D = 2 months, and g H = 0.31,
what is R0? (hint:g L = 1 - g H )
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 5: What happens after invasion...?

page 15 of 38
prev
next
So just by introducing heterogeneity in sexual activity into our model, while keeping the
mean rate of partner change per year constant, we have been able to show that
gonorrhoea can invade the population.
What happens after the infection invades a population depends not only on the value of R0
but also on whether infection is followed by re-susceptibility, immunity or death. For
infections such as gonorrhoea, not evoking an effective immune response in the host (and
therefore modelled using an SIS model structure), infection prevalence rises steadily.
Figure 7. Predicted prevalence and overall incidence of gonorrhoea in a model with heterogeneity in sexual activity. The right-hand figure
also shows the relationship between the basic ( R0 ) and net ( Rn ) reproduction numbers.
As this happens the net reproduction number, Rn, falls from its initial value of R0 (1.36
here) to 1 when equilibrium prevalence is reached. Whereas, if infection leads to immunity
or death, the prevalence of infected and infectious individuals can rise and then fall HIV/AIDS being an example of this effect.

Home
Contents
Module Map
Glossary
Tools
Help
Section 6: Effect of varying heterogeneity in sexual activity

on equilibrium STI prevalence
page 16 of 38
prev
next
More generally, the same average population rate of partner change can result from any of
a wide range of combinations of partner change rates in each activity group. Consequently
equilibrium prevalence results will also vary widely. The figure shows what happens if as
we change the difference between the partner change rates in the high and low-activity
groups:
Figure 8.The effect of increasing heterogeneity in sexual activity in a population on equilibrium STI prevalence.
Parameter values in text.
As can been seen, once heterogeneity increases above a critical level, R0 rises above one
and gonorrhoea can invade. In our example we can now see that this occurs when the
partner change rate is actually below 1.5 per year in the low-activity group and therefore
above around 26 per year in the high-activity group.
The figure also shows that, if we continue to increase heterogeneity in sexual activity,
gonorrhoea prevalence in the overall population (middle line) first increases and then
decreases. In our example, the overall prevalence peaks at around 0.8 and 61 partners
per year in the low and high-activity groups respectively, after which further increases in
heterogeneity lead to a reduction in the overall prevalence.
At first sight this is rather surprising, but the explanation can be seen by examining what is
happening in the two groups separately. Equilibrium prevalence in the low activity group
first increases as partner change rates in this group decrease. This is because, although
the partner rate in the low activity group is decreasing, this factor is initially outweighed by
the increase in the probability of a partner being infectious due to the steep rise in the
infected proportion of the high activity group.
However, further decrease in the partner rate of the low activity group eventually
counterbalances this effect as prevalence in the low activity group reaches a peak and
then starts to decline. In the end the infection goes extinct in the low activity group.
If you would like to see this for yourself, open Model 8.2 again and run the parameter plot
(the graph on bottom right): (note the plot is the mirror image of the plot above)
Click here
to open the Berkeley Madonna Model 8.2 with the flowchart.
Click here
to open the Berkeley Madonna Model 8.2 without the flowchart.

Home
Contents
Module Map
Glossary
Tools
Help
Section 7: Mixing by sexual activity

page 17 of 38
prev
next
In the contact patterns section (MD06 ) we saw that the degree to which the various
groups within a population contact each other strongly determines the impact of
interventions designed to control respiratory infections. This is also the case for sexually
transmitted infections.
Up until now we have been assuming proportionate mixing by sexual activity, but this is
only one possibility. Mixing between population groups can be broadly categorised into
three types:
Proportionate mixing, so called because partnerships are formed between population
groups based on the proportion of all partnerships generated by these groups. This is also
sometimes called random mixing, but this can be confused with randomly selecting
partners based on the proportion of individuals in each group. Therefore describing this
type of mixing as random is discouraged.
With-like mixing, in which individuals preferentially form partnerships with individuals who
have characteristics like their own, for example those with many sexual partners
preferentially choose partners who also have many sexual partners. This is often called
assortative mixing.
With-unlike mixing, in which individuals preferentially form partnerships with individuals
who have characteristics unlike their own, for example those with many sexual partners
preferentially choose partners with few sexual partners. This is often called disassortative
mixing.

Home
Contents
Module Map
Glossary
Tools
Help
Section 7.1: Mixing by sexual activity

page 18 of 38
prev
Similar to the methods used

when we considered
respiratory infections, contact
patterns can be summarised
using mixing matrices in which
each of the elements g jk in the
matrix is the probability that
someone in group k forms a
partnership with someone in
group j. In line with the
convention used earlier, the
second subscript, k, refers to
the "choosing" partner and the
first subscript, j, refers to the
"chosen" partner.
These matrices are similar to
the WAIFW matrices
described in MD06 , but the
matrix elements in MD06
represented the rate at which
two specific individuals come
into effective contact and
therefore included the
transmission probability per
contact, whereas for STIs the
transmission probability per
contact is typically considered
separately.
For proportionate mixing
sexual partners in a given
group are selected randomly
in proportion to the number of
partnerships that the group
generates. For this

assumption the probability
that someone in the highactivity group forms a
Partner j
Partner K
Proportionate
mixing
next
Figure Mixing pattern
Purely with
like mixing
Purely with
unlike mixing
the high-activity group, g HH , is

the same as the probability
that someone in the lowactivity group forms a
the high-activity group, g HL i.e.
g HH = g HL = g H .
In our example g H equals 0.31
(see here for calculation)
Similarly, the probability that
someone in the low-activity
group forms a partnership
with someone in the lowactivity group, g LL , is the
same as the probability that
someone in the high-activity
group forms a partnership
with someone in the lowactivity group, g LH , i.e. g LH =
g LL = g L . Here g Lequals 0.69
(see here for calculation)

Home
Contents
Module Map
Glossary
Tools
Help
Section 7.2: A summary measure of mixing - Q

page 19 of 38
prev
next
The degree of mixing can be summarised using a statistic that has been given the symbol
Q by Gupta and colleagues .
Q is equal to 0 when all partners are selected proportionately,
Q is equal to 1 when all partners are selected purely with-like, and
Q is negative when partners are selected purely with-unlike.
Q1.13 What is Q in our model?

Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 7.3: Data on mixing by sexual activity

page 20 of 38
prev
next
Studies on mixing between different sexual activity groups are uncommon, as data are
required on characteristics of sexual partners. The data from the USA summarised in the
figure below suggest that, on average, the individuals in these groups tended to form
sexual partnerships with individuals more similar to themselves in terms of sexual activity
than the proportionate mixing assumption would predict (i.e. Q was slightly greater than 0).
The data shown in the top three rows of the figure are based on data from late 1990s STI
clinic contact tracing studies. As these individuals are likely to be higher risk than the
general population, these data do not necessarily generalise to the rest of the US
population. However the bottom row of the figure is based on data from a 1992 general
population survey which also indicate slightly with-like mixing.
These findings are intuitively plausible, as in general people tend to socialise with people
similar to themselves (see McPherson et al. ).
Figure 9: The value of the measure Q for mixing between sexual activity groups in four studies of sexual
behaviour in the US, three of which were based on contact tracing in STI clinics, and one (the National Health
and Social Life Survey) was based on a survey of the general population. Figure based on Garnett et al 1996.
Note in this example Q was calculated on data categorised into three groups, and therefore purely with-unlike
mixing was indicated by a Q value of -1/2 rather than -1 as in our two-activity group example.

Home
Contents
Module Map
Glossary
Tools
Help
Section 8: What effect does changing patterns of mixing

have?
page 21 of 38
prev
next
We will now look at the effect mixing has on the spread of STIs in 3 different ways:
1. Effects of mixing on R0
2. Effects of mixing on rate of STI spread and equilibrium STI prevalence (for a fixed R0
value)
3. Effects of mixing on equilibrium STI prevalence (for a given STI natural history and
partner change rates)
Finally in this section we will look at the implications of heterogeneity in sexual activity as
well as the effect that mixing between these groups has on STI control.

Home
Contents
Module Map
Glossary
Tools
Help
Section 9: Effect of mixing on R0

page 22 of 38
prev
next
If other parameters are kept constant, as we model more with-like mixing (Q increases) R0
also increases.
This is because increasing with-like mixing means higher-activity individuals tend to
contact other higher activity individuals more frequently. As higher-activity individuals are
more likely to be infected, increasing with-like mixing leads to an increase in the probability
a "typical-infectee" will be a member of the high-activity group. Higher-activity individuals
have higher partner change rates and therefore generate more secondary infections in a
completely susceptible population. Therefore the value of R0 increases.
Figure 10. Predictions for
the basic reproduction
number, R0 , for a curable
STI in a single-sex, twoactivity group model as
the mixing pattern
between activity groups
varies from "More withunlike" (Q = 0.46, the
most with-unlike mixing
possible without altering
partner change rates),
through proportionate (Q
= 0), to purely with-like
mixing (Q = 1). p = 0.75,
D = 0.167 years, cL = 1.4
partners/year, cH = 31.4
partners/year. Two per
cent of the population
belong to the high-activity
group. An arrow shows
the value of R0 (1.36)
assuming proportionate
mixing as illustrated in
section 8.4. A horizontal
line highlights R0 = 1.
Note that the lower limit of Q (-0.46 in this instance) is the consequence of the different
number of partnerships by the high and low activity groups.
If you would like to understand this better, see section 8.5.3 in the book

Home
Contents
Module Map
Glossary
Tools
Help
Section 9.1: Effect of mixing on R0

page 23 of 38
prev
next
More with-like mixing results in a higher R0 with prevalence in the high activity group rising
more quickly.
Open the model incorporating Q that has been set up for you:
Click here
Click here

Q1.14 By looking at the prevalence plot (bottom-right graph) and varying the value of Q
using the slider, estimate the value of Q at which the infection becomes extinct. What is it?
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 10: Effect of mixing on rate of STI spread and

equilibrium prevalence (for a fixed R0 value)
page 24 of 38
prev
next
We have seen that increasing with-like mixing tends to lead to the infection invading more
rapidly. But this does not also mean that the endemic prevalence will be higher. For a
given R0, increasing with-like mixing does mean a more rapid increase in prevalence, but
increasing with-like mixing tends to result in lower equilibrium prevalence. The more rapid
invasion is because a larger proportion of the partnerships by high activity people are with
other high activity people, generating more secondary infections.
Figure 11. Predictions of the overall prevalence of infectious individuals over time in a curable STI model with
more with-unlike ( Q = 0.4 ), proportionate ( Q = 0 ), or more with-like ( Q = +0.4 ) mixing between activity
groups. R0 was set equal to 1.36 in all scenarios by varying the duration of infection ( D = 0.340, 0.167 and
0.097 years, respectively ). p = 0.75, cL = 1.4 partners/year, cH = 31.4 partners/year. Two per cent of the
population belong to the high-activity group.
But this results in lower endemic prevalence than with proportionate or with-unlike mixing
because more contacts are "wasted" on partners that are already infected. This effect was
called "pre-emptive saturation" by Hethcote & Yorke.

Home
Contents
Module Map
Glossary
Tools
Help
Section 10.1: Effect of mixing on rate of STI spread and

equilibrium prevalence (for a fixed R0 value)
page 25 of 38
prev
next
You can see this effect of mixing on the rate of STI spread for yourself by running Model
8.3 using each of these combinations of the duration of infection (D), and Q in turn.
Prevalence over time is shown in the bottom-right graph.
We need to alter the duration of infection, D, each time to ensure any effect we see is not
due to variation in R0.
Remember you can overlay the three plots on top of each other by pressing the "O" button
on the graph before you start.
For more with-unlike mixing (set D=0.340 & Q = -0.4)
For proportionate mixing (set D=0.167 & Q=0.0)
For more with-like mixing (set D=0.097 & Q = +0.4)
Click here
Click here

You should see (click here to see how the graph should look) that increasing with-like
mixing tends to lead to the infection invading more rapidly. But if R0 is kept constant and
high enough for the infection to invade for all mixing patterns, this also results in lower
endemic prevalence than for proportionate or with-unlike mixing. This is because, for a
given prevalence in the population, more contacts are "wasted" on already-infected
partners if mixing is with-like.
However, increasing with-like mixing does not always lead to lower endemic prevalence if
R0 is not kept constant as we shall see next...

Home
Contents
Module Map
Glossary
Tools
Help
Section 11: Effects of mixing on equilibrium STI prevalence

(for fixed STI natural history & partner change rates)
page 26 of 38
prev
next
We just looked at the effect on equilibrium prevalence of changing the mixing pattern with
R0 held constant by increasing the duration of infection of the STI.
It is also interesting to explore what may happen if the mixing pattern is changed but the
STI natural history and the partner change rates do not change. Plausibly, this could be
the intended, or unintended, consequence of a behaviour change intervention.
The figure shows that increasing with-like mixing may not always lead to a reduction in
equilibrium prevalence. For lower values of R0, increasing with-like mixing may actually
allow the infection to invade a population and lead to a rise in overall prevalence (Figure
12). Conversely, for higher values of R0, increasing with-like mixing will always lead to a
fall in overall prevalence because more contacts are "wasted" on already-infected
individuals in the high-activity group (Figure 13). For moderate values of R0, both effects
may be seen, so that increasing with-like mixing may lead to an initial rise, but subsequent
fall in overall prevalence (Figure 14).
Figure 12
Figure 13
Figure 14
Figure 12. Equilibrium prevalence of a curable STI by mixing pattern and R0 . Assumes partner change rate in
high and low-activity groups are 31.4 and 1.4 partners per year, respectively, the high-activity group is 2 per
cent of the population, the per-partnership transmission probability is 0.75. In the low, moderate and high R0
scenarios, the durations of infection are 0.062, 0.167, and 0.493 years, respectively. Note a log 10 scale is
used on the y axis for clarity. Note that only two lines appear in the low R0 scenario because the prevalence of
infection in the low-activity group remains below 0.1 per cent for all values of Q.

Home
Contents
Module Map
Glossary
Tools
Help
Section 11.1: Effects of mixing on equilibrium STI prevalence

page 27 of 38
prev
next
Q1.15 For an STI with an R0 of 1.36, what would happen to overall STI prevalence if an
intervention resulted in a change (intended or unintended) in mixing by sexual-activity :
a. from proportionate (Q = 0) to slightly with-like (Q = 0.3)?
Answer
b. from slightly with-like (Q = 0.3) to mostly with-like (Q = 0.9)?

Answer
c. from proportionate (Q = 0) to slightly with-unlike (Q = -0.3)?

Answer
All this has implications for STI control as we shall see next...

Home
Contents
Module Map
Glossary
Tools
Help
Section 12: Mixing and STI control

page 28 of 38
prev
next
The impact of screening varies depending on the mixing pattern.
Figure 15. The effect of screening the population on equilibrium curable STI prevalence in the overall population,
by mixing pattern and target group. The mixing patterns shown are: more with-unlike ( Q = 0.4 ), proportionate (
Q = 0 ), more with-like ( Q = +0.4 ). Screening is targeted either randomly, or at high or low activity groups. R0
was set equal to 1.36 in all scenarios by varying the duration of infection ( D = 0.340, 0.167 and 0.097 years,
respectively).
The figure shows that, unsurprisingly, prioritising higher-risk individuals is markedly more
effective per-person-screened than prioritising individuals at random, or prioritising those in
the low-activity group. Looking at the middle figure above, if mixing is proportionate, one
million screenings per year does not markedly affect prevalence when randomly distributed
(solid line) or targeted at the low-activity group (dotted line), but it eradicates the infection if
targeted at the high-activity group (dashed line).
The figure shows also that increased with-like mixing tends to make STIs more difficult to
control. Impact on equilibrium prevalence per individual screened is smaller in populations
with more with-like mixing. Looking across all three panels, above, if 20 million screenings
were randomly distributed each year (solid line), they would reduce the relative equilibrium
prevalence by around a third in the more with-like mixing scenario, and by around a half in
the proportionate mixing scenario, but would almost eradicate the infection in the more
with-unlike mixing scenario.
MD03 Basic Dynamics of Infections

Home
Contents
Module Map
Glossary
Tools
Help
Section 12.1 Mixing and STI control

page 29 of 38
prev
next
To see the effect of increasing with-like mixing, run the models below that have been set
up for you, with the different levels of mixing determined by the parameter Q: Again we
vary the duration of infection, D, to ensure any effect we are seeing is not due to variation
in R0 :
The model is very similar to Model 8.3 but screening has been implemented by adding a
term to the equations determining the rate of change of the infectious and susceptible
individuals in the high and low activity groups, to simulate a higher rate of recovery due to
treatment. These two rates are then altered so that the number of screenings is kept
constant but the screenings are targeted at the high or low activity group, or distributed
randomly.
The model is set up to model the proportionate mixing scenario: (D = 0.167 & Q = 0).
Open the model and run the parameter plot (the bottom-left graph) to see the effect of
increasing the number of screenings per year has on endemic prevalence.
Click here
Click here

By changing the values to those below, you can compare this to the impact in scenarios of
more with-unlike mixing and more with-like mixing:
Remember you can overlay the three plots on top of each other by pressing the "O" button
on the graph before you start.
For more with-unlike mixing (set D=0.340 & Q = -0.4)
For more with-like mixing (set D=0.097 & Q = +0.4)
You should be able to see (click here to see how the graph should look) that as mixing
becomes more with-like, STI control becomes more difficult and people must be screened
more often to get the same relative reduction in prevalence.
MD03 Basic Dynamics of Infections

Home
Contents
Module Map
Glossary
Tools
Help
Section 13: Models of Human Immunodeficiency Virus/AIDS

page 30 of 38
prev
next
Acquired Immune Deficiency Syndrome (AIDS) is the disease caused by the human
immunodeficiency virus (HIV). This retrovirus targets CD4 cells leading, in absence of
treatment, to immune system collapse and ultimately death.
HIV has been the focus of many modelling studies. Many of the insights gained from
modelling curable bacterial STIs also apply to HIV, as there are a number of features in
common. However there are also notable differences between HIV and curable bacterial
STI, some of which are listed below.
All STI
Density dependent force of infection
assumption is not appropriate
Importance of core groups and mixing
patterns
HIV
Less infectious than curable bacterial
STIs
Longer duration of infectivity than
curable bacterial STIs
Infectivity varies over time
No cure
Death from AIDS without treatment

Home
Contents
Module Map
Glossary
Tools
Help
Section 13.1: Modelling HIV

page 31 of 38
prev
next
The SIS model with heterogeneity in sexual activity which we used for gonorrhoea can be
adapted to model HIV. We remove the possibility of recovery/cure and introduce two new
compartments A H and A L for the AIDS stage of infection and a corresponding third
differential equation. As we are modelling long term dynamics of a persistent infection that
leads to death we also need to include mortality, , due to AIDS and to other causes, m,
and to avoid running out of susceptibles, we need to model recruitment to the sexually
active population (a H and a L).
Figure 16. A simple model of HIV transmission and progression.
For simplicity we are ignoring gender and assume proportionate mixing and assume that
there is no sexual activity in the AIDS stage.
Click here for a key to the model labels. The corresponding equations can be seen here .

Home
Contents
Module Map
Glossary
Tools
Help
Section 14: Calculating R0 for HIV

page 32 of 38
prev
next
The R0 for HIV can be calculated in a similar way to the R0 for bacterial STIs:
R0 = cD
We assume that average time from HIV infection to death resulting from AIDS is 9
years. However we are also modelling non-AIDS mortality, m and we need to take this
into account when we calculate the duration of infectivity, D. Here we assume life
expectancy in absence of HIV is 50 years, and the average age of sexual debut is 15
years so m = 1/(50-15) = 1/35 per year. Therefore, overall, we are assuming that:
D=
1
= 7.16 years
1/9 + 1/35
If we assume a transmission probability per partnership of 0.05 and partner change

rates in the low and high activity groups of, 0.2 and 8 per year respectively, then the
number of secondary infections from infected low (RL) and high (RH ) activity individuals will
be:
RL = c LpD = 0.2 0.05 7.16 = 0.07
RH = c H pD = 8 0.05 7.16 = 2.86
If we also assume 15% of people are in the high activity group, as we are assuming
proportionate mixing the probabilities that a new partner will be selected from the low or
high activity groups are:
gH = 0.88 and gL= 0.12

And therefore R0 for HIV in our model population will be :
R0 = g LRL + g H RH = 0.88 2.86 + 0.12 0.07 = 2.52

Home
Contents
Module Map
Glossary
Tools
Help
Section 15: HIV Model predictions

page 33 of 38
prev
next
Those in the higher sexual activity group are more likely to become infected and
subsequently to die from AIDS. A consequence is that, in absence of other changes, the
model predicts equilibrium prevalence and incidence below peak levels (a). This is
because, without increasing the rate of recruitment into the higher activity group to replace
those who do not survive, the average rate of partner change in the population will
decrease over time (b).
Numbers of new infections and AIDS deaths will also come into equilibrium below peak
levels (c).
Such falls in prevalence and incidence tend to make the evaluation of the impact of HIV
interventions more difficult.
Figure 16. Predictions of (a) the incidence and prevalence of HIV and cumulative AIDS deaths; (b) mean partner
change rate in the population; and (c) numbers of new HIV infections and deaths of HIV infecteds.

Home
Contents
Module Map
Glossary
Tools
Help
Section 15.1: Model predictions

page 34 of 38
prev
next
You can explore these effects using models that have been set up for you:
Click here
Click here

Q 1.16 Looking again at the figure16a, how long does it take for HIV to reach its peak in
our model?
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 15.2 Rate of increase in HIV prevalence

page 35 of 38
prev
next
However, data show that HIV has spread faster than suggested by our models in many
countries in sub-Saharan Africa.
Figure 17. Estimated trend in adult HIV (with high and low estimate) prevalence (1549 years).
UNAIDS/WHO, 2008. Note different scales on y axis.
Q1.17 Why do you think HIV may have spread faster in Eastern and Southern Africa than
our models suggested?
Answer

Home
Contents
Module Map
Glossary
Tools
Help
Section 16: Summary

page 36 of 38
prev
next
STI modelling has:

been an enormously useful tool to understand the epidemiology and control of STIs.
helped identify key behaviours important for understanding the spread of STIs, e.g.
risk heterogeneity and mixing patterns.
allowed us to predict likely trends in prevalence and incidence of different STIs and
the impact of existing and hypothetical control strategies.
continued to be an important area for research, as millions of people are newly
infected with HIV each year & the global disease burden of other STIs remains high.

Home
Contents
Module Map
Glossary
Tools
Help
Section 17: Summary of session

page 37 of 38
prev
next
We have described and shown:

what are the important characteristics of STI for the purposes of modelling, and how
STI modelling differs from the modelling of infections described in the previous
sessions.
how simple compartmental models may be used to explore the transmission
dynamics and control of short-duration curable STIs such as gonorrhoea.
how a simple compartmental model may be used to explore the transmission
dynamics of HIV/AIDS.
We hope you have enjoyed this brief introduction to the modelling of STIs. Please see the
references for further reading.

Home
Contents
Module Map
Glossary
Tools
Help
References
page 38 of 38
prev
1. Anderson R, May R. Ch 11: Social heterogeneity and sexually transmitted

diseases.Oxford: Oxford University Press; 1991
2. Boily M-C & Masse B (1997) Mathematical models of disease transmission: A
precious tool of the study of sexually transmitted diseases. Canadian Journal of
Public Health 88:255-265.
3. CASCADE. Time from HIV-1 seroconversion to AIDS and death before widespread
use of highly-active antiretroviral therapy: a collaborative re-analysis. Collaborative
Group on AIDS Incubation and HIV Survival including the CASCADE EU Concerted
Action. Concerted Action on SeroConversion to AIDS and Death in Europe. Lancet
2000;355(9210).
4. Expert Group on Modelling the Impact and Cost of Male Circumcision for HIV Risk
Reduction. Informing decision making on male circumcision for HIV prevention in
high HIV prevalence settings: what mathematical modelling can contribute. PLoS
Med 2009; 6(9): e1000109
5. Garnett GP, Hughes JP, Anderson RM, Stoner BP, Aral SO, Whittington WL, et al.
Sexual mixing patterns of patients attending sexually transmitted diseases clinics.
Sex Transm Dis 1996; 23(3):248-257.
6. Gray R, Azire J, Serwadda D et al. Male circumcision and the risk of sexually
transmitted infections and HIV in Rakai, Uganda. AIDS 2004; 18(18): 657-666.
7. Gupta S, Anderson RM, May RM. Networks of sexual contacts: implications for the
pattern of spread of HIV. AIDS 1989;3(12).
8. Hallett TB, Singh KJ, Smith JA, White RG, Abu-Raddad L, Garnett GP.
Understanding the impact of male circumcision interventions on the spread of HIV in
Southern Africa. PLoS ONE 2008; 3(5): e2212.
9. Hethcote H, Yorke J. Lecture notes in biomathematics: Gonorrhea transmission and
control (vol 56). 1984. Report No.: 56
10. McPherson M, Smith-Lovin L , Cook JM. Birds of a feather: homophily in social
networks. Annual Review of Sociology 2001; 27(1):415-444.
11. Morgan D, Mahe B, Okongo MJ, Lubega R, Whitworth JA. HIV-1 infection in rural
Africa: is there a difference in median time to AIDS and survival compared with that
in industrialized countries? AIDS 2002;16.
12. Nagelkerke NJ, Moses S, De Vlas SJ, Bailey RC. Modelling the public health impact
of male circumcision for HIV prevention in high prevalence areas in Africa. BMC
Infect Dis 2007; 7(1): 16.
13. Schneeberger A, Mercer CH, Gregson SA, Ferguson NM, Nyamukapa CA,
Anderson RM, et al. Scale-free networks and sexually transmitted diseases: a
description of observed patterns of sexual contacts in Britain and Zimbabwe. Sex
Transm Dis 2004;31(6)
14. Todd J, Glynn JR, Marston M, Lutalo T, Biraro S, Mwita W, et al. Time from HIV
seroconversion to death: a collaborative analysis of eight studies in six low and
middle-income countries before highly active antiretroviral therapy. AIDS 2007;21
Suppl 6.
15. Wawer MJ, Gray RH, Sewankambo NK, Serwadda D, Li X, Laeyendecker O, et al.
Rates of HIV-1 Transmission per Coital Act, by Stage of HIV-1 Infection, in Rakai,
Uganda. J Infect Dis 2005;191(9)
16. White RG, Glynn JR, Orroth KK et al. Male circumcision for HIV prevention in subSaharan Africa: who, what and when? AIDS 2008; 22(14): 1841-1850.
17. Williams BG, Lloyd-Smith JO, Gouws E et al. The potential impact of male
circumcision on HIV in Sub-Saharan Africa. PLoS Med 2006; 3(7): e262.
Further reading:
Vynnycky E and White RG (2010) An introduction to infectious disease modelling. Oxford
University Press. Chapters 8 and 9.

md08 PDF

Hochgeladen von

Dokumentinformationen

Originalbeschreibung:

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

md08 PDF

Hochgeladen von

Copyright:

Verfügbare Formate

MD08 Modelling HIV and STIs

EPM302 Modelling and the Dynamics of Infectious Diseases

Section 1: Modelling sexually transmitted infections

MD08 Modelling HIV and STIs

Section 2: Important characteristics of STIs

STIs share a number of characteristics important for how they are

MD08 Modelling HIV and STIs

Section 3: The simplest STI model

Figure 2. A simple model of STI transmission dynamics.

Then describe the same model algebraically

MD08 Modelling HIV and STIs

3.1: Reformulating the force of infection equation for STIs

Figure 2. A simple model of STI transmission dynamics.

MD08 Modelling HIV and STIs

3.2: Simple model prediction

Figure 2. A simple model of STI transmission dynamics.

to open the Berkeley Madonna file with the flowchart.

MD08 Modelling HIV and STIs

Section 4: Importance of heterogeneity in sexual activity

MD08 Modelling HIV and STIs

Section 4.1: Incorporating heterogeneity in sexual activity

Figure 5. Model of gonorrhoea transmission incorporating heterogeneity in activity

MD08 Modelling HIV and STIs

Section 4.2: Incorporating heterogeneity in sexual activity

Model incorporating heterogeneity in sexual activity (2)

MD08 Modelling HIV and STIs

Section 4.3: Probability a partner will be infectious in an STI

MD08 Modelling HIV and STIs

Section 4.4: Force of infection for a STI model with

MD08 Modelling HIV and STIs

Section 4.5: Practical Exercise: Modelling an STI in a

to open the Berkeley Madonna file with the flowchart.

to open the Berkeley Madonna file without the flowchart.

MD08 Modelling HIV and STIs

Section 4.6: Practical Exercise (2)

MD08 Modelling HIV and STIs

Section 4.7: Practical Exercise (3)

MD08 Modelling HIV and STIs

Section 4.8: Calculation of R0

MD08 Modelling HIV and STIs

Section 5: What happens after invasion...?

MD08 Modelling HIV and STIs

Section 6: Effect of varying heterogeneity in sexual activity

to open the Berkeley Madonna Model 8.2 with the flowchart.

to open the Berkeley Madonna Model 8.2 without the flowchart.

MD08 Modelling HIV and STIs

Section 7: Mixing by sexual activity

MD08 Modelling HIV and STIs

Section 7.1: Mixing by sexual activity

Similar to the methods used

generates. For this

Figure Mixing pattern

Figure Mixing pattern

Figure Mixing pattern

the high-activity group, g HH , is

MD08 Modelling HIV and STIs

Section 7.2: A summary measure of mixing - Q

Q1.13 What is Q in our model?

MD08 Modelling HIV and STIs

Section 7.3: Data on mixing by sexual activity

MD08 Modelling HIV and STIs

Section 8: What effect does changing patterns of mixing