See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/47717793

Introduction to Infectious Disease Modelling

Article in Sexually Transmitted Infections · November 2010

DOI: 10.1136/sti.2010.046342 · Source: PubMed

CITATIONS READS

26 21,749

1 author:

Katherine Mary Elizabeth Turner

University of Bristol
166 PUBLICATIONS 6,350 CITATIONS

SEE PROFILE

All content following this page was uploaded by Katherine Mary Elizabeth Turner on 21 August 2014.

The user has requested enhancement of the downloaded file.

An Introduction to
Infectious Disease
Modelling
Solutions to exercises

Emilia Vynnycky and Richard G. White

with an introduction by Paul E.M. Fine
2 | SOLUTIONS TO EXERCISES
 Chapter 2 (Solutions)
How are models set up?
I. An introduction to difference equations

2.1 a) The difference equations are as follows:

Humans: s th1  s th  λht s th  ri th

i th1  i th  λht s th  ri th
Mosquitoes s tv1  s tv  b  λvt s tv  μs tv
i tv1  i tv  λvt s tv  μi tv
Note that the compartments are defined to be the proportions, rather than the numbers
of mosquitoes or humans that are susceptible or infected. Since b is the per capita
birth rate into the mosquito population, we just need to add b into the equation for
susceptible mosquitoes to account for births into the population.

b) The risk of infection among humans needs to account for the prevalence of infected
mosquitoes, the number of mosquitoes per human, the biting rates of mosquitoes, and
the probability that a bite by a mosquito leads to infection in a human.

The risk of infection among mosquitoes needs to account for the biting rate of
mosquitoes, the probability that a bite by a mosquito leads to infection in a mosquito,
and the prevalence of infectious humans.

2.2 We use the symbols b and m to denote the per capita birth and death rates
respectively, and the symbol Nt to denote the population size at time t.

a) The equations can be rewritten as follows:

St+1= bNt + St – λt St - mSt

Et+1= Et + λt St – f Et - mEt
It+1 = It + f Et – r It - mIt
Rt+1 = Rt + r It - mRt

The diagram for this model is provided below, where the expressions next to or above
the arrows reflect the number of individuals who move between categories per unit
time:
4 | SOLUTIONS TO EXERCISES

bNt λtSt fEt rIt

Susceptible Pre-infectious Infectious Immune
St Et It Rt
mSt mEt mIt mRt

b) The model describes the transmission dynamics of an immunizing infection, and is

therefore sufficient for describing the general patterns in incidence for measles and
rubella, which are both immunizing infections. There are several ways of making the
model more realistic than it is at present:

1. Stratify the compartments by age;

2. Include age-dependent contact between individuals;
3. Include changes in mixing patterns during the course of a year because
of school holidays and school terms;
4. Assume that infectious individuals have a different mortality rate from
those who are susceptible or are immune;
5. Include maternal immunity.

These issues are discussed in later chapters.

c) The equations would be rewritten as follows:

St+1= St + b(1-v)Nt – λt St - mSt

Et+1= Et + λt St – f Et - mEt
It+1 = It + f Et – r It - mIt
Rt+1 = Rt + bvNt + r It - mRt

i.e. the proportion of newborns that is immunized enters the immune compartment and
the remainder enters the susceptible compartment.

The diagram for this model is provided below, where the expressions next to or above
the arrows reflect the number of individuals who move between categories per unit
time:

bvNt

b(1-v)Nt λtSt fEt rIt

Susceptible Pre-infectious Infectious Immune
St Et It Rt
mSt mEt mIt mRt

2.3 The following shows the general structure of the model.

λt r2
St It Rt
r1
 CHAPTER 2: HOW ARE MODELS SET UP? AN INTRODUCTION TO DIFFERENCE EQUATIONS |5

This model doesn’t fall naturally into any of the categories presented in Figure 2.2. In
fact, it has been called a “compound model” and has been used to describe hookworm
data (see chapter 5).
6 | SOLUTIONS TO EXERCISES
 Chapter 3 (Solutions)
How are models set up?
II. An introduction to differential equations

3.1 The differential equations are as follows:

dS(t )
  λ(t )S(t )  vS(t )
dt
dSp (t )
  λ(t )S p (t )  w vV (t )  w n R (t )
dt
dI(t )
 λ(t )S(t )  rI (t )
dt
dI p (t )
 λ(t )S p (t )  rI p (t )
dt
dV (t )
 vS(t )  w vV (t )
dt
dR(t )
 rI (t )  w n R(t )  rI p (t )
dt

b) The authors would have chosen to use this model rather than an SIRS model so that
they could allow the duration of immunity and the infectiousness of infectious persons
to depend on whether or not individuals have been infected naturally or vaccinated.
The model structure used also allows the authors to allow the susceptibility to infection
to differ between those who have been vaccinated and those who have been neither
vaccinated nor infected. However, a drawback of having such a high level of detail in
the model is that not all of the input parameters that are needed may be known.

3.2 a) The model diagram is as follows; the expressions next to the arrows reflect the
number of individuals who move between the corresponding categories per unit time:
bvN(t)
λ(t)S(t) fE(t) rI(t)
b(1-v)N(t)
S(t) E(t) I(t) R(t)
mS(t) mE(t) mI(t) mR(t)
8 | SOLUTIONS TO EXERCISES
b) i) m is interpretable as the per capita mortality rate, which is assumed to be identical
for all individuals.

ii) N(t) is the total population size at time t.

c) Since both vaccinated individuals and those who are immune because of natural
infection have been put into the same compartment, the model assumes that natural
infection and vaccination provide the same level of protection..

3.3 a) The differential equations are as follows:

ds(t )
  λs ( t )
dt
dz(t )
 λs ( t )
dt
Notice that we have used the symbol λ for the force of infection in these equations,
rather than λ(t). This reflects the fact that the force of infection is assumed to be the
same over time.

b) i) The equations are: s(t)=e-λt or st=(1-λr)t where λr is the risk of infection in each year
of life.

ii) Assuming that the proportion (ever) infection is just 1-proportion susceptible, then
the proportion ever infected is given by the following:

z(t) = 1- e-λt , or
zt = 1-(1-λr)t

c) The following table provides the corresponding values for the proportion ever
infected by different ages:

Age Force of infection (% per year)

(years) 1% 10% 20%
5 0.0488 0.3935 0.6321
10 0.0952 0.6321 0.8647
20 0.1813 0.8647 0.9817
60 0.4512 0.9975 1.0000

Almost all individuals are predicted to have been infected by age 20 years in the high
transmission setting. Since rubella is an immunizing infection, i.e. once infected,
individuals are immune for life, very few individuals are infected as adults in high
transmission settings. The burden of rubella among adults is therefore likely to be
smallest in the high transmission setting. These issues are discussed further in
chapter 5.
 Chapter 4 (Solutions)
What do models tell us about the
dynamics of infections?

4.1 a) Figure S4.1a plots the observed data for Gothenburg. This shows that two
pandemic waves occurred, with the first occurring in July 1918 and the second
occurring in September-October 1918. The cumulative numbers of cases for these
waves are shown in Figure S4.1b. The natural log of the cumulative numbers of cases
for these two waves are shown in Figure S4.1c.
A. B.
Cumulative numbers of cases

3500 25000
Numbers of cases per week

3000
20000
2500

15000
2000

1500
10000

1000
5000
500

0 0

C. D.
9 12
-ln (Cumulative numbers of
-ln (Cumulative numbers of

8
10
7

6 8
cases)

5
cases)

6
4

3 4

2
2
1

0 0

Week beginning
Figure S4.1: Summary of A. The numbers of cases reported each week, B. The
cumulative numbers of reported cases and C. and D. the natural log of the cumulative
numbers of cases observed during the first and second waves (C. and D. respectively)
of the influenza pandemic in Gothenberg, Sweden in 1918.
10 | SOLUTIONS TO EXERCISES

A straight line can be drawn through the first 4 points of the natural log of the
cumulative numbers of cases for the first wave (corresponding to the period 6/7/1918-
27/7/1918); this line (drawn either by eye or formally by regression) has a slope of
0.367 per day.

A straight line can be drawn through the first 7 points of the natural log of the
cumulative numbers of cases for the second wave (corresponding to the period
7/9/1918-19/10/1918). This line has a slope of 0.104 per day.

The following summarizes the estimates for the net and basic reproduction numbers
obtained using the different formulae in Table 4.1, with R0 estimated to be about 4 for
the first wave and just under 3 for the second wave of the pandemic. These estimates
are slightly higher than the values that have typically been estimated for the 1918
(Spanish) influenza pandemic (see references in the book for details). Notice that
m
 ΛD ' 
ΛD   1
estimates obtained using the formula (1+ΛD)(1+ΛD’) and
 m  are very
  ΛD 
-n

1 -   1 
  n  
 
similar.

Equation used to calculate the Rn R0*

reproduction number: 1 wave 2nd wave
st st
1 wave 2nd wave
(s=0.7) (s=0.5)
1+ΛD 1.73 1.21 2.48 2.42
(1+ΛD)(1+ΛD’) 3.01 1.46 4.30 2.92
m m=n=10 2.94 1.37 4.20 2.75
 ΛD ' 
ΛD   1 m=n=100
 m 
  ΛD 
-n

1 -   1 
  n  
  2.94 1.36 4.20 2.73
* Calculated using the expression Rn/(proportion susceptible (s) at the start of the wave)

b) It would be sensible to apply the epidemic size formula to data from the two waves
separately.

Considering the first wave of the pandemic (taken to be during the period 6/7/1918-
31/8/1918), 4,657 individuals were reported to have experienced disease. If 70% of
individuals were susceptible at the start of the first wave (s0=0.7), the proportion that
were susceptible at the end of the first wave is given by the difference between 0.7 and
the proportion of the population who experienced disease during the first wave. This
calculation assumes that all of those who were reported as cases became immune
(see below). We therefore obtain the following result:

sf = 0.7 – 4,657/196,943 ≈ 0.676

 CHAPTER 4: WHAT DO MODELS TELL US ABOUT THE DYNAMICS OF INFECTIONS? | 11

ln(sf )  ln(s0 )
Substituting for s0 and sf into the equation R0  implies that R0 equals
sf  s0
the following:
ln(0.676)  ln(0.7)
R0   1.45
0.676  0.7

Considering the second wave of the pandemic (the period after 7/9/1918), 19,484
individuals were reported to have experienced disease. Assuming that 50% of
individuals were susceptible at the start of this wave (s0=0.5), then applying a similar
reasoning to that used to calculate sf for the first wave, we obtain the following for the
proportion of the population that was susceptible at the end of the second wave:

sf = 0.5 – 19,484/196,943 ≈ 0.401

ln(sf )  ln(s0 )
Substituting for s0 and sf into the equation R0  implies that R0 equals
sf  s0
the following:
ln(0.401)  ln(0.5)
R0   2.23
0.401 0.5

c) The estimates of R0 that are based on the growth rate are likely to be more reliable
than are those based on the final epidemic size, since they are independent of the
proportion of cases that are reported (unless this changes over time). It is unlikely that
all cases were reported during the pandemic, and therefore R0 based on the epidemic
size is likely to have been underestimated. However, estimates based on both
methods need to make assumptions about the proportion of individuals that are
susceptible at the start of the first and second waves. Whilst the values assumed (70%
and 50% for the start of the first and second waves respectively) are plausible, it is
unclear as to whether they are correct.

d) The lower estimate of R0 for the second wave (calculated using the epidemic growth
rate), as compared with that for the first wave suggests that in Gothenberg, the
transmissibility decreased between the first and second waves. However, the value for
R0 during the first wave seems somewhat high in contrast with estimates obtained
elsewhere (see references cited in the main text) and it seems plausible that the
proportion of cases that were reported changed during the early stages of the first
wave of the pandemic. Such changes in the proportion of cases that were reported
would have led to an overestimate in R0.

4.2 a) According to equation 4.31, the inter-epidemic period is given by the following:
L(D  D' )
T  2π
R0  1
We can rearrange this equation to obtain the following equation for R0:
4π 2L(D  D' )
R0  1 
T2
12 | SOLUTIONS TO EXERCISES

Substituting for L=70×365 days, T=2×365 days, D’=8 days and D=7 days into this
equation leads to the following:

4π 2  70  365  (7  8)
R0  1   29
(2  365) 2

b) According to equation 4.32, the inter-epidemic period (T) is given by the equation:
T  2π A(D  D' )

This equation can be rearranged to give the following for the average age at infection:
T2
A S4.1
4π 2 (D  D' )

Substituting for T=3×365 days, D’=8 days and D=7 days into this equation implies that
(3  365)2
A  2,025 days = 2,025/365≈5.5 years.
4π 2 (7  8)

The limitations of this estimate are as follows:

i) The equation on which it is based assumes that individuals mix randomly,
which is unrealistic (see chapter 7).
ii) The measles vaccination coverage increased after vaccination was
introduced in 1968, and therefore the inter-epidemic period would have
changed over time. This equation does not account for changes in the inter-
epidemic period over time.
 Chapter 5 (Solutions)
Age patterns

5.1 Adapting expression 5.25 in the book, the number of new infections per person
among individuals in age group a can be calculated using the expression:
λsa

where sa is the proportion of individuals in age group a who were seronegative. The
values obtained for the average number of new infections per 100 population are as
follows:

Average number of infections per year per 100

population
(calculated using λsa ×100)
Age group China Fiji UK
(years) λ=20%/yr λ=4%/yr λ=12%/yr
15-19 0.80 1.74 1.54
(=0.2×0.04×100) (=0.04×0.435×100) (=0.12×0.128×100)
20-29 0.86 1.15 1.04
(=0.2×0.043×100) (=0.04×0.288×100) (=0.12×0.087×100)
30-39 1.08 0.77 0.85
(=0.2×0.054×100) (=0.04×0.193×100) (=0.12×0.071×100)

We could have also used the expression sa×risk of infection, which, using the
relationship risk =1-e-rate (Panel 2.2), leads to the expression sa (1  e  λ ) for the
number of new infections per person. This expression leads to the following values for
the average infection incidence:

Average number of infections per year per 100

population
(calculated using sa (1  e  λ ) × 100 )
Age group China Fiji UK
(years) λ=20%/yr λ=4%/yr λ=12%/yr
15-19 0.73 1.71 1.45
20-29 0.78 1.13 0.98
30-39 0.98 0.76 0.80
14 | SOLUTIONS TO EXERCISES
Note that the greatest discrepancy between the values obtained using the expressions
λsa and sa (1  e  λ ) occurs for the estimates for China. This follows from the facts that
the force of infection is higher for China that it is for the UK and Fiji, and the difference
between the value for the risk (1  e  λ ) and the rate (λ) is greatest for large values of
the rate (see Figure 2.5).

The estimates suggest that the highest number of new infections per 100 population
would have been seen among 15-19 year olds in Fiji, followed closely by that for 15-19
year olds in the UK. Therefore, based on these estimates, we would expect the
incidence of CRS to have been greater among babies born to women in these age
groups in these countries, than for babies born to women in other age groups in the
same countries.

However, the overall burden of CRS depends both on the infection incidence and
number of livebirths among women in different age groups. Therefore, to infer the
setting in which the burden of CRS is likely to be the greatest, we would need to
combine the above estimates with the age-specific fertility rate.

5.2 a) Figure S5.1 plots the observed proportion seronegative. The median age at
infection is the point at which the vertical dotted line in this figure crosses the x-axis,
which occurs at about 6 years. This estimate suggests that the average force of
infection is about 100×1/617% per year.

100
90
% seronegative

80
70
60
50
40
30
20
10
0
0 10 20 30 40 50 60

Age (years)
Figure S5.1: Observed proportion of individuals who did not have antibodies to rubella
during 2004-5 in Bangladesh1

b) i)The overall proportion susceptible is calculated using equation 5.17 as the sum of
the proportion susceptible in each age group, weighted by the proportion of the
population that is in that age group (pa×Sa/Na). The final column in the following table
gives the values for pa×Sa/Na in each age group.
 CHAPTER 5: AGE PATTERNS | 15

Age group Number Number % Proportion of the

(years) tested negative negative population that is in the pa×Sa/Na
given age range (pa)
1-5 61 48 78.7 0.1139 0.08964
6-10 61 29 47.5 0.1135 0.05391
11-15 63 21 33.3 0.1109 0.03693
16-20 62 14 22.6 0.1087 0.02457
21-25 83 15 18.1 0.104 0.01882
26-30 67 11 16.4 0.0923 0.01514
31-35 63 12 19.0 0.0775 0.01473
36-40 60 7 11.7 0.0641 0.0075
≥41 62 6 9.7 0.2151 0.02086

The overall proportion susceptible is therefore given by the sum of the values in the
final column, i.e. 0.2821.

ii) The basic reproduction number can be estimated using the expression 1/s (equation
5.20), where s is the proportion of the population that is susceptible. Using the value
for s obtained in part i) implies that R0=1/0.28213.5.

iii) We can obtain an expression for the force of infection in terms of R0 after
rearranging either the expression R0=1+λL or R0=λL, depending on whether the age
distribution of the population is exponential or rectangular respectively. Figure S5.2,
which plots the values for pa for 5 year age groups in Bangladesh in 20052 suggests
that the age distribution was closer to being exponential than to being rectangular.

0.12

0.08
pa

0.04

0
0-4
5-9

15-19
20-24
25-29
30-34

40-44
45-49
50-54
55-59

65-69
70-74
75-79
80-84

90-94
95-99
10-14

35-39

60-64

85-89

100+

Age group (years)

Figure S5.2: Proportion of individuals in different age groups in Bangladesh in 2005
(pa).

Rearranging the expression R0=1+λL gives the following expression for λ:

R0  1
λ
L
16 | SOLUTIONS TO EXERCISES
Substituting for L=65 years and the value for R0 obtained in part ii) implies that the
force of infection equals:
3.5  1
λ  4% per year
65

1
iv) Using the relationship A  (equation 5.10) and substituting for L=65 years
λ  1/ L
and the value for λ obtained in part iii) implies that the average age at infection is
approximately:
1
A  18 years
0.04  1/ 65

Note that the values for A and λ are much smaller than those obtained in part a). This
follows largely from the fact that estimates in part a) did not account for the age
distribution of the population.

c) The following table summarizes the estimates for the average annual risk of infection
calculated using the expression λ  1 sa
1/ a
, where a was taken as the midpoint of the
age group for the corresponding data point. The force of infection was calculated using
the result rate-ln(1-risk) (see page 111).

Age % Average annual risk Average annual force

group negative of infection, of infection
(years) calculated using
λ  1 sa
1/ a

1-5 78.7 0.0767 0.0798

6-10 47.5 0.0889 0.0931
11-15 33.3 0.0811 0.0846
16-20 22.6 0.0793 0.0826
21-25 18.1 0.0716 0.0743
26-30 16.4 0.0625 0.0646
31-35 19.0 0.0491 0.0503
36-40 11.7 0.0549 0.0565
≥41 9.7 0.0415 0.0424

sa 1
We can also use the equation λa  1  . However, when substituting sa and sa+1
sa
into this equation, we obtain the risk of infection between age band a and age band
a+1. Since each age band is of width 5 years, this infection risk is equivalent to a five
year risk. We can convert this five year risk into an annual risk by adapting the logic
described in section 5.2.2.1.4, which leads to the following equation for the average
annual risk in age group a:
λ  1  (1  λa )1/ 5
The force of infection is then calculated using the result rate-ln(1-risk) (see page 111).

The following table summarizes the estimates obtained using this approach:
 CHAPTER 5: AGE PATTERNS | 17

5 year risk of infection, Average annual risk

Age
group calculated using: of infection,
Average annual
(years) s calculated using:
% negative λa  1  a 1 force of
sa λ  1  (1  λa )1/ a infection
1-5 78.7 0.3964 0.0961 0.0798
6-10 47.5 0.2989 0.0686 0.0931
11-15 33.3 0.3213 0.0746 0.0846
16-20 22.6 0.1991 0.0434 0.0826
21-25 18.1 0.0939 0.0195 0.0743
26-30 16.4 -0.1585 -0.0299 0.0646
31-35 19.0 0.3842 0.0924 0.0503
36-40 11.7 0.1709 0.0368 0.0565
≥41 9.7 - - 0.0424

The estimates obtained using both approaches suggest that the force of infection for
adults is lower than that for children, e.g. >8% per year for those aged <20 years and
<8% per year for those aged >20 years. However, the estimates for adults that are
based on the equation λ  1 λa
1/ a
are difficult to interpret, since the proportion of 31-
35 year olds who are seronegative is smaller than that for 26-30 year olds, which leads
to the (unrealistic) estimate that the risk of infection was negative between the ages 21-
25 and 26-30 year olds.

d) Figure S5.3 shows a plot of –ln(observed proportion seronegative) against the age
midpoints for the data from Bangladesh. These figures also clearly highlight the fact
that the datapoint for individuals aged 31-35 years is an outlier.

2.5 2.5

2.0 2.0
seronegative)
-ln(proportion

1.5 1.5

1.0 1.0

0.5 0.5

0.0 0.0
0 10 20 30 40 50 60 0 10 20 30 40 50 60

Age (years)
Figure S5.3: Plot –ln(observed proportion seronegative) against the age midpoints for
the data from Bangladesh in Nessa et al1, with different lines drawn by eye through the
data points for individuals aged <20years (left-hand figure) and for those aged <15
years (right-hand figure).

As shown in the left-hand figure, the gradient of the line through the points for
individuals aged <20 years is steeper than that through the points for individuals aged
>20 years, suggesting that the force of infection is greater for those aged <20 years
than for those aged >20 years.
18 | SOLUTIONS TO EXERCISES

However, based on these plots, we cannot conclude that the force of infection changes
at age 20 years, since, as shown in the right-hand figure, we can also draw a straight
line through the points for individuals aged over 15 years, which would imply that the
force of infection changes at about this age. Ultimately, the age at which the force of
infection is assumed to change needs to be biologically plausible, e.g. consistent with
changes in behaviour, possible exposure to the infection, contact patterns etc at ages
15 or 20 years.

The gradient of the line through the points for individuals aged <15 years is about
1.6/150.11 per year, suggesting that the force of infection in this age group is about
11% per year. The gradient of the line through the points for individuals aged >15
years is about 0.7/350.02 per year, suggesting that the force of infection in this age
group is about 2% per year.

5.3 The following figure shows a plot of –ln(observed proportion seropositive) for the
mumps data in section 5.2.3.2.2. The gradient of the line through the points for
individuals aged <13 years is steeper than that through the points for individuals aged
>13 years (3/130.23 per year vs 1/350.03 per year). This suggests that the force of
infection is also greater for those aged <13 years than for those aged >13 years (23%
vs 3% per year respectively).

5.0

4.0
seronegative)
-ln(proportion

3.0

2.0

1.0

0.0
0 10 20 30 40 50 60

Age (years)
Figure S5.4: Plot of –ln(observed proportion seronegative) for the mumps data in
section 5.2.3.2.2.
 CHAPTER 5: AGE PATTERNS | 19

5.4 i) One informal argument that is sometimes used to obtain this result is that, for
1
realistic values of L, 1/L is small, in comparison with λ, and therefore must be
λ  1/ L
1
approximately equal to .
λ

We can also apply a formal mathematical argument, which uses the result that, for
1
small values of x (i.e. values that are close to zero), the expression is
1 x
approximately equal to 1  x (see proof at the end of the solution to this question).

This argument is as follows:

1 1
We begin by noting that the equation A  can be rewritten in the form
λ  1/ L 1 x
as follows:
 
1  1 

A S5.1
λ 1 
 1 
 λL 
For realistic values of the average life expectancy and for large values of the force of
1 1
infection, is close to zero, and so, according to the result  1  x , the term in
λL 1 x
1
brackets in equation S5.1 is approximately equal to 1 . Substituting this
λL
approximation into equation S5.1 leads to the following:

1 1  1 1
A 1    2
λ λL  λ λ L

If both the force of infection and the life expectancy are sufficiently large, then the
1 1
second term in this equation is negligible (i.e.  0 ) and so A  .
λL
2
λ

1  1  (1  λL)e  λL 
ii) To show that the expression A    approximates to 1/λ, we begin
λ  1  e  λL 
by observing that, for sufficiently large values for the life expectancy and the force of
infection, e-λL is close to zero. Using the result that for small values of x (i.e. values that
1 1
are close to zero),  1  x , we see that  1  e  λL . Substituting this
1 x 1  e  λL
approximation into the equation for A, we obtain the following result:
1  1  (1  λL)e  λL  (1  (1  λL)e  λL )(1  e  λL )
A   
λ  1  e  λL  λ
20 | SOLUTIONS TO EXERCISES
This equation simplifies to the following:
1  e 2 λL (1  λL)  λLe  λL
A
λ
If the force of infection is sufficiently large and for realistic values for the life
1
expectancy, both e 2 λL (1  λL)  0 and λLe  λL  0 , which implies that A 
λ

1
Proof of the result  1  x for small values of x
1 x
1
This result can be derived by using the fact that an expression of the form can
1 x
be written using the following Binomial expansion:

1 ( 1)( 1  1)x 2 ( 1)( 1  1)( 1  2)x 3

 (1  x ) 1  1  x   
1 x 2! 3!
( 1)( 1  1)( 1  2)..( 1  n )x n 1
 
(n  1)!
 1  x  x 2  x 3    ( 1) n 1 x n 1  

For small values of x, terms in x2, x3 etc (known as “higher order terms”) are small and
1
can be ignored. Consequently  1 x .
1 x

1
The result that  1  x follows after repeating the above argument but replacing
1 x
 x for x.

5.5 The following figure shows that the proportion of individuals who had hookworm
ova in their stools (Sa/Na) increases with age and then reaches a plateau or (plausibly)
decreases with increasing age.
1
0.9
Proportion with

0.8
hookworm ova

0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 10 20 30 40 50 60 70 80

Age (years)
 CHAPTER 5: AGE PATTERNS | 21

We might therefore use a reversible model to describe the data, which assumes that
the age-specific proportion infected eventually reaches a plateau with increasing age.
Alternatively, a 2-stage model or a compound catalytic model might be appropriate,
since these assume that the proportion positive peaks before subsequently decreasing
with increasing age. In fact, the authors of 3 used a compound model to analyse the
data.

5.6 a) Assuming that maternal immunity is lost at a constant rate, μ, the rate of change
in the proportion of individuals who have maternal immunity (m(a)) and the proportion
who are susceptible (s(a)) are given by the following equations:
dm(a )
  μm(a )
da
ds(a )
 μm(a )  λs(a )
da

dQ(t )
The differential equation for m(a) is of the form  kQ(t ) and can be solved to
dt
give the following (see section 3.5.1):
m(a)  m(0)e  μa

where m(0) is the proportion of inewborns who have maternal immunity. Since all
individuals are assumed to be born with maternal immunity, m(0)=1, and so,
substituting for m(0)=1 into the above equation gives m(a)=e-μa.

Substituting for m(a)=e-μa into the differential equation for s(a), we obtain the following
equation:
ds(a)
 μe  μa  λs(a)
da
which can be rewritten as follows:
ds(a)
 λs(a)  μe  μa
da

This equation can be solved using the technique of “integrating factors” by following the
steps below:

Step 1. Multiply both sides of the equation by eλa, to obtain the following:
ds(a) λa
e  λs(a)e λa  μe  μa e λa
da

Note that, according to the rules of differentiation (section B.5), the left-hand side of this
equation is equivalent to the derivative of s(a)e λa with respect to a, and so the
equation can be rewritten as follows:

d S5.2
(s(a)e λa )  μe  μa e λa
da
22 | SOLUTIONS TO EXERCISES
Step 2. We now integrate both sides of equation S5.2 between 0 and a to obtain the
following:
a a
d
0 da (s(a)e )da  0 μe e da
λa  μa λa
S5.3

Since integration is the converse of differentiation, the left-hand side equation S5.3
simplifies to:
s(a)e  λa a
0  s(a)e λa  s(0)

However, s(0)=0 (since no individuals are assumed to be susceptible at birth) and

therefore the left-hand side of equation S5.3 simplifies to s(a)e λa .

By the rules of integration (section B.6) the right-hand side of equation S5.3 simplifies
to the following:

a
 μ ( λ  μ )a  μ ( λ  μ )a μ
λ μ e   λμe 
λμ
 0

Step 3. Equating the expressions obtained from integrating the left-hand and right-hand
sides of equation S5.3 leads to the following:
μ μ
s(a)e λa  e ( λ  μ )a 
λμ λμ

Dividing both sides of this equation by eλa leads to the intended result:
μ(e  μa  e  λa )
s(a ) 
λμ

ds(a)
b) Note that when s(a) is at a minimum,  0 . We can therefore obtain the age
da
at which the proportion of the population that is susceptible is at a minimum by
ds(a)
identifying the values for a for which  0.
da
Differentiating the equation for s(a) that is discussed in part a), we obtain the following:

ds(a) μ(  μe  μa  λe  λa )

da λμ

Setting this equation to zero, we see that the following must be satisfied for the
proportion susceptible to be at a minimum:
 μe  μa  λe  λa  0

Multiplying both sides of this equation by e-λa and rearranging the resulting equation,
implies that the following must hold:
 CHAPTER 5: AGE PATTERNS | 23

λ
e ( λ  μ )a 
μ
Taking the natural logs of both sides of this equation and then dividing by λ-μ leads to
the intended result that the minimum in the proportion susceptible occurs when
ln( λ / μ )
a
λμ

1 1
5.7 a) Proof of the result that A  , or equivalently, A  for
λm λ  1/ L
populations with an exponential age distribution, where m=1/L is the average
mortality rate.

Suppose that N0 individuals are born each year. Assuming a constant mortality rate of
m, the number of individuals of age a is given by the equation (see section 3.5.1):

N(a)  N0 e ma

Assuming a constant force of infection, λ, a proportion e-λa of these individuals will be

susceptible, and so the number of susceptible individuals of age a (S(a)) is obtained by
multiplying N(a) by e-λa , i.e.

S(a) = N(a)e-λa = N0e-(λ+m)a

After substituting this expression into equation 5.9, we obtain the following equation:

 
  aλN e
( λ  m ) a
aλ(a )S(a)da 0 da
A 0
 0 S5.4
 
  λN e
( λ  m ) a
λ(a )S(a )da 0 da
0 0

Using the techniques discussed in section B.6, the numerator of this equation simplifies
to the following:
 S5.5
  ( λm )a
ae ( λ  m )a
e  λN 0
0
aλN 0 e ( λm )a da  λN 0   2 

  ( λ  m) ( λ  m)  0 ( λ  m)
2

Similarly, the denominator in equation S5.5 simplifies to the following:

 S5.6
  e ( λm )a  λN 0

( λ  m )a
λN 0 e da  λN 0   
0
  ( λ  m)  0 λ  m

Substituting the right-hand sides of equations S5.5 and S5.6 into the numerator and
denominator of equation S5.4, and cancelling common terms from the numerator and
denominator leads to our intended expression for A:
λN 0
( λ  m)2 1
A 
λN 0 λm
λm
24 | SOLUTIONS TO EXERCISES

b) Proof of the result that, for populations with a rectangular age distribution,
with a life expectancy of L, and assuming random mixing,
1  1  (1  λL)e  λL 
A  
λ  1  e  λL 

Suppose that N0 individuals are born into the population each year. If the population
has a rectangular age distribution in which no individuals die until age L, then the
number of individuals of age a also equals N0.

A proportion e-λa of these individuals will be susceptible, and so the number of

susceptible individuals of age a (S(a)) is obtained by multiplying N0 by e-λa , i.e.

S(a) = N0e-λa

After substituting this expression into equation 5.9, we obtain the following equation:
 L
 aλ(a)S(a)da   aλN e
 λa
0 da
A 0

0
L
 λ(a)S(a)da  λN e
 λa S5.7
0 da
0 0

Using the techniques discussed in section B.5, the numerator of this expression
simplifies to the following:
L
L  ae  λa e  λa   Le  λL e  λL 1 
  2   λN 0   2  2 
 λa
aλ N 0 e da  λN 0 
0
 λ λ 0  λ λ λ 
N 0 (1  (1  λL )e  λL ) S5.8

λ

Similarly, the denominator in equation 5.7 simplifies to the following:

L
L  e  λa  λN 0 (1  e  λL )

 λa
λN 0 e da  λN 0   
0
  λ 0 λ
 N 0 (1  e  λL ) S5.9

Substituting the right-hand sides of equations S5.8 and S5.9 into the numerator and
denominator respectively of equation S5.7, and cancelling out the common term N0
leads to the intended result:
N 0 (1  (1  λL)e  λL ) 1  (1  λL)e  λL
A 
λN 0 (1  e  λL ) λ(1  e  λL )

5.8 For a reversible catalytic model, the differential equations for the rate of change in
the proportion susceptible and proportion currently infected is given by the following:
 CHAPTER 5: AGE PATTERNS | 25

ds(a )
  λs(a )  rs z(a )
da
dz(a )
 λs(a )  rs z(a )
da

In this model, the proportion of individuals of age a that are currently susceptible is
given by 1-proportion currently infected, i.e. s(a) = 1-z(a).

Substituting this expression for s(a) into the differential equation for z(a) leads to the
following equation:
dz(a)
 λ(1  z(a))  rs z(a)
da
 λ  ( λ  rs )z(a)
S5.10

dz(a )
At a given point on the plateau,  0 . Equating equation S5.10 to zero, leads to
da
the following result:
λ  ( λ  rs )z(a)  0

λ
After rearranging this equation, we obtain our intended result, i.e. z(a ) 
λ  rs

A
5.9 a i) With the information provided, we can use the equation A'  (equation
(1  v )
5.34) to work out the long-term average age at infection for mumps following the
introduction of vaccination. Substituting for A=4 years and v=0.6 into this equation
(assuming for now, that the vaccine efficacy is 100%), the long-term average age at
4
infection is given by A'   10 years.
(1  0.6)

Given the debate about the efficacy of the mumps component of the MMR vaccine 4-6,
it would be sensible to assume a vaccine efficacy of <100%. Assuming a vaccine
efficacy of 85%, the long-term average age at infection equals:
4 4
A'    8.2 years
(1  0.85  0.6) (1  0.51)

a ii) The long-term average force of infection λ’ can be obtained after rearranging the
1
expression A'  (equation 5.33) and substituting our estimate of A’ that we
λ'm
obtained in part i) into the resulting expression.

1
For example, the expression A'  can be rearranged to give the following
λ'm
expression for λ’:
26 | SOLUTIONS TO EXERCISES

1
λ'  m
A'

Substituting for A’ = 10 years (based on a vaccine efficacy of 100%), and m=1/60 per
year into this equation leads to the following value for λ’:

1 1 5
λ'     0.0833 per year.
10 60 60

Substituting for A’ = 8.2 years (based on a vaccine efficacy of 85%) leads to an

estimate for the average force of infection of :
1 1
λ'    0.106 per year.
8.2 60

a iii) and a iv) The proportion susceptible and the infection incidence in the long-term
can be calculated using equations 5.31 and 5.36, leading to the following values:

100% vaccine efficacy 85% vaccine efficacy

Age Proportion Average annual Proportion Average annual
(years) susceptible number of new susceptible number of new
(s(a)’= infections per (s(a)’= infections per
(1  v )e  λ'a 100,000 (1  v )e  λ'a 100,000
( λ' s(a)'100,000 ) ( λ' s(a)'100,000 )
15 0.115 955 0.1 1060
25 0.05 415 0.035 368
35 0.022 180 0.012 128

b) As shown by the calculations below, the average annual number of mumps

infections per 100,000 population among 15-35 year olds in the long-term following the
introduction of vaccination is somewhat higher than that before the introduction of
vaccination. You might therefore advise the government to aim to attain a coverage
which is much higher than 60% (e.g. 95%) and to proceed with caution when
introducing MMR vaccination if it thinks that a coverage of only 60% can be achieved.

It might want to consider having a catch-up campaign covering the birth cohorts at
greatest risk, and monitor the age-specific proportion susceptible in the population
through seroprevalence surveys. Most importantly, before proceeding, it should also
consider the effect that 60% coverage of MMR vaccination would have on the burden
of measles, rubella and Congenital Rubella Syndrome.

Calculations of the number of mumps infections per 100,000 population before

the introduction of vaccination:
For these calculations, we first need to estimate the force of infection that is predicted
1
in the absence of vaccination. Rearranging the equation A  (equation 5.10) we
λm
 CHAPTER 5: AGE PATTERNS | 27

obtain the following equation for the force of infection in the absence of vaccination:
1
λ m. Substituting for A=4 years (the average age at infection before the
A
introduction of vaccination) and m=1/60 per year into this equation, we obtain the
following for λ:
1 1
λ   0.233 per year.
4 60

Using this estimate for the force of infection, we obtain the following values for the
proportion susceptible and the average annual numbers of infections per 100,000
population in different age groups before the introduction of vaccination:

Age (years) Proportion Average annual number of

susceptible new infections per 100,000
(s(a)= e  λa ) ( λs(a)  100000 )
15 0.03 705
25 0.003 68
35 (0) 7

c) If the herd immunity effects of vaccination are not accounted for, the proportion
susceptible and the number of infections per 100,000 population in the long-term
following the introduction of vaccination would be given by the values calculated in part
b) but multiplied by (1-v), where v is the proportion of individuals that are effectively
vaccinated. v is given by the expression vaccine coverage×vaccine efficacy. The
values obtained assuming that the vaccine efficacy is 100% and 85% are provided
below. These show that the static model greatly underestimates the long-term numbers
of mumps infections per 100,000 population in 15-35 year olds following the
introduction of vaccination.

100% vaccine efficacy 85% vaccine efficacy

Age Proportion Average annual Proportion Average annual
(years) susceptible number of new susceptible number of new
infections per 100,000 infections per 100,000
15 0.012 282 0.015 345
25 0.001 27 0.001 33
35 0 3 0 3

5.10 a) Multiplying both sides of equation 5.29 by (1-v), we obtain the following:
R0 (1 v )  λ' L  1

Subtracting both sides of this equation by 1 and dividing by L, we obtain the following
equation:
R0 (1  v )  1
 λ' S5.11
L
28 | SOLUTIONS TO EXERCISES

R0 (1  v )  1
b) and d) Figure S5.5 compares the plot of λ'  against that of
L
λ'  λ(1 v ) .
(a) R0=7 (b) R0=12
0.10
0.16
Annual force of infection

0.08
0.12
0.06

0.08
0.04
λ'=λ(1-v)
0.04
0.02
Equation S5.7

0.00 0.00
0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0

Proportion immunized (v)

Figure S5.5: Predictions of the average annual long-term force of infection following
R0 (1  v )  1
the introduction of vaccination, calculated using λ'  (dotted line) for
L
different levels of the immunization coverage among newborns, in a A. low
transmission (R0=7) and B. a high transmission setting (R0=12). The solid line shows
the annual force of infection which would be seen if the force of infection was directly
proportional to the proportion of individuals that are protected by vaccination(v), i.e. if
λ’=λ(1-v).

c) You might have expected the force of infection as predicted by the equation
λ'  λ(1 v ) to decrease more slowly with increased vaccination coverage than that
R0 (1  v )  1
predicted by the line  λ' , since the gradient of the latter line is steeper
L
than that for λ'  λ(1  v ) .

R0 (1  v )  1
For example, recall that the gradient of the line  λ' is the factor by which
L
we multiply “v” (i.e. the “coefficient” of v). The coefficient of v in this equation, and
therefore the gradient of the line is -R0/L. Substituting for R0=1+λL (equation 5.21) into
this equation, we see that the coefficient is equal to:
R0 1  λL
 
L L
The expression for the gradient simplifies to the following:
 1
λ 
 L

In contrast, the gradient of the line λ'  λ(1  v ) is just –λ.

 CHAPTER 5: AGE PATTERNS | 29

1
Since λ  is bigger than λ, we can conclude that the gradient of the line λ'  λ(1  v )
L
will be steeper than that given by equation 5.11.

e) You should find that it is not possible to rearrange the equation to obtain an explicit
expression for λ’ in terms of R0, L and v, since the numerator has a term λ’ and the
denominator has a term e  λ'L .

Instead, we need to use iterative techniques to obtain the value for λ’ which results
from a given value for R0, v and L, as follows:

λ' L
We first rearrange the equation R0  so that we have an expression
(1  v )(1  e  λ' L )
for λ’ in terms of all the other terms in the equation. For example, we could rearrange
the equation to obtain the following:
R0 (1  v )(1  e  λ'L )
λ'  S5.12
L

If we substitute some value for λ’ (denoted by λ0' ) into the right-hand side of this
equation, then for given values for R0, L and v, we will obtain another value for λ’ (we
shall denote it by λ1' ). If we then substitute λ1' into the right-hand side of equation
5.12, we obtain another value for λ’ (we shall denote it by λ2' ). Repeating this process
several times, we eventually obtain a series of values, λ0' , λ1' , λ2' , λ3' , , and we find that
the difference between successive values of λi' becomes progressively smaller, until
the value obtained satisfies equation S5.12 (see Table S5.1). These iterations can be
set up in a spreadsheet.
30 | SOLUTIONS TO EXERCISES
Table S5.1: Illustration of how the post-vaccination force of infection at equilibrium, λ’,
λ' L
which satisfies the equation R0  may be calculated iteratively from
(1  v )(1  e  λ' L )
 λi L
R0 (1  v )(1  e )
the equation λi1  , assuming that R0=7, L=70 years and v=0.1. In
L
this instance, the average force of infection which might be expected if the vaccination
coverage among newborns is 10% is 0.090 per year.
Iteration λi (per year) R0 (1  v )(1  e i )
 λ L

number Value for

L
0 λ0  0.05 7(1  0.1)(1  e 0.0570 )
 0.2 per year.
70
1 λ1  0.2 7(1  0.1)(1  e 0.270 )
 0.0899999 per year
70
2 λ2  0.0899999 7(1  0.1)(1  e 0.089999970 )
 0.0898347 per year
70
3 λ3  0.0898328 7(1  0.1)(1  e 0.089832870 )
 0.0898328 per year
70

References

1. Nessa A, Islam MN, Tabassum S, Munshi SU, Ahmed M, Karim R.

Seroprevalence of rubella among urban and rural Bangladeshi women
emphasises the need for rubella vaccination of pre-pubertal girls. Indian J Med
Microbiol 2008; 26(1):94-95.

2. World Population Prospects: The 2008 Revision and World Urbanization

Prospects: The 2008 Revision. Population Division of the Department of
Economic and Social Affairs of the United Nations Secretariat . 2005.

3. Zhang YX. A compound catalytic model with both reversible and two-stage types
and its applications in epidemiological study. Int J Epidemiol 1987; 16(4):619-621.

4. Kim-Farley R, Bart S, Stetler H et al. Clinical mumps vaccine efficacy. Am J

Epidemiol 1985; 121(4):593-597.

5. Harling R, White JM, Ramsay ME, Macsween KF, van den BC. The effectiveness
of the mumps component of the MMR vaccine: a case control study. Vaccine
2005; 23(31):4070-4074.

6. Cohen C, White JM, Savage EJ et al. Vaccine effectiveness estimates, 2004-

2005 mumps outbreak, England. Emerg Infect Dis 2007; 13(1):12-17.
 Chapter 7 (Solutions)
How do models deal with
contact patterns?

7.1 Using a similar approach to that used in Panel 7.1, we obtain the following two
expressions for the number of new infections among adults which are attributable to
contact with children:
βoy So (t )I y (t ) S7.1

λoy (t )So (t ) S7.2

Equating expressions S7.1 and S7.2, we obtain the following:

λoy (t )So (t )  βoy So (t )I y (t )

Cancelling So(t) from both sides of this equation, we obtain the following equation:

λoy (t )  βoy I y (t ) S7.3

Similarly, we can obtain the following two expressions for the number of new infections
among adults that are attributable to contact with other adults:

βoo So (t )I o (t ) S7.4

λoo (t )So (t ) S7.5

Equating expressions S7.4 and S7.5 and cancelling So(t) from the resulting equation,
we obtain the following equation:
λoo (t )  βoo I o (t ) S7.6

Substituting our expressions for λoy (t )  βoy I y (t ) and λoo (t )  βoo I o (t ) into equation
7.3 in the book, we obtain our intended result, i.e.
λo (t)  βoy I y (t )  βooIo (t )

7.2 We will use the notation and definitions for the symbols provided on pages 183
and 184, and set βyy=2×10-4 per day, βyo=8×10-4 per day, βoy=3×10-4 per day, and
βoo=7×10-5 per day. The answers to the questions are as follows:
32 | SOLUTIONS TO EXERCISES

a) i) λyy (t )  β yy I y (t )  2 10  20  0.004 per day.

-4

ii) λyo (t )  β yoI o (t )  8 10  50  0.04 per day.

-4

iii) λy (t )  λyy (t )  λyo (t )  0.004  0.04  0.044 per day.

i) λoy (t )  βoy I y (t )  3 10  20  0.006 per day.
-4
b)
ii) λoo (t )  βoo I o (t )  7 10 -5  50  0.035 per day.
iii) λo (t )  λoy (t )  λoo (t )  0.006  0.035  0.095 per day.

 β1 0.5 β2 
7.3 Using WAIFW matrix   , and assuming that the average force of
 0.5 β2 β2 
infection among children and adults is 13% and 4% per year respectively, and that the
average numbers of infectious children and adults are 18,956 and 2,859 respectively,
then we would need to solve the following matrix equation to obtain values for β1 and
β2:
 β1 0.5 β2 18,956   0.13 
   
 0.5 β2 β2  2,859   0.04 

This equation can be written out in full as follows:

18,956β1  2,859  0.5β2  0.13 S7.7
18,956  0.5β2  2,859  β2  0.04 S7.8

Equation S7.8 simplifies to the following:

12,237β2  0.04 per year.

Dividing both side of this equation by 12,237 leads to β2  3.24  10 6 per year.

Substituting this value for β2 into equation S7.7 leads to the following:
18,956β1  2,859  0.5  3.24  106  0.13

This equation can be rearranged to give the following:

18,956β1  0.12536 S7.9

Dividing both sides of this equation by 18,956, we obtain β2  6.61 10 6 per year.
Dividing the values obtained for β1 and β2 by 365 to obtain values in units of per day
leads to the following values: β1 = 1.81×10-8 per day and β2 = 8.88×10-9 per day.

 β1 0.5 β2 
Substituting these values for β1 and β2 into the matrix   leads to the
 0.5 β2 β2 
 1.81 10 -8 4.44 10 -9 
following matrix   in units of per day. This matrix is identical
 4.44 10
-9
8.88 10 -9 
to matrix R2 that is presented in section 7.4.2.1.1.
 CHAPTER 7: HOW DO MODELS DEAL WITH CONTACT PATTERNS? | 33

7.4 a) We will use the equation Ii=λiSiD, to calculate the number of infectious persons in
age group i, where λi and Si are the force of infection and the susceptible infectious
persons in age group i respectively, and D is the duration of infectiousness (=7 days or
7/365 years). The age groups 0-1, 2-4, 5-9, 10-14 and 15-74 years will be denoted
using the subscripts 1, 2, 3, 4 and 5 respectively.

The following table summarizes the average numbers of infectious individuals

calculated for each age group:

Age group (years) 0-1 2-4 5-9 10-14 15-74

Average prevaccination 7.7 23.7 51.7 25.5 9.9
force of infection (%/year)
Average number 1,062,861 1,216,541 493,355 59,269 59,182
susceptible
(prevaccination)
Average number of I1=1,570 I2=5,529 I3=4,892 I4=290 I5=112
infectious persons (=0.077 (=0.237 (=0.517 (=0.255 (=0.099
× × × × ×
1,062,861 1,216,541 493,355 59,269 59,182
× × × × ×
7/365) 7/365) 7/365) 7/365) 7/365)

Note that when applying the equation Ii=λiSiD, the units for the force of infection in age
group i and the duration of infectiousness must be consistent: since we used an annual
force of infection, the duration of infectiousness used in the equation is also in units of
years.

b) i) After some calculations (see below) and assuming that α=1, we obtain the
following values for the β parameters for matrix 2:
β1  6.20  10 6 per year
β2  2.11 10 5 per year
β3  7.84  10 5 per year
β4  2.14  10 5 per year
β5  7.99  10 6 per year

The WAIFW matrix is therefore as follows (where the parameters are in units of per
year):
 6.20  10 6 6.20  10 6 6.20  10 6 6.20  10 6 7.99  10 6 
 
 6.20  10 6 2.11 10 5 2.11 10 5 2.11 10 5 7.99  10 6 
 6 
 6.20  10 2.11 10 5 7.84  10 5 2.14  10 5 7.99  10 6 
 6.20  10 6 2.11 10 5 2.14  10 5 7.84  10 5 3 7.99  10 6 
 6 
 7.99  10 7.99  10 6 7.99  10 6 7.99  10 6 7.99  10 6 
34 | SOLUTIONS TO EXERCISES

The equation that we need to solve to obtain these values for the β parameters is as
follows:
 β1 β1 β1 β1 β5  I1   λ1 
    
 β1 β2 β2 β2 β5  I 2   λ2 
β β2 β3 β4 β5  I 3    λ3 
 1    
 β1 β2 β4 αβ3 β5  I 4   λ4 
    
 β5 β5 β5 β5 β5  I 5   λ5 

These equations can be rewritten as follows:

β1(I1  I 2  I 3  I 4 )  β5I 5  λ1 S7.10
β1I1  β2 (I 2  I 3  I 4 )  β5I 5  λ2 S7.11
β1I1  β2I 2  β3I 3  β4I 4  β5I 5  λ3 S7.12
β1I1  β2I 2  β4I 3  αβ3I 4  β5I 5  λ4 S7.13
β5 (I1  I 2  I 3  I 4  I 5 )  λ5 S7.14

Equation S7.14 can be rearranged to give the following:

λ5
β5 
I1  I 2  I 3  I 4  I 5

Substituting for λ5=0.099 per year and for I1  I 2  I 3  I 4  I 5  12,393 into this
equation, we obtain the following value for β5:
β5  0.099 / 12,393  7.99  10 6 per year

Equation S7.10 can be rearranged to give the following expression for β1:

λ1  β5 I 5
β1 
I1  I 2  I 3  I 4

Substituting for β5  7.99  10 6 per year, λ1=0.237 per year and for the corresponding
numbers of infectious persons into this equation, we obtain β1  6.20  10 6 per year.

Equation S7.11 can be rearranged to give the following expression for β2:

λ2  β1I1  β5 I 5
β2 
I2  I3  I 4

Substituting for β5  7.99  10 6 per year, β1  6.20  10 6 per year and for the
corresponding numbers of infectious persons into this equation, we obtain
β2  2.11 10 5 per year.
 CHAPTER 7: HOW DO MODELS DEAL WITH CONTACT PATTERNS? | 35

To obtain β3 and β4, we can solve equations S7.12 and S7.13 simultaneously. To
simplify the notation, we will re-express these two equations as follows:

β3 I 3  β 4 I 4  P S7.12’
αβ3I 4  β4I 3  Q S7.13’

where P  λ3  β1I1  β2I 2  β5I 5 and Q  λ4  β1I1  β2I 2  β5I 5 . Substituting the
corresponding values for the force of infection, the β values and the numbers of
infectious persons into these equations, we see that P=0.3895 and Q=0.1265 (to 4
decimal places).

Multiplying equations S7.12’ and S7.13’ by I3 and I4 respectively, we obtain the

following:
β I 2  β I I  PI
3 3 4 4 3 3
S7.15

αβ I  β4I 3I 4  QI 4
2
3 4
S7.16

Subtracting equation S7.16 from equation S7.15, we obtain the following equation.

β3 (I 32  αI 42 )  PI 3  QI 4
After rearranging, we obtain the following expression for β3:
PI 3  QI 4
β3 
I 32  αI 42
Substituting for P, Q, I3 and I4 into this equation leads to the result that
β3  7.84  10 5 per year.

Rearranging equation S7.12’, we obtain the following expression for β4:

P  β3 I 3
β4 
I4

Substituting for P, I3, I4 and β3 into this equation, we obtain β4  2.14  10 5 per year.

ii) To calculate R0, we first need to calculate the Next Generation Matrix, using the
number of infectious individuals among those in age group i resulting from individuals
in age group j, as obtained using the expression N i βij D . Here, Ni is the number of
individuals in age group i, and βij is the rate at which specific susceptible individuals in
age group i come into effective contact with specific infectious individuals in age group j
and D (=7 days) is the duration of infectiousness.

Ni is given by the width of age group i multiplied by 650,000 (the number of individuals
in each single year age category), as follows:
36 | SOLUTIONS TO EXERCISES

Age group 0-1* 2-4 5-9 10-14 15-74

(years)
Ni 1,137,500 1,950,000 3,250,000 3,250,000 39,000,000
(= (= (= (= (=
1.75×650,000) 3×650,000) 5×650,000) 5×650,000) 60×650,000)
* Note that calculations for 0-1 year olds assume that individuals have maternal immunity for the first 3 months of life.

The Next Generation Matrix should resemble the following:

Age grp
(yrs) 0-1 2-4 5-9 10-14 15-74
0-1  0.135 0.135 0.135 0.135 0.174 
 
2-4  0.232 0.790 0.790 0.79 0.299 
5-9  0.386 1.317 4.884 1.335 0.498 
 
10-14  0.386 1.317 1.335 4.884 0.498 
15-74  5.975 5.975 5.975 5.975 5.975 

Adapting the model files provided for calculating R0, we obtain a value for R0 of 9.1.

iii) The herd immunity threshold for this matrix is 1-1/R0 or 100×(1-1/9.1) 89%

c) Increasing the size of α increases the amount of contact between 10-14 year olds
and leads to an increase in the size of R0 and the herd immunity threshold. For
example, if α=2, R0 equals 11.4 and the herd immunity threshold is about 91%.
Consequently, the greater the amount of contact between 10-14 year olds, the more
difficult it is to control transmission through vaccination.

d) The following summarizes the values for Rn that are obtained for different values for
α:

α 1 1.25 1.5 1.75 2

Rn 0.96 0.99 1.04 1.11 1.2

In general, the greater the value for α, the greater the value for Rn. The values
obtained for Rn are generally consistent with those in Figure 7.17.

 1.784 0.188 
7.5 a) The Next Generation Matrix is given by the following:   , which
 0 .383 0 .766 
results in a value for R0 of 1.85.

b) To answer this question, we calculate the reproduction number using the following
Next Generation Matrix
 β yy S y D β yoS y D 
 
 β S D β S D
 oy o oo o 
 CHAPTER 7: HOW DO MODELS DEAL WITH CONTACT PATTERNS? | 37

where Sy and So are the numbers of susceptible children and adults (defined as those
aged <15 and ≥15 years respectively), calculated after incorporating the appropriate
vaccination coverage for each vaccination scenario. The β parameters are as defined
in the text, and D is the duration of infectiousness (2 days):

The following table summarizes the number of susceptible individuals for each
vaccination scenario, the Next Generation Matrix and the values for the reproduction
number:

Individuals Number of susceptible Next Generation Reproduction

targeted Children (Sy) Adults (So) Matrix number
No vaccination  1.784 0.188 
  1.85
2639 5361  0.383 0.766  (=R0)
Children only 139 5361  0.094 0.010 
 
(=2639-2500) (=5361-0)  0.383 0.766  0.77
Adults only 2639 2861  1.784 0.188 
 
(=2639-0) (=5361-2500)  0.204 0.409  1.81
Same 1814  1.226 0.130 
3686  
0.526 
proportion of
children and (=2639×(1-  0.263
0.3125)) (=5361×(1-0.3125))
adults*
1.27
Equal  0.939 0.099 
 
0.294 0.587 
numbers of 1389 4111
children and 
adults (=2639-1250) (=5361-1250) 1.01
*The proportion of children and adults that need to be targeted with this strategy equals
the number of vaccine doses available ÷ population size = 2500/(2639+5361)=31.25%

The smallest value for the reproduction number is associated with the strategy of
vaccinating only children, which suggests that, of the four strategies, this approach may
be the best way of distributing the vaccine stocks. However, we would also need to
account for the severity of influenza and the mortality rates in other age groups before
making the final decision about which vaccination strategy should be adopted.
 Basic maths (Solutions)

B1 a) log10 1,000,000  6
b) log10 10,000  4
c) log10 0.001  3
d) logp r  q
e) 10  100,000
5

f) q p  r

dy
B2 a)  28
dt

dy
b)  10te 5t  25t 2 e 5t
dt

dy
c)  3e t
dt

dy
d)  8t 3  48t 7
dt

dy
e)  7t 8  2t
dt

B3 In the following expressions, c is some (unknown) constant.

t3
 t dt  c
2
a)
3

 4t dt  t 4  c
3
b)

1 1
c) t 5
dt  
4t 4
c
 BASIC MATHS | 39

20
20  e rt  1  e 20r

rt
d) e dt    
0
 r 0 r


e) 70dt  70t  c

 8 2  x   7   2 6  x   2 
B4 a) i)       ii)      
 2 13  y   3   3 4  y   3 

9x  y  1
b i)
3 x  2y  4

13x  9 y  6
b ii)
4 x  4y  2

B5 a) The determinant is given by: 5  13  2  10  65  20  45

4 0 0 0 0 4
b) The determinant is given by: 3 1 0  3  ( 4  0)  0  12
2 1 1 0 0 2

 a b  x   0 
c) To find the conditions under which the equation       holds for some
c d  y   0 
non-zero values of x and y, we first write out this equation using simultaneous equation
notation:
ax  b y  0
cx  dy  0

Multiplying the first equation by c and the second equation by a we obtain the following
two equations:
acx  b cy  0
acx  ady  0

Subtracting the first equation from the second equation, we obtain the following result:

ady  bcy  0

Factoring out y from this equation, we obtain the following equation:

(ad  bc)y  0

Since y is non-zero, and since this equation equals zero, it follows that the term in
brackets must equal zero, i.e. ad - bc=0.
 40 | SOLUTIONS TO EXERCISES

a b 
Since ad – bc equals the determinant of the matrix   , we have obtained our
c d 
intended result.

5 2
B6 a) To obtain the eigenvalues of   , we need to find the values of ρ for
10 13 
5  ρ 2 
which the determinant of the matrix   is zero.
 10 13  ρ 

The determinant of this matrix is given by the expression (5  ρ)(13  ρ)  20 .

This expression can be rearranged as follows:
(5  ρ)(13  ρ)  20  65  18ρ  ρ 2  20
 ρ 2  18ρ  45
 ( ρ  3)( ρ  15)

Consequently, for the determinant to equal zero, the following equation must hold:
( ρ  3)( ρ  15)  0
This holds when either ρ=3 or ρ=15 , which suggests that the eigenvalues of the matrix
5 2
  are 3 and 15.
10 13 

3 1 0
 
b) To find the eigenvalues of  0 4 0  , we need to find the values of ρ for which the
 0 2 1
 
3  ρ 1 0 
 
determinant of the matrix  0 4ρ 0  is zero, i.e. for which the following
 0 1  ρ 
 2
equation holds:

4ρ 0 0 0 0 4
(3  ρ ) 1 0 0
2 1 ρ 1 ρ 0 0 1

This equation simplifies to the following:

(3  ρ)(4  ρ)(1 ρ)  0

For this equation to hold, ρ (i.e. the eigenvalues) must equal 3, 4 or 1.

View publication stats