MATHEMATICAL MODELS

OF INFECTIOUS DISEASES
Pej Rohani & John Drake

Odum School of Ecology

University of Georgia
GLOBAL CAUSES OF
MORTALITY
Measles & pertussis account for
~300,000 and ~200,000 annual deaths

Total mortality
Infant mortality
MULTIFACETED APPROACH TO
UNDERSTANDING INFECTIOUS DISEASES
Medicine But these approaches don’t
address important questions at
population level ...
Microbiology

Genomics Immunology
Vaccines & Drugs
EMERGING
PATHOGENS
SCHOOL
OUTBREAK

# Boys confined to bed

Boarding School, England
Jan 1978

Raises numerous questions:

• What is etiological 0 2 4 6
Day
8 10 12

• agent?
• Is it novel?
Is a vaccine available?
MODELING QUESTIONS
I. BASICS

Why does
epidemic
What does turn over?
growth rate
# Boys confined to bed

tell us?

Why did it
What go
determines extinct?
0 2 4 6 8 10 12
invasion? Day
MODELING QUESTIONS II.
CONTROL IMPLICATIONS
How to Random or
prevent aimed at age/
spatial When is it core group?
spread? best to
implement
controls? Drugs, Vaccines
Is it or other control
evolving? measures?

How
drastic
?

Probability of
How to prevent
invasion or
invasion/
extinction
reinvasion?
WHAT IS A
MODEL?
Different types of models:
A mathematical/computational model is an abstract
model that uses mathematical language to describe the
behaviour of a system
A Statistical model attempts to describe relationships between
observed quantities and independent variables

Developing a mechanistic model is different from statistical

analyses of data
ABSTRACTIO
N
Purpose Components

Reality Conceptualization
Abstraction

Interpretation

Assumptions Limitation Validation

s
WHAT’S A ‘GOOD’
MODEL?
Choice of model depends crucially on focal question and available
data (hammer & chisel or pneumatic drill?)

Use model principally

for understanding
nature making
predictions
JUDGING A
MODEL…
Three fundamental features of models, often opposing forces:
Accuracy
Capture observed patterns (qualitative or quantitative?) and make
predictions
Increases with model complexity

Transparency
Ability to understand model components
Decreases with model complexity

Flexibility
How easily can model be adapted to new scenarios?
Decreases with model complexity
REALISM VS
TRANSPARENCY
Transparency

Multi-Scale

Network
Resolution

Agent-Based
“Realism”

Structured
Homogeneous

Solution tools
‘HOW’ DO YOU
MODEL?
Analytical Models
Concentrate on problems that can be expressed and
analysed fully using analytical approaches.

Problem-­‐based Models
Construct most “appropriate” model and use
whatever combination of methods for analysis and
prediction.

Ready-­‐Made
Software
ModelMaker
www.modelkinetix.co
m/modelmaker/model
maker.html
GLOBAL
SIMULATORS
RESOURCE
MATERIALS
Keeling & Rohani (2008)

Vynnycky & White (2010)

Anderson & May (1991)

Otto & Day (2007)

MATHEMATICAL MODELLING OF INFECTIOUS
DISEASES

Objective 1: Setting up simple models Objective 4: Heterogeneities

Different transmission modes Risk structure
Basic Reproduction Ratio (R₀), Simple Age-structured transmission
Epidemics, Invasion threshold & Realistic pathogenesis
extinction
Seasonality
Stability analysis
Objective 5: Sensitivity &
Objective 2: Control Variability Stochastic
Infection management
implementation Parameter
uncertainty
Objective 3: Statistical
estimation R0 and other
parameters
STEPS IN DEVELOPING A
MODEL
Formulate problem/objectives

Conceptual model diagram

Dynamic equations

Computer code
THE SIMPLEST
MODELS
Let’s develop a model for Boarding School influenza outbreak
Some important choices need to be made at outset
1. What do we want to keep track of?
Amount of virus in population?
Antibody titre of everyone in population (school)?
Cities in which infected people have been found?
CATEGORISING
INDIVIDUALS
Healthy Incubating Diseased Clinical status

Susceptible Exposed/latent Infectious Recovered/Immune Infection status

Incubating Infectious, Infectious, Infectious,

latent
symptomatic quarantined
THE SIMPLEST
MODELS
Pragmatic choice: categorise individuals in population according to their
infection status, eg:

Susceptible
Infectious
Recovered/Immune
These are our
“system variables”
THE SIMPLEST
MODELS
2. What model structure?
-- Determined by pathogen biology

Susceptible Infectious SI – signifies fatal infection

Susceptible Infectious Recovered SIR – recovery after infection

Susceptible Exposed Infectious Recovered SEIR – latency

Susceptible Infectious SIS – no immunity elicited

THE SIMPLEST
MODELS
2. What model structure?
-- Determined by pathogen biology

Carrier

Susceptible Infectious Recovered SIR – with carriers

Susceptible Exposed Infectious Recovered

Infectious Exposed Susceptible Vectored transmission

THE SIMPLEST
MODELS
What model structure?
Depends on what do we know about the pathogen (eg, influenza) It’s
directly transmitted (aerosol)
An acute infection
Lifelong immunity (to that strain)

Transmission Recovery

Susceptible Infectious Recovered

THE SIMPLEST
MODELS
Transmission Recovery

Susceptible Infectious Recovered

Flow between classes/compartments determined by details of host population

structure and pathogen biology

Host population size

Contact rates
These are our “parameters”
Pathogen infectivity
THE SIMPLEST
MODELS
3. Deterministic or
stochastic?

Deterministic 50 independent
stochastic
realizations

On average, stochastic simulations identical to deterministic

predictions, though individual realizations may be quite
different
REALISM VS
TRANSPARENCY
Multi-Scale

Network
Resolution

Agent-Based

Structured
Homogeneous

Solution tools
THE SIMPLEST
MODELS
We’ve settled on a deterministic SIR model – now what?

How do we write down some equations to describe spread of ‘flu in this

population?
Assign each system variable a unique Roman letter, eg:
Susceptible, S (proportion) or X (number)
Infectious, I (proportion) or Y (number)
Recovered/Immune, R (proportion) or Z (number)
Assign parameters a unique (typically Greek) letter, eg:
Contact rate, 
Pathogen infectivity, 
VERY IMPORTANT!
NOTHING SPECIAL ABOUT MY CHOICE OF NOTATION –
USE OF PARTICULAR LETTERS HIGHLY IDIOSYNCRATIC

OTHER AUTHORS MAY USE DIFFERENT LETTERS TO

DENOTE SAME VARIABLES OR PARAMETERS.

YOU CANNOT AUTOMATICALLY ASSUME THAT β IN TWO

DIFFERENT PAPERS MEANS THE SAME THING!
3. MODEL
EQUATIONS
BATH TUB
EXAMPLE
Water inflow rate,
I(t)

Let W(t) be amount of water in bathtub

(ml)

Need a dynamic equation that tells us

how W(t) will change through time
Water outflow
rate, O(t)
Consider a small time interval, δt

Then,

W(t+ δt) = W(t) + Inflow rate × elapsed time - Outflow rate × elapsed time
BATH TUB
EXAMPLE
Water inflow rate,
I(t)
W (t + 6t) = W (t) + I ⇥ 6t — O
⇥ 6t

Rearrange
W (t + 6t) — W
=I —
(t) Water outflow
O rate, O(t)
6t
Left hand side is a difference quotient for derivative of W with
respect to time

Let δt → 0 dW
=I —
dt
O
MANY BATHTUBS =
COMPARTMENT
MODELS
MODEL
EQUATIONS
If we knew X t and Yt, could we predict Xt+  t and Yt+t, where t is
some (very short) time later?

X t+t = X t – ( t) X t Yt /N

Yt+t = Yt + ( t) X t Yt /N - ( t) Yt

And
Z t+t = Z t + ( t) Yt

 is probability of transmission given contact

 is contact rate
BASIC
QUESTIONS?
=

X t+t = X t – ( t) X t Yt /N
Yt+t = Yt + ( t) X t Yt /N - ( t)
Yt Z t+t = Z t + ( t) Yt

Average infectious period given by 1/

[why?]
MEAN LIFE TIME
CALCULATION
Consider recovery of a single infectious individual I(t) = e—çt
Z 1 c
1= ce— dt =
çt ц
o

Hence, probability density function is e-t

1
=
ц
Z 1
x fmean
For a random variable x, with probability density function f(x), the (x)dx is given
by 0
AN ODE
MODEL
Consider the equation describing Susceptible dynamics
Xt+  t = X t – ( t) X t Yt/N

Re-write as
Xt+  t - X t = - ( t) X t Yt/N
(X t+t – X t )/ t =  X t Yt /N

By fundamental theorem of calculus, as t → 0,

dX/dt = -  X Y/N
AN ODE SIR
MODEL
dX Y
dt = —βX N
dY Y
= βX —
dt цY N
dZ
=
dt
цY
o By definition, X+Y+Z = N
o These equations describe rates of change in state variables
o Parameters   represent instantaneous rates
AN ODE SIR
MODEL
dX
In my lectures (as in K&R 2008),
Y
variables = X,—βX Y & Z refer to the
numbers dt of individuals in each class.
Variables N S, I, & R refer to the
proportions dY of the Y population in
= βX each— class
цY dt N
dZ
Parameters   represent instantaneous=
o These equations describe rates of change in state variables

o цY rates

dt
AN ODE SIR
MODEL dX Y
dt = —βX N
dY Y
= βX —
dt цY N
dZ
=
dt
цY
Important to notice: transmission rate is assumed to depend on frequency
of infecteds in population (Y/N). Hence, this is frequency-
dependent transmission
SIMULATING
EPIDEMICS Transmission rate,
β = 10 yr-1
Transmission rate,
β = 50 yr-1
Transmission rate,
β = 100 yr-1
1
Transmission rate,
β = 200 yr-1

Infected
Susceptible

Infectious period (1/γ) = 3 days 0.5

0.1
0 0
0 0.5 1 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2

 = 10; 1/ = 10d  = 50; 1/ = 10d  = 1e+02; 1/ = 10d  = 2e+02; 1/ = 10d
1 1 1 0.8
0.9
0.8 0.8 0.8
0.6
0.7

Infected
Susceptible

0.6 0.6 0.6

0.5
Infectious period (1/γ) = 10 days 0.4
0.4
0.4
0.4
0.3
0.2 0.2
0.2 0.2
0.1
0 0 0 0
0 0.5 1 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2

 = 10; 1/ = 20d  = 50; 1/ = 20d  = 1e+02; 1/ = 20d  = 2e+02; 1/ = 20d
1 1 1

Infected
Susceptible

0.5 0.5
Infectious period (1/γ) = 20 days 0.5 0.4 0.5 0.5

0.2

0 0 0 0 0 0
0 0.5 1 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2

 = 10; 1/ = 30d  = 50; 1/ = 30d  = 1e+02; 1/ = 30d  = 2e+02; 1/ = 30d
1 1 1

0.8

Infected
Susceptible

0.6 0.5 0.5 0.5

Infectious period (1/γ) = 30 days 0.4
0.4 0.5 0.5
0.3
0.2
0.2
0.1
0 0 0 0 0 0
0 0.5 1 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2 0 0.05 0.1 0.15 0.2

Time (years) Time (years) Time (years) Time (years)

MODEL
DYNAMICS
As parameters are varied, model predicts different
outcomes

Can we anticipate trajectories without resorting to

numerical integration?

Question: under what conditions will an infectious

disease invade a system?
THE INVASION
THRESHOLD
When can an infectious disease invade a population?

Initial conditions: X)0( = N, Y)0( = 1,

Z)0( = 0 Invasion only if dY/dt > 0
ie, βXY/N – γY > 0 → Y)βX/N - γ( > 0

If and only if X/N > γ/β

Since X=N, requires 1>
γ/β Or β/γ > 1
K
e
BASIC REPRODUCTIVE
RATIO, R0
Ratio β/γ gives number of cases before infected individual recovers
Universally referred to as R0 or Basic Reproductive Ratio

Definition: Number of secondary cases generated by a typical infected in an

entirely susceptible population

R₀ < 1 R₀ =4
No invasion Successful invasion
R0 AND MODEL
PARAMETERS
Transmission Rate (β, yr⁻¹)

R0 < 1

Infectious period (1/ days)

 R0 ESTIMATES OF
Hepatitis C
R0
Seasonal Influenza

1918 Influenza

Ebola
SARS

Phocine Distemper

HIV (MSM)

HIV (FSW)

Mumps

Pertussis
THE DEATH OF AN
EPIDEMIC
In SIR equations, let’s divide equation for dX/dt by dZ/dt:
dX/dZ = - ( X Y/N)/(Y)
= - R0 X/N

Integrate with respect to Z

X(t) = X(0) e –Z(t) R0/N

When epidemic is over, by definition, we have X(∞),Y(∞) (=0), and

Z(∞)

X(∞) = N – Z(∞) = X(0) e –Z(∞) R0/N

THE DEATH OF AN
EPIDEMIC
So, N – Z(∞) - X(0) e –Z(∞) R0/N = 0
Solve this numerically (‘transcendental’ equation)

Epidemic dies out because

there are too few infectives,
not because of too few
susceptibles

Kermack & McKendrick (1927)

SIMPLE
EPIDEMICS
β 1/Ɣ R₀

“Measles” 886 /yr 0.019 yr 17

“Influenza” 180 /yr 0.011 yr 2

“Chickenpox” 315 /yr 0.022 yr 7

“Rubella” 200 /yr 0.025 yr 5

FREQUENCY- OR DENSITY-
DEPENDENT TRANSMISSION?

Assumed contact rate, , constant: ‘mixing’ is independent of population

size: frequency-dependent transmission. Reasonable?
If we assume contact rate to be N (increases with ‘crowding’), then
transmission rate is
dX/dt = -XY
Called density-dependent transmission

N


Contact Rate
Population Size
DOES IT
MATTER?
Again, pathogen invasion if dY/dt > 0
If initially everyone susceptible (X=N),
NY – Y>0  Y(N - ) > 0
In this case, we define R0  N so
need R0>1

Hence, for any particular β and γ, there’s now a threshold population

density required for invasion
INCORPORATING
VIRULENCE
Assume infectious individuals die at rate α

dY
= . . . — цY —
dt
↵Y
 TRANSMISSION &
Rβ=0.0426,
Density
0 Dependent
γ=24, α=18, μ=0.02
Frequency
β=426, γ=24, α=18, μ=0.02
D ependent
NT = 1000 No invasion threshold
 = 0.0426; 1/ = 15d  = 426; 1/ = 15d
 = 426; 1/ = 15d
10000 10000
10000

11

Size
Population Size

10 10

Size 9

0
5000

Populati
5000

R
R
5000 0

on
5000 0
Infected
NT detcefnI

1 Population
0
0 0 0 100 200 300 400 500 600 700 800 900 1000
0 100 200 300 400 500 600 700 800 900 1000
Time (years) Time (years)

0
00 100100 200200 300300 400400 500500 600600 700700 800800 900900 FD
0
00
transmission → pathogen can wipe out
100100 200200 300300 400400 500500 600600 700700 800800
10001000
Time (years)
900900 10001000
Time host(years)
WHAT SHOULD WE
DO?
If population size doesn’t change, FD & DD equivalent (βFD = N x
βDD)
Otherwise:
Frequency-dependence generally more appropriate in large
populations with heterogenous mixing, STDs, vector-borne
pathogens
Density-dependence representative of wildlife & livestock
diseases (especially with smaller population sizes)