Mathematical modelling of infectious disease

The modeling of infectious diseases is a tool that has been used to study the mechanisms by which diseases spread, to predict the future course of an outbreak and to evaluate strategies to control an epidemic.[1]

The first scientist who systematically tried to quantify causes of death was John Graunt in his book Natural and Political Observations made upon the Bills of Mortality, in 1662. The bills he studied were listings of numbers and causes of deaths published weekly. Graunt's analysis of causes of death is considered the beginning of the "theory of competing risks" which according to Daley and Gani [1] is "a theory that is now well established among modern epidemiologists".

The earliest account of mathematical modelling of spread of disease was carried out in 1760 by Daniel Bernoulli. Trained as a physician, Bernoulli created a mathematical model to defend the practice of inoculating against smallpox.[2] The calculations from this model showed that universal inoculation against smallpox would increase the life expectancy from 26 years 7 months to 29 years 9 months.[3] Daniel Bernoulli's work preceded the modern understanding of germ theory.

In the early 20th century, William Hamer[4] and Ronald Ross[5] applied the law of mass action to explain epidemic behaviour.

The 1920s saw the emergence of compartmental models. The Kermack–McKendrick epidemic model (1927) and the Reed–Frost epidemic model (1928) both describe the relationship between susceptible, infected and immune individuals in a population. The Kermack–McKendrick epidemic model was successful in predicting the behavior of outbreaks very similar to that observed in many recorded epidemics.[6]

Recently, agent-based models (ABMs) have been used in exchange for simpler compartmental models, e.g.,.[7] For example, epidemiological ABMs have been used to inform public health (nonpharmaceutical) interventions against the spread of SARS-CoV-2.[8] Epidemiological ABMs, in spite of their complexity and requiring high computational power, have been criticized for simplifying and unrealistic assumptions.[9][10] Still, they can be useful in informing decisions regarding mitigation and suppression measures in cases when ABMs are accurately calibrated.[11]

Models are only as good as the assumptions on which they are based. If a model makes predictions that are out of line with observed results and the mathematics is correct, the initial assumptions must change to make the model useful.

• Rectangular and stationary age distribution, i.e., everybody in the population lives to age L and then dies, and for each age (up to L) there is the same number of people in the population. This is often well-justified for developed countries where there is a low infant mortality and much of the population lives to the life expectancy.
• Homogeneous mixing of the population, i.e., individuals of the population under scrutiny assort and make contact at random and do not mix mostly in a smaller subgroup. This assumption is rarely justified because social structure is widespread. For example, most people in London only make contact with other Londoners. Further, within London then there are smaller subgroups, such as the Turkish community or teenagers (just to give two examples), who mix with each other more than people outside their group. However, homogeneous mixing is a standard assumption to make the mathematics tractable.

The basic reproduction number (denoted by R₀) is a measure of how transferable a disease is. It is the average number of people that a single infectious person will infect over the course of their infection. This quantity determines whether the infection will spread exponentially, die out, or remain constant: if R₀ > 1, then each person on average infects more than one other person so the disease will spread; if R₀ < 1, then each person infects fewer than one person on average so the disease will die out; and if R₀ = 1, then each person will infect on average exactly one other person, so the disease will become endemic: it will move throughout the population but not increase or decrease.

An infectious disease is said to be endemic when it can be sustained in a population without the need for external inputs. This means that, on average, each infected person is infecting exactly one other person (any more and the number of people infected will grow exponentially and there will be an epidemic, any less and the disease will die out). In mathematical terms, that is:

${\displaystyle \ R_{0}S\ =1.}$

The basic reproduction number (R₀) of the disease, assuming everyone is susceptible, multiplied by the proportion of the population that is actually susceptible (S) must be one (since those who are not susceptible do not feature in our calculations as they cannot contract the disease). Notice that this relation means that for a disease to be in the endemic steady state, the higher the basic reproduction number, the lower the proportion of the population susceptible must be, and vice versa. This expression has limitations concerning the susceptibility proportion, e.g. the R₀ equals 0.5 implicates S has to be 2, however this proportion exceeds to population size.

Assume the rectangular stationary age distribution and let also the ages of infection have the same distribution for each birth year. Let the average age of infection be A, for instance when individuals younger than A are susceptible and those older than A are immune (or infectious). Then it can be shown by an easy argument that the proportion of the population that is susceptible is given by:

${\displaystyle S={\frac {A}{L}}.}$

We reiterate that L is the age at which in this model every individual is assumed to die. But the mathematical definition of the endemic steady state can be rearranged to give:

${\displaystyle S={\frac {1}{R_{0}}}.}$

Therefore, due to the transitive property:

${\displaystyle {\frac {1}{R_{0}}}={\frac {A}{L}}\Rightarrow R_{0}={\frac {L}{A}}.}$

This provides a simple way to estimate the parameter R₀ using easily available data.

For a population with an exponential age distribution,

${\displaystyle R_{0}=1+{\frac {L}{A}}.}$

This allows for the basic reproduction number of a disease given A and L in either type of population distribution.

Compartmental models are formulated as Markov chains.[12] A classic compartmental model in epidemiology is the SIR model, which may be used as a simple model for modeling epidemics. Multiple other types of compartmental models are also employed.

The SIR model

Diagram of the SIR model with initial values ${\textstyle S(0)=997,I(0)=3,R(0)=0}$, and rates for infection ${\textstyle \beta =0.4}$ and for recovery ${\textstyle \gamma =0.04}$

Animation of the SIR model with initial values ${\textstyle S(0)=997,I(0)=3,R(0)=0}$, and rate of recovery ${\textstyle \gamma =0.04}$. The animation shows the effect of reducing the rate of infection from ${\textstyle \beta =0.5}$ to ${\textstyle \beta =0.12}$. If there is no medicine or vaccination available, it is only possible to reduce the infection rate (often referred to as "flattening the curve") by appropriate measures such as social distancing.

In 1927, W. O. Kermack and A. G. McKendrick created a model in which they considered a fixed population with only three compartments: susceptible, ${\displaystyle S(t)}$; infected, ${\displaystyle I(t)}$; and recovered, ${\displaystyle R(t)}$. The compartments used for this model consist of three classes:[13]

• ${\displaystyle S(t)}$ is used to represent the individuals not yet infected with the disease at time t, or those susceptible to the disease of the population.
• ${\displaystyle I(t)}$ denotes the individuals of the population who have been infected with the disease and are capable of spreading the disease to those in the susceptible category.
• ${\displaystyle R(t)}$ is the compartment used for the individuals of the population who have been infected and then removed from the disease, either due to immunization or due to death. Those in this category are not able to be infected again or to transmit the infection to others.

Mathematical models need to integrate the increasing volume of data being generated on host-pathogen interactions. Many theoretical studies of the population dynamics, structure and evolution of infectious diseases of plants and animals, including humans, are concerned with this problem. A model to assess the probability for a worldwide spreading and declare pandemic has been recently developed by Valdez et al.[15] Research topics include:

• Pandemic
• transmission, spread and control of infection
• epidemiological networks
• spatial epidemiology
• persistence of pathogens within hosts
• intra-host dynamics
• immuno-epidemiology
• virulence
• Strain (biology) structure and interactions
• antigenic shift
• phylodynamics
• pathogen population genetics
• evolution and spread of resistance
• role of host genetic factors
• statistical and mathematical tools and innovations
• role and identification of infection reservoirs

If the proportion of the population that is immune exceeds the herd immunity level for the disease, then the disease can no longer persist in the population. Thus, if this level can be exceeded by vaccination, the disease can be eliminated. An example of this being successfully achieved worldwide is the global smallpox eradication, with the last wild case in 1977. The WHO is carrying out a similar vaccination campaign to eradicate polio.

The herd immunity level will be denoted q. Recall that, for a stable state:

${\displaystyle R_{0}\cdot S=1.}$

In turn,

${\displaystyle R_{0}={\frac {N}{S}}={\frac {\mu N\operatorname {E} (T_{L})}{\mu N\operatorname {E} [\min(T_{L},T_{S})]}}={\frac {\operatorname {E} (T_{L})}{\operatorname {E} [\min(T_{L},T_{S})]}},}$

which is approximately:

${\displaystyle {\frac {\operatorname {\operatorname {E} } (T_{L})}{\operatorname {\operatorname {E} } (T_{S})}}=1+{\frac {\lambda }{\mu }}={\frac {\beta N}{v}}.}$

S will be (1 − q), since q is the proportion of the population that is immune and q + S must equal one (since in this simplified model, everyone is either susceptible or immune). Then:

${\displaystyle {\begin{aligned}&R_{0}\cdot (1-q)=1,\\[6pt]&1-q={\frac {1}{R_{0}}},\\[6pt]&q=1-{\frac {1}{R_{0}}}.\end{aligned}}}$

Remember that this is the threshold level. If the proportion of immune individuals exceeds this level due to a mass vaccination programme, the disease will die out.

We have just calculated the critical immunisation threshold (denoted q_c). It is the minimum proportion of the population that must be immunised at birth (or close to birth) in order for the infection to die out in the population.

${\displaystyle q_{c}=1-{\frac {1}{R_{0}}}.}$

Because the fraction of the final size of the population p that is never infected can be defined as:

${\displaystyle \lim _{t\to \infty }S(t)=e^{-\int _{0}^{\infty }\lambda (t)\,dt}=1-p.}$

Hence,

${\displaystyle p=1-e^{-\int _{0}^{\infty }\beta I(t)\,dt}=1-e^{-R_{0}p}.}$

Solving for ${\displaystyle R_{0}}$, we obtain:

${\displaystyle R_{0}={\frac {-\ln(1-p)}{p}}.}$

Models have the advantage of examining multiple outcomes simultaneously, rather than making a single forecast. Models have shown broad degrees of reliability in past pandemics, such as SARS, Swine flu, MERS and Ebola.[16]

Software

• Model-Builder: Interactive (GUI-based) software to build, simulate, and analyze ODE models.
• GLEaMviz Simulator: Enables simulation of emerging infectious diseases spreading across the world.
• STEM: Open source framework for Epidemiological Modeling available through the Eclipse Foundation.
• R package surveillance: Temporal and Spatio-Temporal Modeling and Monitoring of Epidemic Phenomena