Skip to main content

SIR Extensions: Vaccination and Herd Immunity

Module 1: Low-Dimensional Dynamics

NoteReading
  • Recommended: Hethcote (2000), the sections on vaccination and on SEIR and SIRS models.
  • Background: Keeling and Rohani (2008).

The SIR model rests on several assumptions: all individuals start susceptible, vaccinated individuals are fully protected, infection makes people infectious at once, immunity is permanent and the population is closed and well mixed. This page follows a simple rule: break one assumption at a time and change the model as little as possible. Each change gives a new behaviour and often a new mathematical object, such as a higher-dimensional system of ODEs, a PDE or a network.

We use the normalized SIR model of Compartmental Models with \(S + I + R = 1\),

\[ \dot S = -\beta S I, \qquad \dot I = \beta S I - \gamma I, \qquad \dot R = \gamma I, \qquad R_0 = \frac{\beta}{\gamma}. \]

Vaccination with a perfect vaccine

Suppose a fraction \(v\) of the population is vaccinated and fully protected at \(t = 0\). The equations do not change; only the initial condition does,

\[ S(0) = 1 - v - I_0, \qquad I(0) = I_0, \qquad R(0) = v . \]

An outbreak grows if \(\beta S(0) > \gamma\), that is if the effective reproduction number \(R_{\text{eff}}(0) = R_0(1 - v)\) exceeds \(1\). The herd immunity threshold is

\[ v_c = 1 - \frac{1}{R_0}, \tag{1}\]

the same expression as the herd immunity formula in The Reproduction Number. Two points are easy to misread:

  • \(v_c\) is a threshold for growth. Above \(v_c\) an introduced case does not start an epidemic in the deterministic model, but a few cases can still occur.
  • It assumes homogeneous mixing and a vaccine that gives complete protection.

How large is the threshold for real diseases?

Equation 1 grows quickly with \(R_0\). The table uses typical \(R_0\) ranges, published estimates vary between studies and populations, so treat the ranges as orders of magnitude. The thresholds are computed from Equation 1.

Disease Typical \(R_0\) \(v_c = 1 - 1/R_0\)
Seasonal influenza 1.2 to 1.6 0.17 to 0.38
COVID-19 (original strain) 2 to 3 0.50 to 0.67
Smallpox 5 to 7 0.80 to 0.86
Polio 5 to 7 0.80 to 0.86
Measles 12 to 18 0.92 to 0.94

For measles, a drop of a few percentage points in coverage below about 94% is enough for epidemics to be possible again, which is why measles elimination is fragile.

An imperfect (“leaky”) vaccine

Now relax the assumption that vaccinated people are fully protected. A leaky vaccine reduces susceptibility by a factor \(\varepsilon\) with \(0 \le \varepsilon \le 1\). Split the susceptibles into unvaccinated \(S_u\) and vaccinated-but-still-susceptible \(S_v\):

\[ \dot S_u = -\beta S_u I, \qquad \dot S_v = -\varepsilon \beta S_v I, \qquad \dot I = \beta (S_u + \varepsilon S_v) I - \gamma I, \qquad \dot R = \gamma I . \tag{2}\]

With coverage \(v\), we have \(S_u(0) \approx 1 - v\) and \(S_v(0) \approx v\), so \(R_{\text{eff}}(0) = R_0\,(1 - v + \varepsilon v)\). Setting this below \(1\) gives

\[ v_c^{\text{leaky}} = \frac{1 - 1/R_0}{1 - \varepsilon} . \tag{3}\]

For \(\varepsilon = 0\) we recover Equation 1. For \(\varepsilon > 0\) the threshold is higher, and if \(\varepsilon > 1/R_0\) then \(v_c^{\text{leaky}} > 1\): even vaccinating everyone does not prevent an epidemic. Coverage and efficacy are two separate quantities.

Figure 1 integrates Equation 2 with \(R_0 = 3\) and checks Equation 3. The fraction ever infected falls to zero at the predicted thresholds.

Figure 1: Fraction of the population ever infected as a function of vaccine coverage v, for R0 = 3. Dashed lines mark the thresholds from the formula. With eps = 0.6 the threshold is above 1, so vaccination alone cannot prevent an epidemic.

Why smallpox could be eradicated

Smallpox is the standard case where the SIR assumptions hold well. It has no animal reservoir, so transmission is only between humans. Infection and vaccination both give long-lasting immunity, and the disease is easy to recognize. With \(R_0\) about 5 to 7 the threshold is about 80 to 86%. The WHO declared smallpox eradicated in 1980, after the last natural case in 1977.

Other diseases break one or more assumptions, so the SIR model is not enough for them:

Disease Why the plain SIR model fails or is fragile
Measles SIR-like, but a very high \(R_0\) means a threshold above 90%.
Influenza Waning immunity, antigenic drift and animal reservoirs.
Tuberculosis Long latent periods, partial immunity and chronic infection.

In modelling terms, eradication requires the disease-free equilibrium to be reachable and globally stable, not merely locally stable.

A menu of extensions

Each row breaks one assumption and gives the smallest model change that does so. You have already seen the first rows above.

Assumption broken Model change New behaviour
All start susceptible \(S(0) = 1 - v\) Herd immunity threshold \(v_c\).
Vaccine fully protective Two susceptible classes, factor \(\varepsilon\) (Equation 2) Higher threshold, coverage separated from efficacy.
Infection means infectious at once Add an exposed class \(E\) with latent period \(1/\sigma\) (SEIR) Delay between infection and cases, later and lower peak.
Immunity is permanent Waning at rate \(\omega\) (SIRS) Endemic equilibrium in a closed population, possible recurrent outbreaks.
Closed population Births and deaths at rate \(\mu\) Endemic equilibria, vaccination as a steady-state control problem.
Homogeneous mixing Groups with a contact matrix \(\beta_{ij}\) \(R_0\) is the spectral radius of a next-generation matrix, targeted vaccination.
Space is irrelevant Diffusion terms (reaction-diffusion SIR, a PDE) Spatial waves and hotspots.
A single population Patches coupled by a mobility matrix Importation, exportation and staggered regional waves.
Risk depends only on fractions Contacts on a network (Module 5) Superspreading, different thresholds, targeted vaccination of high-degree nodes.
Constant infectiousness Infectiousness depending on time since infection (renewal equation) Direct link to generation-interval data.
Constant \(\beta\) \(\beta(t)\) or \(\beta(I) = \beta_0/(1 + kI)\) Multiple waves without waning immunity.
We observe infections directly Observation model: reported cases \(\approx \rho\, i(t - \tau)\) Explains mismatches between model and reported data.

The last row connects to Fitting an Epidemic to Data and Predicting the Peak, where reporting noise is the reason the peak is hard to predict.

Two extensions in simulation

The SEIR model adds one equation and one time scale,

\[ \dot S = -\beta S I, \quad \dot E = \beta S I - \sigma E, \quad \dot I = \sigma E - \gamma I, \quad \dot R = \gamma I, \]

and the SIRS model returns recovered individuals to the susceptible class,

\[ \dot S = -\beta S I + \omega R, \quad \dot I = \beta S I - \gamma I, \quad \dot R = \gamma I - \omega R . \]

In SIRS, the endemic equilibrium has \(S^* = 1/R_0\) and \(I^* = \dfrac{\omega}{\gamma + \omega}\left(1 - \dfrac{1}{R_0}\right)\), which exists for \(R_0 > 1\).

Figure 2: Left: SIR and SEIR with the same beta, gamma and initial infectious fraction 0.001 (sigma = 0.2 per day). The latent period delays and lowers the peak. Right: SIRS with omega = 0.01 per day settles to the endemic level given by the formula.

Choosing the right model

Extend the SIR model only as far as the question and the data require. More realism is not more truth: every extension adds parameters and uncertainty. Use these models to understand mechanisms and thresholds and to compare intervention scenarios, not to produce a perfect forecast.

NoteExercise 1

For measles with \(R_0 = 15\), compute the herd immunity threshold. A vaccine reduces susceptibility by a factor \(\varepsilon = 0.05\) (95% efficacy). What coverage is needed?

\(v_c = 1 - 1/15 \approx 0.933\). With a leaky vaccine, Equation 3 gives \(v_c^{\text{leaky}} = 0.933/(1 - 0.05) \approx 0.98\). Almost the whole population must be vaccinated, which shows why measles elimination is fragile.

NoteExercise 2

Use Figure 2 as a starting point. Increase the latent period (reduce \(\sigma\)) and describe how the peak time and the height of the peak change. Does the final size change?

A longer latent period delays the peak and lowers it. The final size does not change, because it depends only on \(R_0\) and the initial condition.

NoteOptional project

Consider a hypothetical human-only disease with \(R_0\) between 5 and 7 and fatality ratio about 0.3, with outbreaks starting in five regions at once. Choose a model structure (SEIR, single population or metapopulation), model a vaccination campaign with a start delay and a daily dose capacity, estimate cumulative infections and deaths until you declare control, and identify the two or three assumptions that most affect your estimate.

Next

Simulating the SIR Model implements the SIR and SIS models, and Fitting an Epidemic to Data fits them to data.

References

Hethcote, Herbert W. 2000. “The Mathematics of Infectious Diseases.” SIAM Review 42 (4): 599–653. https://doi.org/10.1137/S0036144500371907.
Keeling, Matt J., and Pejman Rohani. 2008. Modeling Infectious Diseases in Humans and Animals. Princeton University Press.