Comparison with a compartmental model

The final size equation is derived analytically from a compartmental epidemiological model, the susceptible-infectious-recovered (SIR) model. Yet it can only be solved numerically — thus with perfect solvers the final epidemic size from an SIR model and a final size calculation should be the same.

This vignette compares the final epidemic sizes from a simple SIR model with those estimated using finalsize::final_size().

New to finalsize? It may help to read the “Get started” vignette first!

Code
library(finalsize)

SIR model definition

First, define a simple SIR model and a helper function to replicate the contact matrix by rows and columns. Here, the model is defined as an iteration of transitions between the epidemiological compartments. These transitions form a system of ordinary differential equations (ODEs), whose solution through time could also have been obtained by using a solver such as deSolve::lsoda().

Interested in compartmental modelling to obtain timeseries of epidemic outcomes?

The epidemics package might be useful!

Code
# Replicate the contact matrix
repmat <- function(X, m, n) {
  mx <- dim(X)[1]
  nx <- dim(X)[2]
  matrix(t(matrix(X, mx, nx * n)), mx * m, nx * n, byrow = TRUE)
}

# Calculate the proportion of individuals infected
final_size_r <- function(r0, contact_matrix,
                         demography_vector, susceptibility, p_susceptibility) {
  # check p_susceptibility
  stopifnot(all(rowSums(p_susceptibility) == 1))

  p_susc <- as.vector(p_susceptibility)
  susc <- as.vector(susceptibility)
  demo_v <- rep(demography_vector, ncol(p_susceptibility)) * p_susc
  nsteps <- 10000

  # epidemiological parameters
  h <- 0.1
  beta <- r0
  gamma <- 1

  # initial conditions
  I <- 1e-8 * demo_v
  S <- demo_v - I
  R <- 0

  # prepare contact matrix
  m <- repmat(contact_matrix, ncol(p_susceptibility), ncol(p_susceptibility))

  # multiply contact matrix by susceptibility
  m <- m * susc

  # iterate over compartmental transitions
  for (i in seq_len(nsteps)) {
    force <- (m %*% I) * S * h * beta
    S <- S - force
    g <- I * gamma * h
    I <- I + force - g
    R <- R + g
  }

  # return proportion of individuals recovered
  as.vector(R / demo_v)
}

Prepare \(R_0\).

Code
r0 <- 2.0

Final size in a uniformly mixing population

Here, the assumption is of a uniformly mixing population where social contacts do not differ among demographic groups (here, age groups).

Code
# estimated population size for the UK is 67 million
population_size <- 67e6

# estimate using finalsize::final_size
final_size_finalsize <- final_size(
  r0 = r0,
  contact_matrix = matrix(1) / population_size,
  demography_vector = population_size,
  susceptibility = matrix(1),
  p_susceptibility = matrix(1),
  solver = "newton"
)

# estimate from SIR model
final_size_sir <- final_size_r(
  r0 = r0,
  contact_matrix = 0.99 * matrix(1) / population_size,
  demography_vector = population_size,
  susceptibility = matrix(1),
  p_susceptibility = matrix(1)
)

# View the estimates
final_size_finalsize
#>     demo_grp   susc_grp susceptibility p_infected
#> 1 demo_grp_1 susc_grp_1              1  0.7968121

final_size_sir
#> [1] 0.7968781

There is a small discrepancy between the final epidemics sizes returned by final_size() and the SIR model. One quick check to identify the source of the error would be to calculate the difference between the two sides of the finalsize equation, i.e. \(x\), the estimated final size, against \(1 - e^{-R_0x}\) (for a difference of \(x - (1 - e^{-R_0x})\)) to see which of the final size estimates is incorrect — a small error, below the error tolerance value (\(1^{-6}\)), would indicate that the method being checked is more or less correct, while a larger error would suggest the method is not correct.

This can be written as a temporary function.

Code
# function to check error in final size estimates
final_size_error <- function(x, r0, tolerance = 1e-6) {
  error <- abs(x - (1.0 - exp(-r0 * x)))
  if (any(error > tolerance)) {
    print("Final size estimate error is greater than tolerance!")
  }
  # return error
  error
}
Code
# error for estimate using finalsize::final_size()
final_size_error(final_size_finalsize$p_infected, r0)
#> [1] 0

# error for estimate from SIR model
final_size_error(final_size_sir, r0)
#> [1] "Final size estimate error is greater than tolerance!"
#> [1] 3.915923e-05

This suggests that it is the SIR model which makes a small error in the final size estimate, while finalsize::final_size() provides an estimate that is within the specified solver tolerance (\(1^{-6}\)).

Final size with heterogeneous mixing

Here, the comparison considers scenarios in which social contacts vary by demographic groups, here, age groups. See the vignette on “Modelling heterogeneous social contacts” for more guidance on how to implement such scenarios.

Prepare population data

First, prepare the population data using the dataset provided with finalsize. This includes preparing the population’s age-specific contact matrix and demography distribution vector, as well as the susceptibility and demography-in-susceptibility distribution vectors. See the “Guide to constructing susceptibility matrices” for help understanding this latter step, and see the the vignette on “Modelling heterogeneous susceptibility” for worked out examples.

Code
# Load example POLYMOD data included with the package
data(polymod_uk)

# Define contact matrix (entry {ij} is contacts in group i
# reported by group j)
contact_matrix <- polymod_uk$contact_matrix

# Define population in each age group
demography_vector <- polymod_uk$demography_vector

Examine the contact matrix and demography vector.

Code
contact_matrix
#>                  
#> contact.age.group         [,1]         [,2]         [,3]
#>           [0,20)  4.514575e-08 1.600071e-08 8.965770e-09
#>           [20,40) 1.600071e-08 2.489595e-08 1.346051e-08
#>           40+     8.965770e-09 1.346051e-08 1.464763e-08

demography_vector
#> [1] 14799290 16526302 28961159

Susceptibility varies between groups

In this scenario, susceptibility to infection only varies between groups, with older age groups (20 and above) only half as susceptible as individuals aged (0 – 19).

Code
# Define susceptibility of each group
susceptibility <- matrix(
  data = c(1.0, 0.5, 0.5),
  nrow = length(demography_vector),
  ncol = 1
)

# Assume uniform susceptibility within age groups
p_susceptibility <- matrix(
  data = 1.0,
  nrow = length(demography_vector),
  ncol = 1
)

Examine the final size estimates.

Code
# from {finalsize} using finalsize::final_size
final_size(
  r0 = r0,
  contact_matrix = contact_matrix,
  demography_vector = demography_vector,
  susceptibility = susceptibility,
  p_susceptibility = p_susceptibility,
  solver = "newton"
)
#>   demo_grp   susc_grp susceptibility p_infected
#> 1   [0,20) susc_grp_1            1.0  0.7119330
#> 2  [20,40) susc_grp_1            0.5  0.3252457
#> 3      40+ susc_grp_1            0.5  0.2334420

# using SIR model
final_size_r(
  r0 = r0,
  contact_matrix = contact_matrix * 0.99,
  demography_vector = demography_vector,
  susceptibility = susceptibility,
  p_susceptibility = p_susceptibility
)
#> [1] 0.7057450 0.3187776 0.2281473

The final size estimates are very close, and the difference does not fundamentally alter our understanding of the potential outcomes of an epidemic in this population.

Susceptibility varies within groups in different proportions

In this scenario, susceptibility to infection varies within age groups, with a fraction of each age group much less susceptible. Additionally, the proportion of individuals with low susceptibility varies between groups, with individuals aged (0 – 19) much more likely to have low susceptibility than older age groups.

Code
# Define susceptibility of each group
# Here
susceptibility <- matrix(
  c(1.0, 0.1),
  nrow = length(demography_vector), ncol = 2,
  byrow = TRUE
)

# view the susceptibility matrix
susceptibility
#>      [,1] [,2]
#> [1,]    1  0.1
#> [2,]    1  0.1
#> [3,]    1  0.1

# Assume variation in susceptibility within age groups
# A higher proportion of 0 -- 20s are in the lower susceptibility group
p_susceptibility <- matrix(1, nrow = length(demography_vector), ncol = 2)
p_susceptibility[, 2] <- c(0.6, 0.5, 0.3)
p_susceptibility[, 1] <- 1 - p_susceptibility[, 2]

# view p_susceptibility
p_susceptibility
#>      [,1] [,2]
#> [1,]  0.4  0.6
#> [2,]  0.5  0.5
#> [3,]  0.7  0.3

Examine the final size estimates for this scenario.

Code
# estimate group-specific final sizes using finalsize::final_size
final_size(
  r0 = r0,
  contact_matrix = contact_matrix,
  demography_vector = demography_vector,
  susceptibility = susceptibility,
  p_susceptibility = p_susceptibility
)
#>   demo_grp   susc_grp susceptibility p_infected
#> 1   [0,20) susc_grp_1            1.0 0.27211694
#> 2  [20,40) susc_grp_1            1.0 0.24819087
#> 3      40+ susc_grp_1            1.0 0.19449495
#> 4   [0,20) susc_grp_2            0.1 0.03126239
#> 5  [20,40) susc_grp_2            0.1 0.02812422
#> 6      40+ susc_grp_2            0.1 0.02139636

# estimate group-specific final sizes from the SIR model
final_size_r(
  r0 = r0,
  contact_matrix = contact_matrix,
  demography_vector = demography_vector,
  susceptibility = susceptibility,
  p_susceptibility = p_susceptibility
)
#> [1] 0.27261723 0.24864863 0.19485488 0.03132008 0.02817620 0.02143593

Here too, the final size estimates are very similar, and provide essentially the same understanding of the potential outcomes of an epidemic in this population.

Complete immunity to infection in part of the population

In this scenario, a fraction of each age group is completely immune to infection. Additionally, individuals aged (0 – 19) are much more likely to have complete immunity than older age groups.

Code
# Define susceptibility of each group
susceptibility <- matrix(1, nrow = length(demography_vector), ncol = 2)
susceptibility[, 2] <- 0.0

# view susceptibility
susceptibility
#>      [,1] [,2]
#> [1,]    1    0
#> [2,]    1    0
#> [3,]    1    0

# Assume that some proportion are completely immune
# complete immunity is more common among younger individuals
p_susceptibility <- matrix(1, nrow = length(demography_vector), ncol = 2)
p_susceptibility[, 2] <- c(0.6, 0.5, 0.3)
p_susceptibility[, 1] <- 1 - p_susceptibility[, 2]

# view p_susceptibility
p_susceptibility
#>      [,1] [,2]
#> [1,]  0.4  0.6
#> [2,]  0.5  0.5
#> [3,]  0.7  0.3
Code
final_size(
  r0 = r0,
  contact_matrix = contact_matrix,
  demography_vector = demography_vector,
  susceptibility = susceptibility,
  p_susceptibility = p_susceptibility
)
#>   demo_grp   susc_grp susceptibility p_infected
#> 1   [0,20) susc_grp_1              1 0.07787362
#> 2  [20,40) susc_grp_1              1 0.07227733
#> 3      40+ susc_grp_1              1 0.05594389
#> 4   [0,20) susc_grp_2              0 0.00000000
#> 5  [20,40) susc_grp_2              0 0.00000000
#> 6      40+ susc_grp_2              0 0.00000000

final_size_r(
  r0 = r0,
  contact_matrix = contact_matrix,
  demography_vector = demography_vector,
  susceptibility = susceptibility,
  p_susceptibility = p_susceptibility
)
#> [1] 0.07791032 0.07231121 0.05597004 0.00000001 0.00000001 0.00000001

Here, the method from finalsize::final_size() correctly provides estimates of 0.0 for the susceptibility group (in each age group) that is completely immune, while the SIR model method estimates that a very small fraction of the completely immune groups are still infected. This discrepancy is due to the assumption of initial conditions for the SIR model, wherein the same fraction (\(1^{-8}\)) are assumed to be initially infected.