simex is an R package for simulating the spread of infectious diseases, as well as interventions such as social distancing measures, isolation and vaccination. It uses an age-structured SEIR compartmental model with country-specific age demographics and contact rates. The simplest way to interact with the tool is to use the Shiny App.
To install the development version from github:
remotes::install_github("WHO-Collaboratory/simex")Load the package using:
library("simex")You can use simex interactively by launching the shiny app locally with
simex::run_shiny(). If you want to use it programmatically, follow the steps
below.
Most settings are specified via the get_parameters function. The arguments and
their default values are given below:
| Argument | Description | Default value |
|---|---|---|
| iso3 | The ISO3 code of the country used to draw age-distributions and contact rates from. | "USA" |
| R0 | The basic reproduction number. | 3 |
| generation_time | The mean generation time in days. | 10 |
| incubation_period | The mean incubation period in days. | 6 |
| infectiousness_presymp | Relative infectiousness of presymptomatic cases to symptomatic cases. | 0.25 |
| frac_symp | The proportion of cases that eventually develop symptoms. | 0.8 |
| ifr | Infection fatality rate provided either as a single value or as a vector of the same length as the number of age categories. Use the function 'age_to_ifr' to calculate a COVID-like IFR from a vector of ages. | age_to_ifr(get_age_median()) |
| hosp_mortality | The probability of death given a case is admitted to hospital, either as a single value or as a vector of the same length as the number of age categories. The inverse of the number of cases admitted to hospital per death. | 1/seq(20, 5, length = 16) |
| hosp_protection_death | Given a case requires hospitalisation, the proportion of deaths admission to hospital averts. | 0.75 |
| hosp_duration | The mean duration of stay in the hospital in days, either as a single value or as a vector of the same length as the number of age categories. | seq(7, 21, length = 16) |
| hosp_capacity | Total hospital bed capacity given as a proportion of the population. | 0.0025 |
| comm_mortality | The probability of death of cases that remain in the community, either as a single value or as a vector of the same length as the number of age categories. | rep(0, 16) |
| vax_rate | The daily rate of vaccination as a proportion of the population. | 0.001 |
| vax_infectiousness | The reduction (as a proportion) in infectioussness of an individual due to vaccination. | 0.3 |
| vax_infection | The protection (as a proportion) against infection provided by vaccination. | 0.5 |
| vax_hosp | The protection (as a proportion) against hospitalisation provided by vaccination, given infection. | 0.5 |
| vax_death | The protection (as a proportion) against death provided by vaccination, given hospitalisation. | 0.8 |
| isolation_adherence | The proportion of symptomatic individuals that adhere to isolation measures. | 0 |
| isolation_effectiveness | The reduction in daily transmission potential of a given individual due to adherence to isolation measures. | 0.8 |
| isolation_delay | The mean delay from symptom onset to isolation in days. | 3 |
| social_distancing | A named vector of length 4 containing the proportion reduction in contacts due to social distancing of 'home', 'school', 'work' and 'other'. | c(home = 0, school = 0, work = 0, other = 0) |
| vax_prioritised | A logical indicating whether older age groups are vaccinated first. | TRUE |
| hosp_prioritised | A logical indicating whether older age groups are hospitalised first when hospital capacity is exceeded. | TRUE |
To run the model using default settings, specify a parameter object par and
feed this into the run_model function.
## set parameters using defaults
pars <- get_parameters()
## run model using default parameters
output <- run_model(pars)
## look at output
print(output)##
## [Simex Object]
## - Days: 1 to 365
## - Age Categories: age_1 to age_16
## - Compartments: S_u | E_u | C_u | H_u | R_u | D_u | S_v | E_v | C_v | H_v | R_v | D_v
To visualise the results, use the generic plot function defined for the
simex class. Below, we first visualise prevalence and then incidence,
specified using the what argument.
## visualise prevalence
plot(output, what = "prevalence")
## visualise incidence
plot(output, what = "incidence")
Hospital capacity can be displayed by toggling the show_hosp_capacity
argument.
## visualise prevalence with hospital capaciy
plot(output, what = "prevalence", show_hosp_capacity = TRUE)
To get a summary of the model state at a given point in time, use the summary
function with the optional day argument. In the example below, we display the
summary statistics for day 150:
## get summary at day 150
summary(output, day = 150)| statistic | value |
|---|---|
| incidence_last_7 | 0.0026936 |
| incidence_cumulative | 0.0370007 |
| incidence_weekly_change | 1.0051376 |
| death_last_7 | 0.0000007 |
| death_cumulative | 0.0000103 |
| death_weekly_change | 1.0257307 |
| hosp_admission_last_7 | 0.0000211 |
| hosp_admission_cumulative | 0.0002897 |
| hosp_admission_weekly_change | 1.0261607 |
| hospital_occupancy | 0.0698724 |
| proportion_cases_hospitalised_last_7 | 0.0077865 |
| ifr_last_7 | 0.0002768 |
| ifr_cumulative | 0.0002774 |
The output object is of class simex and is a list containing four items:
prevalenceis an array with 3 dimensions: time (365 days) x age (16 categories) x state (12 compartments). The compartments are S (susceptible), E (exposed, pre-symptomatic), C (symptomatic in community), H (sympomatic in hospital), R (recovered), D (dead). Each comparment is split into unvaccinated (with subscript _u) and vaccinated (with subscript _v). Each value in the array represents the proportion of the population in that given age-disease compartment on a given day.deltasis an array with the same dimensions asprevalence. Each value represents the change in a given age-disease compartment from the previous day.incidenceis an array with the same dimensions asprevalence. Each value represents the new additions to a given age-disease comparment on a given day. For the exposed category E, this represents incidence of new infections, for the hospitalised category H this represents new hospital admissions, and for the dead category D this represents new deaths.parscontains the parameter set initially fed intorun_model.
Accessing the data is easiest using array indexing. Remember, the dimensions are
time, age and compartment. For example, accessing the prevalence on the 250th
day is done with output$prevalence[250,,]:
## state
## age S_u E_u C_u H_u R_u D_u S_v E_v C_v H_v R_v D_v
## age_1 1.3 0 0.1 0 4.1 0 0.0 0 0.0 0 0.0 0
## age_2 0.6 0 0.1 0 5.2 0 0.0 0 0.0 0 0.0 0
## age_3 0.3 0 0.1 0 5.8 0 0.0 0 0.0 0 0.0 0
## age_4 0.1 0 0.0 0 6.4 0 0.0 0 0.0 0 0.0 0
## age_5 0.6 0 0.1 0 5.9 0 0.0 0 0.0 0 0.0 0
## age_6 0.6 0 0.1 0 5.9 0 0.0 0 0.0 0 0.0 0
## age_7 0.7 0 0.1 0 6.2 0 0.0 0 0.0 0 0.0 0
## age_8 0.5 0 0.1 0 6.1 0 0.0 0 0.0 0 0.0 0
## age_9 0.1 0 0.1 0 5.9 0 0.4 0 0.0 0 0.0 0
## age_10 0.0 0 0.1 0 5.2 0 0.6 0 0.0 0 0.0 0
## age_11 0.0 0 0.1 0 5.2 0 0.8 0 0.0 0 0.0 0
## age_12 0.0 0 0.1 0 4.6 0 1.4 0 0.0 0 0.1 0
## age_13 0.0 0 0.0 0 2.4 0 3.2 0 0.1 0 0.6 0
## age_14 0.0 0 0.0 0 0.1 0 4.2 0 0.1 0 1.3 0
## age_15 0.0 0 0.0 0 0.0 0 3.6 0 0.0 0 0.9 0
## age_16 0.0 0 0.0 0 0.0 0 6.3 0 0.0 0 1.1 0
Accessing the prevalence of the 1st age compartment (0-4) and 1st infectious
compartment (unvaccinated susceptible) for days 130 to 135 is done with
output$prevalence[130:135,1,1]:
## day_130 day_131 day_132 day_133 day_134 day_135
## 5.5 5.5 5.5 5.5 5.5 5.5
The outputs can also be extracted in tibble form using the extract function,
once again using the what argument to specify whether prevalence or incidence
is extracted.
## extract prevalence
extract(output, what = "prevalence")## # A tibble: 4,380 × 4
## day vax compartment value
## <int> <lgl> <fct> <dbl>
## 1 1 FALSE S 1.00
## 2 1 FALSE E 0
## 3 1 FALSE C 0.00000000293
## 4 1 FALSE H 0
## 5 1 FALSE R 0
## 6 1 FALSE D 0
## 7 1 TRUE S 0
## 8 1 TRUE E 0
## 9 1 TRUE C 0
## 10 1 TRUE H 0
## # ℹ 4,370 more rows
If we want to filter this list for a sequence of days, we can then do basic dataframe manipulation:
## define start and end days
days_from <- 10
days_to <- 20
## extract prevalence
df <- extract(output, what = "prevalence")
df <- df[df$day %in% seq(days_from, days_to),]
df## # A tibble: 132 × 4
## day vax compartment value
## <int> <lgl> <fct> <dbl>
## 1 10 FALSE S 9.91e- 1
## 2 10 FALSE E 3.60e- 9
## 3 10 FALSE C 3.33e- 9
## 4 10 FALSE H 5.35e-11
## 5 10 FALSE R 2.48e- 9
## 6 10 FALSE D 1.45e-12
## 7 10 TRUE S 9.00e- 3
## 8 10 TRUE E 2.10e-12
## 9 10 TRUE C 6.97e-13
## 10 10 TRUE H 2.06e-13
## # ℹ 122 more rows
We use vaccination as an example intervention. Referencing the table above, we
can see that vaccination rate is specified using the vax_rate argument and
set it to 0.5% of the population per day.
## define vaccination rate
pars <- get_parameters(vax_rate = 0.005)
## run model
output <- run_model(pars)
## visualise prevalence
plot(output)
We can see that the daily increase in number of vaccinated individuals, as well as the impact on infection and disease severity.
The default initial state begins with an entirely susceptible, unvaccinated population and a single infection. We can also specify a custom initial state by passing a matrix specifying the number of individuals in each age-infection compartment. In the example below, we extract the initial state from the previous run, and modify the compartments so half the susceptible population is assigned to the vaccinated compartment.
## extract starting point from previous run
state <- output$prevalence[1,,]
## assign half of susceptibles to vaccinated
state[,"S_u"] <- state[,"S_v"] <- state[,"S_u"]/2
## run model
output <- run_model(pars, init_state = state)
## visualise prevalence
plot(output)
We can see that the simulation now begins with a 50% vaccinated population, significantly reducing the size and impact of the pandemic.
Sometime we only want to introduce interventions at a certain time, or we want to model multiple, successive interventions. We can do this easily by generating a named list of parameters, where the name gives the time that parameter set should be used from. In the example below, we introduce isolation measures with an adherence of 50% on the 125th day.
## introduce isolation on the 125th day
parlist <- list(
"1" = get_parameters(),
"125" = get_parameters(isolation_adherence = 0.5)
)
## run model
output <- run_model(parlist)
## visualise prevalence
plot(output)
Comparing this figure with the first model run with no interventions, we can see the proportion of deaths drops from about 0.7% to 0.4%; a reduction in deaths of more than 40%!
It is useful to directly compare different scenarios visually. The
vis_comarison function does exactly this, and accepts a named list of simex
objects. In the example below, we generate a named list of lists that compares
the default scenario (no interventions) with the a scenario where isolation is
introduced on the 125th day.
## define two scenarios, one without intervention and one with isolation
parlists <- list(
"No intervention" = get_parameters(),
"Isolation on day 125" = list(
"1" = get_parameters(),
"125" = get_parameters(isolation_adherence = 0.5)
)
)
## run model across set of parameter lists
outputs <- lapply(parlists, run_model)
## compare scenarios
vis_comparison(outputs)
- Finlay Campbell (campbellf@who.int)
- Prabasaj Paul (ppaul@who.int)
Maintainer: Finlay Campbell
The simex software is made available by Collaboratory (2024) under an MIT license (citation file).
The demographic data used in this software is made available by United Nations (2024) under a Creative Commons license CC BY 3.0 IGO. Source:
United Nations, Department of Economic and Social Affairs, Population Division (2024). World Population Prospects 2024, Online Edition.
The contact data used in this software is made available by Prem et al. (2017) under a Creative Commons license CC BY 4.0. Source:
Kiesha Prem, Alex R. Cook, Mark Jit, Projecting social contact matrices in 152 countries using contact surveys and demographic data, PLoS Comp. Biol. (2017), https://doi.org/10.1371/journal.pcbi.1005697.