This is an introductory vignette to the {simulist} R package. {simulist} simulates two types of common epidemiological data collected during infectious disease outbreaks: 1) a line list, which provides individual-level descriptions of cases in an outbreak; 2) a contact dataset, which provides details of which others individuals were in contact with each of the cases.
The main function in the {simulist} package is sim_linelist(). This functions takes in arguments that control the dynamics of the outbreak, such as the infectious period, and outputs a line list table (<data.frame>) with case information for each infected individual.
For this introduction we will simulate a line list for the early stages of a COVID-19 (SARS-CoV-2) outbreak. This will require two R packages: {simulist}, to produce the line list, and {epiparameter} to provide epidemiological parameters, such as onset-to-death delays.
Code
First we load in some data that is required for the line list simulation. Data on epidemiological parameters and distributions are read from the {epiparameter} R package.
Code
# create contact distribution (not available from {epiparameter} database)
contact_distribution <- epidist(
disease = "COVID-19",
epi_dist = "contact distribution",
prob_distribution = "pois",
prob_distribution_params = c(mean = 2)
)
#> Citation cannot be created as author, year, journal or title is missingCode
# create infectious period (not available from {epiparameter} database)
infect_period <- epidist(
disease = "COVID-19",
epi_dist = "infectious period",
prob_distribution = "gamma",
prob_distribution_params = c(shape = 1, scale = 1)
)
#> Citation cannot be created as author, year, journal or title is missingCode
# get onset to hospital admission from {epiparameter} database
onset_to_hosp <- epidist_db(
disease = "COVID-19",
epi_dist = "onset to hospitalisation",
single_epidist = TRUE
)
#> Using Linton N, Kobayashi T, Yang Y, Hayashi K, Akhmetzhanov A, Jung S, Yuan
#> B, Kinoshita R, Nishiura H (2020). "Incubation Period and Other
#> Epidemiological Characteristics of 2019 Novel Coronavirus Infections
#> with Right Truncation: A Statistical Analysis of Publicly Available
#> Case Data." _Journal of Clinical Medicine_. doi:10.3390/jcm9020538
#> <https://doi.org/10.3390/jcm9020538>..
#> To retrieve the citation use the 'get_citation' functionCode
# get onset to death from {epiparameter} database
onset_to_death <- epidist_db(
disease = "COVID-19",
epi_dist = "onset to death",
single_epidist = TRUE
)
#> Using Linton N, Kobayashi T, Yang Y, Hayashi K, Akhmetzhanov A, Jung S, Yuan
#> B, Kinoshita R, Nishiura H (2020). "Incubation Period and Other
#> Epidemiological Characteristics of 2019 Novel Coronavirus Infections
#> with Right Truncation: A Statistical Analysis of Publicly Available
#> Case Data." _Journal of Clinical Medicine_. doi:10.3390/jcm9020538
#> <https://doi.org/10.3390/jcm9020538>..
#> To retrieve the citation use the 'get_citation' functionThe seed is set to ensure the output of the vignette is consistent. When using {simulist}, setting the seed is not required unless you need to simulate the same line list multiple times.
Code
set.seed(123)The first argument in sim_linelist() is the contact distribution (contact_distribution), which here we specify as Poisson distribution with a mean (average) number of contacts of 2, and with the infectious period and probability of infection per contact (prob_infect) will control the growth rate of the simulated epidemic. Here we set the probability of infection as 0.5 (on average half of contacts become infected). The minimum requirements to simulate a line list are the contact distribution, the infectious period, probability of infection, onset-to-hospitalisation delay and onset-to-death delay.
Code
linelist <- sim_linelist(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = onset_to_hosp,
onset_to_death = onset_to_death
)
head(linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 Hassan Lacsina confirmed m 39 2023-01-01 <NA> recovered
#> 2 3 Joshua Cowan confirmed m 90 2023-01-01 <NA> recovered
#> 3 4 Annabelle Vu probable f 9 2023-01-01 2023-01-07 died
#> 4 5 Emily Fu confirmed f 71 2023-01-01 2023-01-01 died
#> 5 6 Kirsten Barna suspected f 48 2023-01-02 <NA> recovered
#> 6 7 Ty Henderson suspected m 77 2023-01-01 <NA> recovered
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> 25.2
#> 2 <NA> 2022-12-31 2023-01-03 25.2
#> 3 2023-01-16 2022-12-29 2023-01-02 NA
#> 4 2023-01-08 2023-01-02 2023-01-07 25.2
#> 5 <NA> 2022-12-30 2023-01-03 NA
#> 6 <NA> 2022-12-26 2023-01-02 NAControlling outbreak size
The reproduction number (\(R\)) has a strong influence on the size of an outbreak. For {simulist}, the reproduction number is, not provided directly, but rather is determined by the mean number of contacts and the probability of infection. However, the {simulist} package generates line list data using a stochastic algorithm, so even when \(R < 1\) it can produce a substantial outbreak by chance, or an \(R >> 1\) will sometimes not produce a vast epidemic in one simulation (i.e. one replicate) due to the stochasticity.
Alert
The reproduction number (\(R\)) of the simulation results from the contact distribution (contact_distribution) and the probability of infection (prob_infect); the number of infections is a binomial sample of the number of contacts for each case with the probability of infection (i.e. being sampled) given by prob_infect. If the average number of secondary infections from each primary case is greater than 1 (\(R > 1\)) then this can lead to the outbreak becoming extremely large.
There is currently no depletion of susceptible individuals in the simulation model (i.e. infinite population size), so the maximum outbreak size (second element of the vector supplied to the outbreak_size argument) can be used to return a line list early without producing an excessively large data set.
If \(R > 1\), the simulation may return early after reaching the maximum outbreak size. In these scenarios when \(R > 1\), the \(R\) value is controlling the rate at which the maximum outbreak size is reached rather than the size of the outbreak (not all simulations with \(R > 1\) will reach the maximum outbreak size due to stochasticity).
The simulation is therefore sensitive to the contact distribution and probability of infection resulting in an \(R\) just above or below 1.
When requiring a minimum or maximum outbreak size we can specify the outbreak_size argument in sim_linelist(). By default this is set to 10 and 10,000 for the minimum and maximum, respectively. In the case of the minimum outbreak size, this means that the simulation will not return a line list until the conditioning has been met. In other words, the simulation will resimulate a branching process model until an outbreak infects at least 10 people. In the case of the maximum outbreak size, if the number of infected individuals exceeds the maximum the simulation will end, even if there are still infectious individuals capable of continuing transmission, the function will return the data early with a warning that the number of infections in the data has reached the maximum and stating how many cases and contacts are in the data output.
When requiring a line list that represents a large outbreak, such as the COVID-19 outbreak, setting the outbreak_size to a larger number guarantees a line list of at least that size. Here we simulate a line list requiring at least 250 cases (and fewer than 10,000 cases). The maximum number of cases can also be increased when simulating outbreaks such as global pandemics.
Code
linelist <- sim_linelist(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = onset_to_hosp,
onset_to_death = onset_to_death,
outbreak_size = c(250, 1e4)
)
head(linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 Traegon Vasquez-Luna confirmed m 87 2023-01-01 <NA> recovered
#> 2 2 Xiaoyun Mattern confirmed f 61 2023-01-01 <NA> recovered
#> 3 3 Charles Thompson confirmed m 29 2023-01-01 2023-01-04 recovered
#> 4 4 Eduardo Lopez confirmed m 71 2023-01-01 2023-02-04 recovered
#> 5 5 Yaneli Leyba probable f 23 2023-01-01 2023-02-07 recovered
#> 6 6 Renata Tinner confirmed f 7 2023-01-01 <NA> recovered
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> 25.9
#> 2 <NA> 2023-01-03 2023-01-03 25.9
#> 3 <NA> 2022-12-31 2023-01-07 25.9
#> 4 <NA> 2022-12-25 2023-01-02 25.9
#> 5 <NA> 2023-01-04 2023-01-05 NA
#> 6 <NA> 2023-01-01 2023-01-04 25.9The amount of time the simulation takes can be determined by the mean of the contact distribution (contact_distribution), the probability of infection (prob_infect) and conditioning the outbreak size (outbreak_size). If the minimum outbreak_size is large, for example hundreds or thousands of cases, and the mean number of contacts and probability of infection mean the reproduction number is below one, it will take many branching process simulations until finding one that produces a sufficiently large epidemic.
Case type
During an infectious disease outbreak it may not be possible to confirm every infection as a case. A confirmed case is typically defined via a diagnostic test. There are several reasons why a case may not be confirmed, including limited testing capacity and mild or non-specific early symptoms, especially in fast growing epidemics. We therefore include two other categories for cases: probable and suspected. For example, probable cases may be those that show clinical evidence for the disease but have not, or cannot, be confirmed by a diagnostic test. Suspected cases are those that are possibly infected but do not show clear clinical or epidemiological evidence, nor has a diagnostic test been performed. Hence the distribution of suspected/probable/confirmed will depend on the pathogen characteristics, outbreak-specific definitions, and resources available.
The line list output from the {simulist} simulation contains a column (case_type) with the type of case.
{simulist} can simulate varying probabilities of each case being suspected, probable or confirmed. By default the sim_linelist() function uses probabilities of suspected = 0.2, probable = 0.3 and confirmed = 0.5.
Code
linelist <- sim_linelist(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = onset_to_hosp,
onset_to_death = onset_to_death
)
head(linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 Jordan Leishman probable m 20 2023-01-01 <NA> recovered
#> 2 2 Bronson Esposito probable m 75 2023-01-01 2023-01-05 recovered
#> 3 5 Daisy Vasquez-Torres confirmed f 72 2023-01-01 <NA> recovered
#> 4 9 Meron Parker confirmed f 7 2023-01-01 <NA> recovered
#> 5 10 Collin Vahabi probable m 21 2023-01-02 <NA> recovered
#> 6 11 Kevin Ortiz suspected m 13 2023-01-02 2023-01-06 died
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> NA
#> 2 <NA> 2022-12-29 2023-01-01 NA
#> 3 <NA> 2023-01-04 2023-01-05 25.6
#> 4 <NA> 2023-01-04 2023-01-05 25.6
#> 5 <NA> 2022-12-26 2023-01-01 NA
#> 6 2023-01-11 2023-01-05 2023-01-05 NATo alter these probabilities, supply a named vector to the sim_linelist() argument case_type_probs. The vector should contain three numbers, with the names suspected, probable and confirmed, with the numbers summing to one. Here we change the values to simulate an outbreak in which the proportion of cases confirmed by laboratory testing is high.
Code
linelist <- sim_linelist(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = onset_to_hosp,
onset_to_death = onset_to_death,
case_type_probs = c(suspected = 0.05, probable = 0.05, confirmed = 0.9)
)
head(linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 Ryan Mcpherson confirmed m 12 2023-01-01 2023-01-01 recovered
#> 2 2 James Nava confirmed m 18 2023-01-02 <NA> died
#> 3 4 Marcus Benjamin confirmed m 34 2023-01-02 2023-01-09 died
#> 4 6 Tina Danz confirmed f 70 2023-01-02 <NA> recovered
#> 5 7 Safar el-Mahfouz confirmed m 60 2023-01-02 2023-01-03 recovered
#> 6 8 Rafael Paolo Chen confirmed m 64 2023-01-02 <NA> recovered
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> 24.7
#> 2 2023-01-15 2023-01-05 2023-01-06 24.7
#> 3 2023-01-14 2023-01-01 2023-01-03 24.7
#> 4 <NA> 2022-12-31 2023-01-04 24.7
#> 5 <NA> 2023-01-01 2023-01-03 24.7
#> 6 <NA> 2023-01-03 2023-01-05 24.7It is also possible to set one of these categories to 1, in which case every case will be of that type.
The way {simulist} assigns case types is by pasting case types onto existing case data. Thus, it could be viewed that the true underlying data is that all cases in the simulation are confirmed, but that there is a lack of information in some cases. There are no cases in the output line list that are incorrectly attributed as probable or suspected that have not been infected. That is to say, all individuals in the line list, whatever their case type, are infected during the outbreak.
Anonymous line list
By default sim_linelist() provides the name of each individual in the line list. If an anonymised line list is required the anonymise argument of sim_linelist() can be set to TRUE.
Code
linelist <- sim_linelist(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = onset_to_hosp,
onset_to_death = onset_to_death,
anonymise = TRUE
)
head(linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 TZEHlV9qWk confirmed f 70 2023-01-01 <NA> recovered
#> 2 4 YOVXleJLDP confirmed f 12 2023-01-01 <NA> recovered
#> 3 5 GCZVQbf799 suspected f 10 2023-01-01 <NA> died
#> 4 6 zyQOkL4DOx confirmed m 43 2023-01-01 <NA> recovered
#> 5 7 r8HQKWI7fe confirmed m 90 2023-01-01 2023-01-02 died
#> 6 10 eebBmOYMxb probable m 81 2023-01-01 <NA> died
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> 24.8
#> 2 <NA> 2022-12-28 2023-01-02 24.8
#> 3 2023-01-22 2022-12-31 2023-01-03 NA
#> 4 <NA> 2023-01-03 2023-01-06 24.8
#> 5 2023-01-18 2023-01-05 2023-01-06 24.8
#> 6 2023-01-13 2023-01-03 2023-01-03 NAThe names used in the line list are produced at random by the {randomNames} R package. Therefore, even when anonymise = FALSE there is no personal data of real individuals being produced or shared. The anonymise argument only changes the $case_name column of the line list, as this is deemed the only personally identifiable information (PII) in the line list data.
Population age
For an overview of how a line list can be simulated with a uniform or structured population age distribution see the vignette dedicated to this topic.
Age-stratified hospitalisation and death risks
For an overview of how a line list can be simulated with age-stratified (or age-dependent) hospitalisation and death risks see the vignette dedicated to this topic.
Simulate contacts table
To simulate a contacts table, the sim_contacts() function can be used. This requires the same arguments as sim_linelist(), but does not require the onset-to-hospitalisation delay and onset-to-death delays.
Code
contacts <- sim_contacts(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5)
head(contacts)
#> from to age sex date_first_contact
#> 1 Karin Mendicino Robert Tran 72 m 2023-01-01
#> 2 Karin Mendicino Ashley Ellerman 78 f 2023-01-01
#> 3 Karin Mendicino Joshua Gaytan-Gonzalez 33 m 2023-01-02
#> 4 Karin Mendicino Laneace Smith 39 f 2022-12-26
#> 5 Ashley Ellerman Mackenzie Alt 13 f 2023-01-04
#> 6 Joshua Gaytan-Gonzalez Marcie Medo 2 f 2022-12-29
#> date_last_contact was_case status
#> 1 2023-01-05 N under_followup
#> 2 2023-01-04 Y case
#> 3 2023-01-03 Y case
#> 4 2023-01-01 N lost_to_followup
#> 5 2023-01-04 Y case
#> 6 2023-01-03 Y caseSimulate both line list and contacts table
To produce both a line list and a contacts table for the same outbreak, the sim_linelist() and sim_contacts() cannot be used separately due to the stochastic algorithm, meaning the data in the line list will be discordant with the contacts table.
In order to simulate a line list and a contacts table of the same outbreak the sim_outbreak() function is required. This will simulate a single outbreak and return a line list and a contacts table. The inputs of sim_outbreak() are the same as the inputs required for sim_linelist().
Code
outbreak <- sim_outbreak(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = onset_to_hosp,
onset_to_death = onset_to_death
)
head(outbreak$linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 Alan Henn suspected m 10 2023-01-01 <NA> recovered
#> 2 2 Lauryn County confirmed f 72 2023-01-01 <NA> recovered
#> 3 4 Dawson Davis probable m 85 2023-01-01 <NA> recovered
#> 4 5 Shawnee Bemski suspected f 62 2023-01-02 <NA> recovered
#> 5 6 Jaadallah el-Akram confirmed m 40 2023-01-01 2023-01-02 died
#> 6 8 Mahfoodha al-Nassif confirmed f 82 2023-01-01 <NA> recovered
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> NA
#> 2 <NA> 2022-12-29 2023-01-02 24.9
#> 3 <NA> 2022-12-29 2023-01-04 NA
#> 4 <NA> 2023-01-01 2023-01-05 NA
#> 5 2023-01-23 2023-01-02 2023-01-04 24.9
#> 6 <NA> 2023-01-01 2023-01-04 24.9Code
head(outbreak$contacts)
#> from to age sex date_first_contact date_last_contact
#> 1 Alan Henn Lauryn County 72 f 2022-12-29 2023-01-02
#> 2 Alan Henn James Lefebre 80 m 2023-01-02 2023-01-03
#> 3 Alan Henn Dawson Davis 85 m 2022-12-29 2023-01-04
#> 4 Alan Henn Shawnee Bemski 62 f 2023-01-01 2023-01-05
#> 5 Lauryn County Jaadallah el-Akram 40 m 2023-01-02 2023-01-04
#> 6 Lauryn County Landis Veasley 57 m 2023-01-02 2023-01-05
#> was_case status
#> 1 Y case
#> 2 N under_followup
#> 3 Y case
#> 4 Y case
#> 5 Y case
#> 6 N under_followupsim_outbreak() has the same features as sim_linelist() and sim_contacts(), this includes simulating with age-stratified risks of hospitalisation and death, the probability of case types or contact tracing status can be modified.
Advanced
The sim_*() functions, by default, use an excess degree distribution to account for a network effect when sampling the number of contacts in the simulation model (\(q(n) \sim (n + 1)p(n + 1)\) where \(p(n)\) is the probability density function of a distribution, e.g., Poisson or Negative binomial, within the .sim_network_bp() internal function). This network effect can be turned off by using the config argument in any sim_*() function and setting network = "unadjusted" (create_config(network = "unadjusted")) which will instead sample from a probability distribution \(p(n)\).
Using functions for distributions instead of <epidist>
It is possible to use an anonymous function instead of an <epidist> object when specifying the parameters of the delay and contact distributions. We recommend using the <epidist> objects but here outline the alternative approach.
Code
outbreak <- sim_outbreak(
contact_distribution = function(x) dpois(x = x, lambda = 2),
infect_period = function(x) rgamma(n = x, shape = 2, scale = 2),
prob_infect = 0.5,
onset_to_hosp = function(x) rlnorm(n = x, meanlog = 1.5, sdlog = 0.5),
onset_to_death = function(x) rweibull(n = x, shape = 0.5, scale = 0.2)
)
head(outbreak$linelist)
#> id case_name case_type sex age date_onset date_admission outcome
#> 1 1 Pierre Salat probable m 34 2023-01-01 <NA> recovered
#> 2 2 Matthew Piske confirmed m 13 2023-01-05 <NA> recovered
#> 3 5 Carsen Harris suspected f 75 2023-01-06 <NA> recovered
#> 4 7 Fat'hiyaa el-Salam suspected f 1 2023-01-03 <NA> recovered
#> 5 9 Nelle Nible confirmed f 20 2023-01-06 <NA> recovered
#> 6 10 Megan Collie probable f 79 2023-01-06 <NA> recovered
#> date_outcome date_first_contact date_last_contact ct_value
#> 1 <NA> <NA> <NA> NA
#> 2 <NA> 2023-01-01 2023-01-04 25.2
#> 3 <NA> 2023-01-02 2023-01-04 NA
#> 4 <NA> 2022-12-29 2023-01-03 NA
#> 5 <NA> 2023-01-03 2023-01-05 25.2
#> 6 <NA> 2023-01-04 2023-01-07 NACode
head(outbreak$contacts)
#> from to age sex date_first_contact date_last_contact
#> 1 Pierre Salat Matthew Piske 13 m 2023-01-01 2023-01-04
#> 2 Pierre Salat Chandler Geyer 62 f 2023-01-02 2023-01-03
#> 3 Pierre Salat Josette De La Torre 7 f 2023-01-01 2023-01-04
#> 4 Pierre Salat Carsen Harris 75 f 2023-01-02 2023-01-04
#> 5 Pierre Salat Ruth Anaya 49 f 2022-12-29 2023-01-04
#> 6 Pierre Salat Fat'hiyaa el-Salam 1 f 2022-12-29 2023-01-03
#> was_case status
#> 1 Y case
#> 2 N under_followup
#> 3 N under_followup
#> 4 Y case
#> 5 N under_followup
#> 6 Y caseThe contact_distribution requires a density function instead of a random number generation function (i.e. dpois() or dnbinom() instead of rpois() or rnbinom()). This is due to the branching process simulation adjusting the sampling of contacts to take into account the random network effect.
The same approach of using anonymous functions can be used in sim_linelist() and sim_contacts().
Simulating without hospitalisations and/or deaths
The onset-to-hospitalisation (onset_to_hosp) and onset-to-death (onset_to_death) arguments can also be set to NA in which case the date of admission ($date_admission) and date of death ($date_death) column in the line list will contains NAs.
Code
linelist <- sim_linelist(
contact_distribution = contact_distribution,
infect_period = infect_period,
prob_infect = 0.5,
onset_to_hosp = NA,
onset_to_death = NA,
hosp_risk = NA,
hosp_death_risk = NA,
non_hosp_death_risk = NA
)
head(linelist)
#> id case_name case_type sex age date_onset date_admission
#> 1 1 Jeffree O'Blia probable m 79 2023-01-01 <NA>
#> 2 2 Daifallah el-Popal probable m 79 2023-01-01 <NA>
#> 3 3 Tuhfa al-Morad probable f 80 2023-01-01 <NA>
#> 4 8 Margarita Goodbear confirmed f 50 2023-01-01 <NA>
#> 5 9 Emma Yang probable f 44 2023-01-01 <NA>
#> 6 12 Farajallah al-Baddour confirmed m 10 2023-01-02 <NA>
#> outcome date_outcome date_first_contact date_last_contact ct_value
#> 1 recovered <NA> <NA> <NA> NA
#> 2 recovered <NA> 2023-01-01 2023-01-03 NA
#> 3 recovered <NA> 2022-12-30 2023-01-02 NA
#> 4 recovered <NA> 2022-12-31 2023-01-04 23.1
#> 5 recovered <NA> 2022-12-31 2023-01-02 NA
#> 6 recovered <NA> 2023-01-01 2023-01-03 23.1This same functionality also applies to sim_outbreak(). In the above example, hosp_risk, hosp_death_risk and non_hosp_death_risk are set to NA. If the risk (*_risk) arguments are left as numeric inputs but the corresponding onset-to-event distribution (i.e. hosp_risk for onset_to_hosp and hosp_death_risk and non_hosp_death_risk for onset_to_death) are set to NA a warning will be produced. The example above simulates with neither hospitalisation or deaths, but these do not need to be turned off together, and one or the other can be set to NA with their corresponding risk arguments.