getting-started.RmdWork in progress
EpiNow is designed to be used at scale with few changes to the defaults and a single function call or to be used in an ad-hoc fashion via individual function calls. See the quick start section for a dicussion of the wrapper functions available. For more on using each function see the function documentation and introductory vignette.
EpiNow assumes that limited information is available on the dates of onset for cases and so is primarily based on using case counts by date of report. As imported cases may bias estimates of the time-varying reproduction number we support analysis that is based both on local cases and local + imported cases (where imported cases are used to adjust the local reproduction number estimates but are not otherwise used). An example of the required format is supplied in the EpiSoon package (see below). Here data manipulation is done using data.table but this is strictly optional and EpiNow makes no assumptions about the form of the inputted data (except that it has the columns required).
library(EpiNow) library(EpiSoon) library(data.table)
cases <- data.table::as.data.table(EpiSoon::example_obs_cases) ## Here just using local cases if also using imported cases define imports as "imported" cases <- cases[, `:=`(confirm = as.integer(cases), import_status = "local")] tail(cases) #> cases date confirm import_status #> 1: 296.0 2020-03-17 296 local #> 2: 343.0 2020-03-18 343 local #> 3: 399.5 2020-03-19 399 local #> 4: 454.0 2020-03-20 454 local #> 5: 605.5 2020-03-21 605 local #> 6: 367.5 2020-03-22 367 local
In order to map from cases by date of report to cases by date of infection information is needed on reporting delay, and incubation period. This may either be specified (with uncertainty) in the format EpiNow supports or can be derived via fitting to supplied data. We first cover specifying the distirbutions.
Here we define a lognormal distribution for reporting delay with a mean of 5 days (SD: 1 day) and a standard deviation of 3 days (SD: 1 day). Internally this uncertainty is then sampled to produce a data.frame that encodes a single distribution on each line in a format that can be processed by EpiNow::dist_skel (used internally when dealing with distributions).
Define the distribution
delay_def <- EpiNow::lognorm_dist_def(mean = 5, mean_sd = 1, sd = 3, sd_sd = 1, max_value = 30, samples = 10, to_log = TRUE) head(delay_def) #> model params max_value #> 1: lognorm <list[2]> 30 #> 2: lognorm <list[2]> 30 #> 3: lognorm <list[2]> 30 #> 4: lognorm <list[2]> 30 #> 5: lognorm <list[2]> 30 #> 6: lognorm <list[2]> 30
Example of sampling from one of the distributions
dist_skel(10, model = delay_def$model[1], params = delay_def$params[[1]]) #> [1] 3 5 0 3 12 4 0 1 12 5