Skip to content

Latest commit

 

History

History
231 lines (190 loc) · 8.9 KB

README.md

File metadata and controls

231 lines (190 loc) · 8.9 KB
Raw

README

The estimateR package provides tools to estimate the effective reproductive number through time from delayed and indirect observations of infection events. This package is currently in a beta version.

Installation

First, make sure you have installed the devtools package locally. If not, run:

install.packages("devtools")

in RStudio or other.

Then run:

library(devtools)
install_github("covid-19-Re/estimateR")

Documentation

The full documentation is available here.

Quick example

We demonstrate below a basic use of the estimateR package. Check out the estimateR vignette and additional vignettes for more details.

We start with arbitrary incidence data representing daily counts of confirmed cases for a disease of interest. This incidence data counts the number of people tested positive for a particular infectious disease and reported on day t.

library(estimateR)
date_first_data_point <- as.Date("2020-02-24")
toy_incidence_data <- c(4,9,19,14,36,16,39,27,46,
                          77,78,113,102,134,165,183,
                          219,247,266,308,304,324,346,
                          348,331,311,267,288,254,239)

Below we explain the required user input. It contains assumptions that need to be adapted to the particular disease and setting studied.

Reporting delay

In this toy example, we assume that the incidence data represents cases by date of report. To infer the time series of cases by date of symptom onset, we need an assumption about the reporting delay, i.e. the time between symptom onset and reporting. This delay is specific to each reporting system and can be estimated from data by public health authorities (e.g. from line list data). The delay can vary for each case and is therefore described by a delay distribution.

Here we assume that the delay is Gamma distributed with the following parameters:

shape_onset_to_report = 2.7
scale_onset_to_report = 1.6
onset_to_report <- list(name="gamma", shape = shape_onset_to_report, scale = scale_onset_to_report)

Incubation period

With the reporting delay, we can estimate the time series of cases by date of symptom onset. However, this is still delayed from the actual time series of infections. We also need to account for an incubation period, i.e. the time between infection and symptom onset in patients. Such information is typically obtained from clinical studies. The incubation period varies between patients and is therefore described by a distribution.

Here we assume that the incubation period is Gamma distributed with the following parameters:

shape_incubation = 3.2 
scale_incubation = 1.3
incubation <- list(name="gamma", shape = shape_incubation, scale = scale_incubation)

Generation time / serial interval

To estimate the effective reproductive number from the time series of infections, we need an assumption about the generation interval, i.e. the time from infection of a primary case to the infection of a secondary case. In theory, we are interested in the so-called intrinsic generation interval distribution, which describes the average infectiousness of a patient over time and does not depend on the epidemic phase. In practice however, estimates of the serial interval (time between symptom onset of the primary case and symptom onsets of the secondary case) is used as a proxy for the generation interval. Estimates of the serial interval distribution are typically obtained from household or contact-tracing studies.

Please note that the validity of using the serial interval as a proxy for the generation interval depends on specific assumptions about disease progression and infection, which may be more or less violated for a certain disease.

In estimateR, we assume that the generation interval / serial interval is gamma distributed. We supply the mean and standard deviation of this gamma distribution:

mean_serial_interval = 4.8
std_serial_interval = 2.3

Note that we specify these parameters in the same time unit as the time steps of the original observation data. For instance, if the original data represents daily reports (as in this toy example), then the parameters must be specified in days.

Estimation window

The estimation window corresponds to the size of the smoothing window used in EpiEstim. See help(estimate_Re) for additional details. By default, it is set to three days. The wider the window, the stronger the smoothing.

estimation_window = 3

Estimating the effective reproductive number

Now we have provided all relevant disease- and setting-specific parameters. With the function estimate_Re_from_noisy_delayed_incidence(), we can compute the reproductive number through time from the incidence data and our further specifications.

toy_estimates <- estimate_Re_from_noisy_delayed_incidence(toy_incidence_data,
  smoothing_method = "LOESS",
  deconvolution_method = "Richardson-Lucy delay distribution",
  estimation_method = "EpiEstim sliding window",
  delay = list(incubation, onset_to_report),
  estimation_window = estimation_window,
  mean_serial_interval = mean_serial_interval,
  std_serial_interval  = std_serial_interval,
  output_Re_only = FALSE,
  ref_date = date_first_data_point,
  time_step = "day"
)

Let’s look at the most recent Re estimates:

tail(toy_estimates, n = 20)
#>          date observed_incidence smoothed_incidence deconvolved_incidence
#> 19 2020-03-05                 78           104.7215              233.9893
#> 20 2020-03-06                113           119.7628              245.8158
#> 21 2020-03-07                102           135.7005              257.2442
#> 22 2020-03-08                134           152.7363              266.8594
#> 23 2020-03-09                165           171.0718              275.6173
#> 24 2020-03-10                183           187.9392              284.3650
#> 25 2020-03-11                219           201.4078              293.8851
#> 26 2020-03-12                247           212.9347              303.3842
#> 27 2020-03-13                266           223.9770              311.9033
#> 28 2020-03-14                308           235.9919              319.9628
#> 29 2020-03-15                304           247.8836              328.0520
#> 30 2020-03-16                324           258.1096              336.6381
#> 31 2020-03-17                346           267.4566                    NA
#> 32 2020-03-18                348           276.7116                    NA
#> 33 2020-03-19                331           286.6611                    NA
#> 34 2020-03-20                311           296.5506                    NA
#> 35 2020-03-21                267           305.4301                    NA
#> 36 2020-03-22                288           313.7929                    NA
#> 37 2020-03-23                254           322.1324                    NA
#> 38 2020-03-24                239           330.9420                    NA
#>    Re_estimate
#> 19    1.531783
#> 20    1.442508
#> 21    1.374039
#> 22    1.321471
#> 23    1.278438
#> 24    1.242481
#> 25    1.215222
#> 26    1.195959
#> 27    1.181623
#> 28    1.169092
#> 29    1.157416
#> 30    1.147870
#> 31          NA
#> 32          NA
#> 33          NA
#> 34          NA
#> 35          NA
#> 36          NA
#> 37          NA
#> 38          NA
  • The column observed_incidence shows the originally reported case counts.
  • The column smoothed_incidence shows the case counts after smoothing.
  • The column deconvolved_incidence shows the estimated number of infections.
  • The column Re_estimate shows point estimates for the reproductive number.

Why are there missing values (NAs)?

We see that no estimates are available for the 8 most recent days - the earliest available estimate is 9 days delayed. This is because the reporting delay and incubation period prevent us from inferring the number of infections for the most recent past. If individuals are infected today, it will take time until they develop symptoms and are reported as cases. Thus, cases reported today mostly tell us something about past infection, and not much about infections today. As a result, the reproductive number can only be estimated with a certain delay. This delay may be partially reduced depending on what data is available, and through the use of nowcasting. See this vignette for more details.

How to cite?

When using this package for a publication, please cite: