SignalDetectionTool: R Package for signal detection on infectious disease surveillance data

The SignalDetectionTool (SDT) is an R package which contains an R shiny app to perform signal detection on infectious disease surveillance data and visualise results. Furthermore the functions of this R package can also be used standalone without running the shiny application. This vignette gives guidance on how to use the functions within the R package.

Running the App

This section is extensively described in the README of the package thus we give minimal instructions here. First install and load the SDT package.

library(SignalDetectionTool)

Run the app entering the following command in the console.

It is possible to run the app with customised settings using a yaml file. Please have a look at the README.

Using functions from the SDT

Data

The input to the SDT is a linelist of infectious disease cases structured according to the format defined in input_metadata. A sample linelist (input_example) and its corresponding metadata specification (input_metadata) is available in the package as internal datasets and can be accessed using:

data("input_example")
data("input_metadata")

Applying signal detection

To apply signal detection to your data for the most recent 6 weeks you first need to preprocess your linelist and then apply get_signals(). Inside get_signals() the linelist is aggregated to a timeseries containing weekly counts of cases.
To apply different signal detection methods use the parameter method. To generate stratified signals use the parameter stratification. Please have a look at the function documentation for more details. This is an example for signals generated using the default paramters with FarringtonFlexible and no stratification:

data_prepro <- input_example %>% preprocess_data()
signals <- data_prepro %>% get_signals()

The generated signals output is a data.frame containing the aggregated number of cases per week with additional columns cases_in_outbreak, alarms, upperbound, expected, category, stratum, method and number_of_weeks.

  • cases_in_outbreak is only generated if the column outbreak_status was available in your provided linelist. It counts how many cases in this week were already part of a known outbreak.
  • alarms is a column with booleans indicating for which weeks a signal has been generated. Of course the majority of weeks in the aggregated timeseries contain NA in this column as only the most recent selected number_of_weeks are filled with TRUE of FALSE.
  • upperbound contains a numeric threshold calculated with the signal detection methods. As for alarms the majority of weeks in the aggregated timeseries contain NA as this is only calculated for the number_of_weeks for which signals are generated. Furthermore some algorithms such as EARS do not calculate a threshold.
  • expected is a numeric value giving the expected number of cases by the signal detection algorithm. Only filled for the weeks for which signals are generated.
  • category Character string specifying to which stratum timeseries belongs to when having added stratification parameters (more details in the next example). For unstratified timeseries it is NA.
  • stratum Character string specifying to which category of the stratum the timeseries belongs to when having added stratification parameters (more details in the next example). For unstratified timeseries it is NA.
  • method Character string specifying the signal detection method which was used.
  • number_of_weeks Integer specifying the number of weeks for which signals were generated.

Generating stratified signals using EARS:

data_prepro <- input_example %>% preprocess_data()
signals_ears_stratified <- data_prepro %>% get_signals(
  method = "ears",
  stratification = c("age_group", "county")
)

The signals_ears_stratified output is a data frame in long format, containing multiple time series of weekly aggregated case counts. Each time series corresponds to a unique combination of the specified stratification variables, with all strata stacked vertically.

The columns category and stratum identify the individual timeseries. In this example we obtain the following values:
* category is filled with the column names provided in the stratification parameter.
* stratum with the values of these variables.

In this example we obtain:
* category is filled with “age_group” and “county”
* stratum contains “00–04”, “05–09”, “10–14”, “15–19”, “20–24”, “25–29”, “30–34”, “35–39”, “40–44”, “45–49”, “50–54”, “55–59”, “60–64”, “65–69”, “70–74”, “75–79”, “80–84”, “85–89”, “90–94”, “95–99”, “100–104”, “105–109” for rows with category == "age_group". For rows with category == "county" the values of stratum are “Burgenland”, “Kärnten”, “Niederösterreich”, “Oberösterreich”, “Salzburg”, “Steiermark”, “Tirol”, “Vorarlberg”, “Wien”.

Create visualisations and tables

Coming soon…

Run the report

You can directly generate an HTML or Word report containing signal results and visualizations using run_report(). Simply pass your surveillance linelist—structured according to the format defined in input_metadata—to run_report(). The data will be preprocessed automatically within the function. The following code chunk generates a Word report using EARS without any stratification (default) for the past six weeks (default):

run_report(input_example,
  report_format = "DOCX",
  method = "EARS"
)

This generates an HTML report (default) with stratification by age group, county and sex for the last 2 weeks using FarringtonFlexible (default):

run_report(input_example,
  strata = c("age_group", "county", "sex"),
  number_of_weeks = 2
)