An introduction to linelist

library(linelist)

Motivations

Outbreak analytics pipelines often start with case line lists, which are data tables in which every line is a different case/patient, and columns record different variables of potential epidemiological interest such as date of events (e.g. onset of symptom, case notification), disease outcome, or patient data (e.g. age, sex, occupation). Such data is typically held in a data.frame (or a tibble) and used in various downstream analysis. While this approach is functional, it often means that each analysis step will:

  1. need to check the required inputs are present in the data, and for the user to specify where (e.g. ‘This is the column where dates of onset are stored.’)

  2. need to validate the required data (e.g. ‘Check that the field storing dates of onset are indeed dates, and not a character.’)

The aim of linelist is to take care of these pre-requisites once and for all before downstream analyses, thus helping to make data pipelines more robust and straightforward.

linelist in a nutshell

Outline

linelist is an R package which implements basic data representation for case line lists, alongside accessors and basic methods. It essentially provides three types of functionalities:

  1. tagging: a tags system permits to pre-identify key epidemiological variables needed in downstream analyses (e.g. dates of case notification, symptom onset, age, gender, disease outcome)

  2. validation: functions checking that tagged variables are indeed present in the data.frame/tibble, and that they have the expected type (e.g. checking that dates are Date, integer or numeric)

  3. secured methods: generic functions which could lead to the loss of tagged variables have dedicated methods for linelist objects with adapted behaviours, either updating tags as needed (e.g. rename(), names() <- ...) or issuing warnings/errors when tagged variables are lost (e.g. select(), x[], x[[]])

Should I use linelist?

linelist is designed to add a robust, foundational layer to your data pipelines, but it might add unnecessary complexity to your analysis scripts. Here are a few hints to gauge if you should consider using the package.

You may have use for linelist if …:

  • your data changes/updates over time (e.g. new entries, new variables, renamed variables)

  • you build data pipelines entailing multiple layers of data processing and analysis

  • you are looking to build re-useable analysis scripts, i.e. which will work on other datasets with minimal added changes

Conversely, you probably do not need it if …:

  • you work on historical data, which has likely already been curated/validated and will no longer change

  • you perform some quick, simple analysis of your data, which you will not need to expand on later

  • your analysis scripts are very specific and will not be re-used elsewhere

Getting started

Installation

Our stable versions are released periodically on CRAN, and can be installed using:

install.packages("linelist", build_vignettes = TRUE)

If you prefer using the latest features and bug fixes, you can alternatively install the development version of linelist from GitHub using the following commands:

if (!require(remotes)) {
  install.packages("remotes")
}
remotes::install_github("epiverse-trace/linelist", build_vignettes = TRUE)

Once installed, you can load the package in your R session using:

library(linelist)

Key functionalities

A linelist object is an instance of a data.frame or a tibble in which key epidemiological variables have been tagged. The main features of the packages are broken down into the 3 categories outlined above.

Tagging system

Tags are paired keys pointing a reference epidemiological variables to the name of a column in a data.frame or tibble. The tagging system permits to construct linelist objects, modify tags in existing objects, check and access existing tags and the corresponding variables.

  • make_linelist(): to create a linelist object by tagging key epi variables in a data.frame or a tibble

  • set_tags(): to add, remove, or modify tags in a linelist

  • tags(): to list variables which have been tagged in a linelist

  • tags_names(): to list all recognized tag names; details on what the tags represent can be found at ?make_linelist

  • tags_df(): to obtain a data.frame of all the tagged variables in a linelist

Validation

Basic routines are provided to validate linelist objects. More advanced validation e.g. looking at compatibility of dated events will be implemented in a separate package.

  • validate_tags(): check that tagged variables are present in the dataset, that tags match the pre-defined list of tagged variables

  • validate_types(): check that tagged variables have an acceptable class, as defined in tags_types()

  • validate_linelist(): general validation of linelist objects, equivalent to running both validate_tags() and validate_types(), and checking the class of the object

Secured methods

These are dedicated S3 methods for existing generics which can be used to prevent the loss of tagged variables.

  • lost_tags_action(): to set the behaviour to adopt when tagged variables would be lost by an operation: issue a warning (default), an error, or ignore

  • get_lost_tags_action(): to check the current behaviour for lost tagged variables

  • names<-(): the ‘base R’ approach to renaming columns of a linelist; will rename tags as needed to match the new column names

  • x[] and x[[]]: for subsetting columns using ‘base R’ syntax; will behave according to get_lost_tags_actions() if tagged variables are lost

Worked example

Example dataset

In this example, we use the case line list of the Hagelloch 1861 measles outbreak, distributed by the outbreaks package as measles_hagelloch_1861 .

data(measles_hagelloch_1861, package = "outbreaks")

# overview of the data
head(measles_hagelloch_1861)
#>   case_ID infector date_of_prodrome date_of_rash date_of_death age gender
#> 1       1       45       1861-11-21   1861-11-25          <NA>   7      f
#> 2       2       45       1861-11-23   1861-11-27          <NA>   6      f
#> 3       3      172       1861-11-28   1861-12-02          <NA>   4      f
#> 4       4      180       1861-11-27   1861-11-28          <NA>  13      m
#> 5       5       45       1861-11-22   1861-11-27          <NA>   8      f
#> 6       6      180       1861-11-26   1861-11-29          <NA>  12      m
#>   family_ID class complications x_loc y_loc
#> 1        41     1           yes 142.5 100.0
#> 2        41     1           yes 142.5 100.0
#> 3        41     0           yes 142.5 100.0
#> 4        61     2           yes 165.0 102.5
#> 5        42     1           yes 145.0 120.0
#> 6        42     2           yes 145.0 120.0

Creating a linelist object

Let us assume we want to tag the following variables to facilitate downstream analyses, after having checked their tag name in ?make_linelist:

  • the date of symptom onset, here called prodrome (tag: date_onset)
  • the date of death (tag: date_death)
  • the age of the patient (tag: age)
  • the gender of the patient (tag: gender)

We first load a few useful packages, and create a linelist with the above information:

library(tibble) # data.frame but with nice printing
library(dplyr) # for data handling
library(magrittr) # for the %>% operator
library(linelist) # this package!

x <- measles_hagelloch_1861 %>%
  tibble() %>%
  make_linelist(date_onset = "date_of_prodrome",
                date_death = "date_of_death",
                age = "age",
                gender = "gender")
head(x)
#> 
#> // linelist object
#> # A tibble: 6 × 12
#>   case_ID infector date_of_prodrome date_of_rash date_of_death   age gender
#>     <int>    <int> <date>           <date>       <date>        <dbl> <fct> 
#> 1       1       45 1861-11-21       1861-11-25   NA                7 f     
#> 2       2       45 1861-11-23       1861-11-27   NA                6 f     
#> 3       3      172 1861-11-28       1861-12-02   NA                4 f     
#> 4       4      180 1861-11-27       1861-11-28   NA               13 m     
#> 5       5       45 1861-11-22       1861-11-27   NA                8 f     
#> 6       6      180 1861-11-26       1861-11-29   NA               12 m     
#> # ℹ 5 more variables: family_ID <int>, class <fct>, complications <fct>,
#> #   x_loc <dbl>, y_loc <dbl>
#> 
#> // tags: date_onset:date_of_prodrome, date_death:date_of_death, gender:gender, age:age

The printing of the object confirms that the tags have been added. If we want to double-check which variables have been tagged:

tags(x)
#> $date_onset
#> [1] "date_of_prodrome"
#> 
#> $date_death
#> [1] "date_of_death"
#> 
#> $gender
#> [1] "gender"
#> 
#> $age
#> [1] "age"

Changing tags

Tags can be added / removed / changed using set_tags().

Let us assume we also want to record the outcome: it is currently missing, but can be built from dates of deaths (missing date = survived). This can be done by using mutate() on x to create the new variable (remember x is not only a linelist, but also a regular tibble and this is compatible with dplyr verbs), and setting up a new tag using set_tags():

x <- x %>%
  mutate(
    inferred_outcome = if_else(is.na(date_of_death), "survived", "died")
  ) %>%
  set_tags(outcome = "inferred_outcome")
x
#> 
#> // linelist object
#> # A tibble: 188 × 13
#>    case_ID infector date_of_prodrome date_of_rash date_of_death   age gender
#>      <int>    <int> <date>           <date>       <date>        <dbl> <fct> 
#>  1       1       45 1861-11-21       1861-11-25   NA                7 f     
#>  2       2       45 1861-11-23       1861-11-27   NA                6 f     
#>  3       3      172 1861-11-28       1861-12-02   NA                4 f     
#>  4       4      180 1861-11-27       1861-11-28   NA               13 m     
#>  5       5       45 1861-11-22       1861-11-27   NA                8 f     
#>  6       6      180 1861-11-26       1861-11-29   NA               12 m     
#>  7       7       42 1861-11-24       1861-11-28   NA                6 m     
#>  8       8       45 1861-11-21       1861-11-26   NA               10 m     
#>  9       9      182 1861-11-26       1861-11-30   NA               13 m     
#> 10      10       45 1861-11-21       1861-11-25   NA                7 f     
#> # ℹ 178 more rows
#> # ℹ 6 more variables: family_ID <int>, class <fct>, complications <fct>,
#> #   x_loc <dbl>, y_loc <dbl>, inferred_outcome <chr>
#> 
#> // tags: date_onset:date_of_prodrome, date_death:date_of_death, gender:gender, age:age, outcome:inferred_outcome

If we wanted to undo the above, i.e. remove the outcome tag, we only need to set it to NULL:

x <- x %>%
  set_tags(outcome = NULL)
tags(x)
#> $date_onset
#> [1] "date_of_prodrome"
#> 
#> $date_death
#> [1] "date_of_death"
#> 
#> $gender
#> [1] "gender"
#> 
#> $age
#> [1] "age"

Accessing tagged variables

Now that key variables have been tagged in x, we can used these pre-defined fields in downstream analyses, without having to worry about variable names and types. We could access tagged variables using any of the following means:

# select tagged variables only
x %>%
  select(has_tag(c("date_onset", "date_death")))
#> Warning: The following tags have lost their variable:
#>  gender:gender, age:age
#> 
#> // linelist object
#> # A tibble: 188 × 2
#>    date_of_prodrome date_of_death
#>    <date>           <date>       
#>  1 1861-11-21       NA           
#>  2 1861-11-23       NA           
#>  3 1861-11-28       NA           
#>  4 1861-11-27       NA           
#>  5 1861-11-22       NA           
#>  6 1861-11-26       NA           
#>  7 1861-11-24       NA           
#>  8 1861-11-21       NA           
#>  9 1861-11-26       NA           
#> 10 1861-11-21       NA           
#> # ℹ 178 more rows
#> 
#> // tags: date_onset:date_of_prodrome, date_death:date_of_death

# select tagged variables only with renaming on the fly
x %>%
  select(onset = has_tag("date_onset"))
#> Warning: The following tags have lost their variable:
#>  date_death:date_of_death, gender:gender, age:age
#> 
#> // linelist object
#> # A tibble: 188 × 1
#>    onset     
#>    <date>    
#>  1 1861-11-21
#>  2 1861-11-23
#>  3 1861-11-28
#>  4 1861-11-27
#>  5 1861-11-22
#>  6 1861-11-26
#>  7 1861-11-24
#>  8 1861-11-21
#>  9 1861-11-26
#> 10 1861-11-21
#> # ℹ 178 more rows
#> 
#> // tags: date_onset:onset

# get all tagged variables in a data.frame
x %>%
  tags_df()
#> # A tibble: 188 × 4
#>    date_onset date_death gender   age
#>    <date>     <date>     <fct>  <dbl>
#>  1 1861-11-21 NA         f          7
#>  2 1861-11-23 NA         f          6
#>  3 1861-11-28 NA         f          4
#>  4 1861-11-27 NA         m         13
#>  5 1861-11-22 NA         f          8
#>  6 1861-11-26 NA         m         12
#>  7 1861-11-24 NA         m          6
#>  8 1861-11-21 NA         m         10
#>  9 1861-11-26 NA         m         13
#> 10 1861-11-21 NA         f          7
#> # ℹ 178 more rows

Using safeguards

Because x remains a valid tibble, we can use any data handling operations implemented in dplyr. However, some of these operations may cause accidental removal of key tagged variables. linelist provides a safeguard mechanism against this. For instance, let’s assume we want to select only some columns of x:

x %>%
  select(1:2)
#> Warning: The following tags have lost their variable:
#>  date_onset:date_of_prodrome, date_death:date_of_death, gender:gender, age:age
#> 
#> // linelist object
#> # A tibble: 188 × 2
#>    case_ID infector
#>      <int>    <int>
#>  1       1       45
#>  2       2       45
#>  3       3      172
#>  4       4      180
#>  5       5       45
#>  6       6      180
#>  7       7       42
#>  8       8       45
#>  9       9      182
#> 10      10       45
#> # ℹ 178 more rows
#> 
#> // tags: [no tagged variable]

Here, the above command gave a meaningful warning, in which select() removes some of the variables that were tagged.

We can also use the has_tag() select helper to select columns via their tag. For example, to retain the first 2 variables, and the gender tag:

# hybrid selection
x %>%
  select(1:2, has_tag("gender"))
#> Warning: The following tags have lost their variable:
#>  date_onset:date_of_prodrome, date_death:date_of_death, age:age
#> 
#> // linelist object
#> # A tibble: 188 × 3
#>    case_ID infector gender
#>      <int>    <int> <fct> 
#>  1       1       45 f     
#>  2       2       45 f     
#>  3       3      172 f     
#>  4       4      180 m     
#>  5       5       45 f     
#>  6       6      180 m     
#>  7       7       42 m     
#>  8       8       45 m     
#>  9       9      182 m     
#> 10      10       45 f     
#> # ℹ 178 more rows
#> 
#> // tags: gender:gender

Again, we observe a warning as before due to the loss of tagged variables in the operation. This behaviour can be silenced if needed, or could be changed to issue an error (for stronger pipelines for instance):

# hybrid selection
x %>%
  select(1:2, has_tag("gender"))
#> Warning: The following tags have lost their variable:
#>  date_onset:date_of_prodrome, date_death:date_of_death, age:age
#> 
#> // linelist object
#> # A tibble: 188 × 3
#>    case_ID infector gender
#>      <int>    <int> <fct> 
#>  1       1       45 f     
#>  2       2       45 f     
#>  3       3      172 f     
#>  4       4      180 m     
#>  5       5       45 f     
#>  6       6      180 m     
#>  7       7       42 m     
#>  8       8       45 m     
#>  9       9      182 m     
#> 10      10       45 f     
#> # ℹ 178 more rows
#> 
#> // tags: gender:gender

# hybrid selection - no warning
lost_tags_action("none")
#> Lost tags will now be ignored.

x %>%
  select(1:2, has_tag("gender"))
#> 
#> // linelist object
#> # A tibble: 188 × 3
#>    case_ID infector gender
#>      <int>    <int> <fct> 
#>  1       1       45 f     
#>  2       2       45 f     
#>  3       3      172 f     
#>  4       4      180 m     
#>  5       5       45 f     
#>  6       6      180 m     
#>  7       7       42 m     
#>  8       8       45 m     
#>  9       9      182 m     
#> 10      10       45 f     
#> # ℹ 178 more rows
#> 
#> // tags: gender:gender

# hybrid selection - error due to lost tags
lost_tags_action("error")
#> Lost tags will now issue an error.

x %>%
  select(1:2, has_tag("gender"))
#> Error: The following tags have lost their variable:
#>  date_onset:date_of_prodrome, date_death:date_of_death, age:age

# note that `lost_tags_action` sets the behavior for any later operation, so we
# need to reset the default
get_lost_tags_action() # check current behaviour
#> [1] "error"
lost_tags_action() # reset default
#> Lost tags will now issue a warning.

Changing tag loss action permanently

If you wish to change the lost_tags_action in a way that persists across R sessions, you can do so by setting the LINELIST_LOST_ACTION environment variable. For example, your .Renviron file could contain the following line:

LINELIST_LOST_ACTION="error"