The Problem

REDCap exports a “choose all that apply” question into a series of similarly-named, binary indicator variables (i.e., the variables are equal to either “checked” or “unchecked”). Using these variables individually, there is no obvious way to detect common patterns that people pick together.

Example: In the Nacho Craving Index (NCI), respondents can indicate which of eight ingredients they are currently craving (i.e., Chips, Yellow cheese, Orange cheese, White cheese, Meat, Beans, Tomatoes, Peppers). These are exported into variables with names like ingredients___1, ingredients___2, etc.

In REDCap, it is simple to get a summary of those individual variables by using the “Data Exports, Reports, and Stats” application within the REDCap interface and selecting “Stats & Charts”. Once the data is in R, simple tables can be produced with the table() function, or beautiful tables can be created with the tabyl() and adorn_pct_formatting() functions from the janitor package. However, from these univariate tables, it is impossible to judge which patterns of answers are marked together. In the above example, using the univariate tables, it is difficult to tell what pecentage of people are craving both chips and yellow cheese.

redcap <- readRDS(file = "./redcap.rds")

# Chips
janitor::tabyl(redcap$ingredients___1) %>% 
  janitor::adorn_pct_formatting() %>% 
  knitr::kable()

redcap$ingredients___1	n	percent
Unchecked	21	70.0%
Checked	9	30.0%


# Yellow cheese
janitor::tabyl(redcap$ingredients___2) %>% 
  janitor::adorn_pct_formatting() %>% 
  knitr::kable()

redcap$ingredients___2	n	percent
Unchecked	23	76.7%
Checked	7	23.3%

Aside: Loading REDCap Data into R

The redcapAPI package can be used to load data directly into R. To learn more about it, take a look here. Normally the code to automatically pull data with an API includes a person’s secret code “key”. Because I want to keep this hidden, I have hidden this API key in my user profile and the code below includes a call to Sys.getenv() to grab the key. To learn more about working with APIs, look here. Also notice that the data is saved using the saveRDS() function. REDCap data loaded with the API has the variable labels added as an extra attribute. To allow this vignette to run without sharing my secret key, I have saved the data to the package website.

rcon <- redcapAPI::redcapConnection(
  url = 'https://redcap.miami.edu/api/', 
  token = Sys.getenv("NCI_API_Key")
)

redcap <- redcapAPI::exportRecords(rcon)

saveRDS(redcap, file = "redcap.rds")

Make Analysis Data

Even after subsetting the REDCap data to only include the ingredients variables, it is still difficult to detect common patterns in the eight ingredients.

redcap <- readRDS(file = "./redcap.rds")

analysis <- redcap %>% 
  select(starts_with("ingredients___")) 
  
knitr::kable(tail(analysis))

	ingredients___1	ingredients___2	ingredients___3	ingredients___4	ingredients___5	ingredients___6	ingredients___7	ingredients___8
25	Checked	Checked	Unchecked	Unchecked	Unchecked	Checked	Unchecked	Unchecked
26	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked
27	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked
28	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked
29	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked	Unchecked
30	Checked	Checked	Unchecked	Unchecked	Unchecked	Unchecked	Checked	Unchecked

The Solution

Default Lettering

The make_binary_word() function combines responses from the individual variables into a single “word” that indicates which choices were selected. For example, if the first option from the NCI ingredient question, chips (i.e., ingredients___1), was checked, the word created by make_binary_word() will begin with a; or if it was not checked, the word will begin with _. If the second option, Yellow cheese (i.e., ingredients___2), was checked, the next letter will be a b, otherwise a _ will be used as a placeholder. Following this pattern, if somebody is not craving any of the eight nacho ingredients, the “word” will be eight underscores, one for each ingredient (i.e., ________). Conversely, if they are craving every ingredient, the “word” will be abcdefgh.

patterns <- make_binary_word(analysis) 
janitor::tabyl(patterns)
#>  patterns  n    percent
#>  ________ 20 0.66666667
#>  ______gh  1 0.03333333
#>  a_c__f_h  1 0.03333333
#>  a_cdefgh  1 0.03333333
#>  ab____g_  1 0.03333333
#>  ab___f__  1 0.03333333
#>  ab___f_h  1 0.03333333
#>  ab__efgh  1 0.03333333
#>  ab_de_gh  1 0.03333333
#>  ab_defgh  1 0.03333333
#>  abcdef_h  1 0.03333333

Custom Lettering

While the default lettering is somewhat useful, using meaningful (mnemonic) letters makes the binary words easier to understand. In this case, the first letter for each choice can be used as a useful mnemonic.

Abbreviation	Ingredient
C	Chips
Y	Yellow cheese
O	Orange cheese
W	White cheese
M	Meat
B	Beans
T	Tomatoes
P	Peppers

To use custom lettering, specify a vector of single letter abbreviations and pass it to the the_labels argument. Be sure to include one unique abbreviation for each column of the data frame. For example:

labels <- c("C", "Y", "O", "W", "M", "B", "T", "P")

patterns <- make_binary_word(analysis, the_labels = labels)

janitor::tabyl(patterns)
#>  patterns  n    percent
#>  CYOWMB_P  1 0.03333333
#>  CY_WMBTP  1 0.03333333
#>  CY_WM_TP  1 0.03333333
#>  CY__MBTP  1 0.03333333
#>  CY___B_P  1 0.03333333
#>  CY___B__  1 0.03333333
#>  CY____T_  1 0.03333333
#>  C_OWMBTP  1 0.03333333
#>  C_O__B_P  1 0.03333333
#>  ______TP  1 0.03333333
#>  ________ 20 0.66666667