recapFactorFlip equivalent

[Wunsei]

redcapAPI has the helpful redcapFactorFlip function. Does REDCapR have an equivalent? If no, how might I replicate the same thing?

[wibeasley]

yeah, @nutterb has done some cool things in redcapAPI.

Here's the code for his function: https://github.com/nutterb/redcapAPI/blob/master/R/redcapFactorFlip.R. I haven't studied it too closely, but it looks like he gets the raw values & labels and makes them factors. Here are the parts that strike me as most relevant: https://github.com/nutterb/redcapAPI/search?q=redcapLabels

Regarding how to replicate it in REDCapR, I'm guess it would start with REDCapR::redcap_metadata_read() --run that example from your own machine to see get a feel for the returned dataset. The selection_choice_or_calculations column exposes values like "0, Female | 1, Male" for gender and "0, Unknown / Not Reported | 1, NOT Hispanic or Latino | 2, Hispanic or Latino" for ethnicity. That could be split at each pipe (ie, |), and a regex could pull out the number and the value.

Here's a proof of concept that I haven't tested. Suppose the relevant parts of the metadata dataset are:

ds <- 
  tibble::tribble(
    ~field_name, ~select_choices_or_calculations,
    "record_id", NA_character_,
    "age"      , NA_character_,
    "gender"   , "0, Female | 1, Male",
    "race"     , "1, American Indian/Alaska Native | 2, Asian | 3, Native Hawaiian or Other Pacific Islander | 4, Black or African American | 5, White | 6, Unknown / Not Reported",
    "ethnicity", "0, Unknown / Not Reported | 1, NOT Hispanic or Latino | 2, Hispanic or Latino" 
)

Something like this would extract each level. Notice rematch2, which is one of my favorite & underappreciated packages, does the hard work.

pattern <- "^(?<level>\\d+),(?<label>.+)$"
ds |> 
  dplyr::select(
    field     = field_name,
    choice    = select_choices_or_calculations
  ) |> 
  tidyr::drop_na(choice) |> 
  tidyr::separate_rows(
    choice, 
    sep     = " \\| "
  ) |> 
  rematch2::bind_re_match(
    choice, 
    pattern
  )

Here's resulting dataset, which (I think) can be piped into a purrr function to apply the levels & labels to each factor variable.

       field                                       choice level                                      label
1     gender                                    0, Female     0                                     Female
2     gender                                      1, Male     1                                       Male
3       race             1, American Indian/Alaska Native     1              American Indian/Alaska Native
4       race                                     2, Asian     2                                      Asian
5       race 3, Native Hawaiian or Other Pacific Islander     3  Native Hawaiian or Other Pacific Islander
6       race                 4, Black or African American     4                  Black or African American
7       race                                     5, White     5                                      White
8       race                    6, Unknown / Not Reported     6                     Unknown / Not Reported
9  ethnicity                    0, Unknown / Not Reported     0                     Unknown / Not Reported
10 ethnicity                    1, NOT Hispanic or Latino     1                     NOT Hispanic or Latino
11 ethnicity                        2, Hispanic or Latino     2                         Hispanic or Latino

Does that help? If there's enough interest, I'll put it into the package. I'd love fed back from anyone who is interested in this potential. @pbchase, you usually have an opinion?

[skadauke]

I also think being able to move between raw data and labels would be useful. We built the parse_labels function that takes a string from selection_choice_or_calculations and returns a tibble:

parse_labels <- function(string){
  out <- string %>%
    strsplit(" \\| |, ") %>% # split either by ' | ' or ', '
    unlist() %>%
    matrix(
      ncol = 2,
      byrow = TRUE,
      dimnames = list(
        c(),               # row names
        c("raw", "label")) # column names
    ) %>%
    dplyr::as_tibble()
  
  out
}

string <- "1, American Indian/Alaska Native | 2, Asian | 3, Native Hawaiian or Other Pacific Islander | 4, Black or African American | 5, White | 6, Unknown / Not Reported"

parse_labels(string)
#> # A tibble: 6 × 2
#>   raw   label                                    
#>   <chr> <chr>                                    
#> 1 1     American Indian/Alaska Native            
#> 2 2     Asian                                    
#> 3 3     Native Hawaiian or Other Pacific Islander
#> 4 4     Black or African American                
#> 5 5     White                                    
#> 6 6     Unknown / Not Reported 

This function should work well in a purrr context.

One improvement I would suggest to your above code is not to filter on choice == NA but on categorical field types (dropdown, radio, checkbox, +/- yesno, +/- truefalse) because that field is also used to report calculations for calculated field and I'm assuming you don't want those to go into your result set.