Strategy

As described in the Retrieving Longitudinal and Repeating Structures vignette, the best way to read and write data from projects with longitudinal/repeating elements is to break up the “block matrix” dataset into individual datasets. Each rectangle should have a coherent grain.

Following this strategy, we’ll write to the REDCap server in two distinct steps:

1. Upload the patient-level instrument(s)
2. Upload the each repeating instrument separately.

The actual upload phase is pretty straight-forward –it’s just a call to REDCapR::redcap_write(). Most of the vignette’s code prepares the dataset so that the upload will run smoothly.

Pre-requisites

See the Typical REDCap Workflow for a Data Analyst vignette and

1. Verify REDCapR is installed
2. Verify REDCap Access
3. Review Codebook

Retrieve Token

Please closely read the Retrieve Protected Token section, which has important security implications. The current vignette imports a fake dataset into REDCap, and we’ll use a token stored in a local file.

# retrieve-credential
path_credential <- system.file("misc/dev-2.credentials", package = "REDCapR")
credential  <- REDCapR::retrieve_credential_local(
  path_credential = path_credential,
  project_id      = 3748
)

c(credential$redcap_uri, credential$token)

Datasets to Write to Server

To keep this vignette focused on writing/importing/uploading to the server, we’ll start with the data that needs to be written. These example tables were prepared by Raymond Balise for our 2023 R/Medicine workshop, “Using REDCap and R to Rapidly Produce Biomedical Publications”.

There are two tables, each with a different granularity:

• ds_patient: each row represents one patient,
• ds_daily: each row represents one daily measurement per patient.

# load-patient
ds_patient <-
  "test-data/vignette-repeating-write/data-patient.rds" |>
  system.file(package = "REDCapR") |>
  readr::read_rds()

ds_patient

# load-repeating
ds_daily <-
  "test-data/vignette-repeating-write/data-daily.rds" |>
  system.file(package = "REDCapR") |>
  readr::read_rds()

ds_daily

Prep: Stoplight Fields

If the REDCap project isn’t longitudinal and doesn’t have arms, uploading a patient-level data.frame to REDCap doesn’t require adding variables. However we typically populate the *_complete variables to communicate the record’s status.

If the row is needs a human to add more values or inspect the existing values consider marking the instrument “incomplete” or “unverified”; the patient’s instrument record will appear red or yellow in REDCap’s Record Dashboard. Otherwise consider marking the instrument “complete” so it will appear green.

With this example project, the only patient-level instrument is “enrollment”, so the corresponding variable is enrollment_complete.

# patient-complete
ds_patient <-
  ds_patient |>
  dplyr::mutate(
    enrollment_complete   = REDCapR::constant("form_complete"),
  )

Prep: REDCapR::validate_for_write()

REDCapR::validate_for_write() inspects a data frame to anticipate potential problems before writing with REDCap’s API. A tibble is returned, with one row per potential problem (and a suggestion how to avoid it). Ideally an 0-row tibble is returned.

REDCapR::validate_for_write(ds_patient, convert_logical_to_integer = TRUE)

If you encounter problems that can be checked with automation, please tell us in an issue. We’ll work with you to incorporate the new check into REDCapR::validate_for_write().

When a dataset’s problems are caught before reaching the server, the solutions are easier to identify and implement.

Prep: Write Small Subset First

If this is your first time with a complicated project, consider loading a small subset of rows and columns. In this case, we start with only three columns and two rows.

# patient-subset
ds_patient |>
  dplyr::select(              # First three columns
    id_code,
    date,
    is_mobile,
  ) |>
  dplyr::slice(1:2) |>        # First two rows
  REDCapR::redcap_write(
    ds_to_write = _,
    redcap_uri  = credential$redcap_uri,
    token       = credential$token,
    convert_logical_to_integer = TRUE
  )

Prep: Recode Variables where Necessary

Some variables in the data.frame might be represented differently than in REDCap.

A common transformation is changing strings into the integers that underlie radio buttons. Common approaches are dplyr::case_match() and using joining to lookup tables (if the mappings are expressed in a csv). Here’s an in-line example of dplyr::case_match().

ds_patient <-
  ds_patient |>
  dplyr::mutate(
    race =
      dplyr::case_match(
        race,
        "White"                       ~  1L,
        "Black or African American"   ~  2L,
        "Asian"                       ~  3L,
        "Native American"             ~  4L,
        "Pacific Islander"            ~  5L,
        "Multiracial"                 ~  6L,
        "Refused or don't know"       ~  7L
      )
  )

Write Entire Patient-level Table

If the small subset works, we usually jump ahead and try all columns and rows.

If this larger table fails, split the difference between (a) the smaller working example and (b) the larger failing example. See if this middle point (that has fewer rows and/or columns than the failing point) succeeds or fails. Then repeat. This “bisection” or “binary search” debugging technique is helpful in many areas of programming and statistical modeling.

# patient-entire
ds_patient |>
  REDCapR::redcap_write(
    ds_to_write = _,
    redcap_uri  = credential$redcap_uri,
    token       = credential$token,
    convert_logical_to_integer = TRUE
  )

Add Plumbing Variables

As stated in the vignette’s intro, the structure of the dataset uploaded to the server must be precise. When uploading repeating instruments, there are several important columns:

1. record_id: typically indicates the patient’s id. (This field can be renamed for the project.)
2. redcap_event_name: If the project is longitudinal or has arms, this indicates the event. Otherwise, you don’t need to add this variable.
3. redcap_repeat_instrument: Indicates the instrument/form that is repeating for these columns.
4. redcap_repeat_instance: Typically a sequential positive integer (e.g., 1, 2, 3, …) indicating the order.

The combination of these variables needs to be unique. Please read the Retrieving Longitudinal and Repeating Structures vignette for details of these variables and their meanings.

You need to pass specific variables so that the REDCap server understands the hierarchical structure of the data points.

# repeat-plumbing
ds_daily <-
  ds_daily |>
  dplyr::group_by(id_code) |>
  dplyr::mutate(
    redcap_repeat_instrument  = "daily",
    redcap_repeat_instance    = dplyr::row_number(da_date),
    daily_complete            = REDCapR::constant("form_complete"),
  ) |>
  dplyr::ungroup() |>
  dplyr::select(
    id_code,                        # Or `record_id`, if you didn't rename it
    # redcap_event_name,            # If the project is longitudinal or has arms
    redcap_repeat_instrument,       # The name of the repeating instrument/form
    redcap_repeat_instance,         # The sequence of the repeating instrument
    tidyselect::everything(),       # All columns not explicitly passed to `dplyr::select()`
    daily_complete,                 # Indicates incomplete, unverified, or complete
  )

# Check for potential problems.  (Remember zero rows are good.)
REDCapR::validate_for_write(ds_daily, convert_logical_to_integer = TRUE)

ds_daily

Writing Repeating Instrument Variables

# daily-entire
ds_daily |>
  REDCapR::redcap_write(
    ds_to_write = _,
    redcap_uri  = credential$redcap_uri,
    token       = credential$token,
    convert_logical_to_integer = TRUE
  )

More Complexity

This vignette required only two data.frames, but more complex projects sometimes need more. For example, each repeating instrument should be its own data.frame and writing step. Arms and longitudinal events need to be considered too.

Batching

By default, REDCapR::redcap_write() requests datasets of 100 patients as a time, and stacks the resulting subsets together before returning a data.frame. This can be adjusted to improve performance; the ‘Details’ section of REDCapR::redcap_write() discusses the trade offs.

I usually shoot for ~10 seconds per batch.

Manual vs API

Manual downloading/uploading might make sense if you’re do the operation only once. But when does it ever stop after the first time?

If you have trouble uploading, consider adding a few fake patients & measurements and then download the csv. It might reveal something you didn’t anticipate. But be aware that it will be in the block matrix format (i.e., everything jammed into one rectangle.)

REDCap’s CDIS

The Clinical Data Interoperability Services (CDIS) use FHIR to move data from your institution’s EMR/EHR (eg, Epic, Cerner) to REDCap. Research staff have control over which patient records are selected or eligible. Conceptually it’s similar to writing to REDCap’s with the API, but at much bigger scale. Realistically, it takes months to get through your institution’s human layers. Once established, a project would be populated with EMR data in much less development time
–assuming the desired data models corresponds with FHIR endpoints.