Paul W. Egeler, M.S., GStat
Spectrum Health Office of Research Administration
13 July 2017
So the new buzz in the REDCap world seems to be Repeating Instruments and Events. Certainly there is potential for a lot of utility in this feature and I was excited to try it out. I know I will be using this feature a lot in the future.
Unfortunately, I was not very happy with the way the data was exported either via CSV or API call. When you conceptualize the data model for a Repeating Instrument, you probably think of a multi-table model. You might expect that the non-repeating instruments may constitute one table that would be related to Repeating Instruments tables via a one-to-many relationship. In reality, the data is outputted as one table with all possible fields; this has the effect of nesting the output table in a way that is not useful in most analysis software.
The normalized data can be retrieved by downloading repeating instruments individually then doing a little data munging or by writing a few custom parameters in a series of API calls (then doing more data munging), but this is a lot of extra steps that can make reproducible research more difficult.
REDCapRITS is a programmatic solution to handle the problem in both SAS and R.
For example, consider this mocked-up data exported from a REDCap project with repeating instruments.
The data contains information on a subset of cars in R's built-in mtcars
dataset [1].
Within the table there is also a repeating instrument, sales, which has sales transaction
data for some of those cars.
car_id | redcap_repeat_instrument | redcap_repeat_instance | make | model | mpg | cyl | motor_trend_cars_complete | price | color | customer | sale_complete |
---|---|---|---|---|---|---|---|---|---|---|---|
1 | AMC | Javelin | 15.2 | 8 | 1 | ||||||
1 | sale | 1 | 12000.50 | 1 | Bob | 0 | |||||
1 | sale | 2 | 13750.77 | 3 | Sue | 2 | |||||
1 | sale | 3 | 15004.57 | 2 | Kim | 0 | |||||
2 | Cadillac | Fleetwood | 10.4 | 8 | 0 | ||||||
3 | Camaro | Z28 | 13.3 | 8 | 0 | ||||||
3 | sale | 1 | 7800.00 | 2 | Janice | 2 | |||||
3 | sale | 2 | 8000.00 | 3 | Tim | 0 | |||||
4 | Chrysler | Imperial | 14.7 | 8 | 0 | ||||||
4 | sale | 1 | 7500.00 | 1 | Jim | 2 | |||||
5 | Datsun | 710 | 22.8 | 4 | 0 | ||||||
6 | Dodge | Challenger | 15.5 | 8 | 0 | ||||||
7 | Duster | 360 | 14.3 | 8 | 0 | ||||||
7 | sale | 1 | 8756.40 | 4 | Sarah | 1 | |||||
7 | sale | 2 | 6800.88 | 2 | Pablo | 0 | |||||
7 | sale | 3 | 8888.88 | 1 | Erica | 0 | |||||
7 | sale | 4 | 970.00 | 4 | Juan | 0 | |||||
8 | Ferrari | Dino | 19.7 | 6 | 0 | ||||||
9 | Mazda | RX4 Wag | 21 | 6 | 0 | ||||||
10 | Merc | 230 | 22.8 | 4 | 0 | ||||||
10 | sale | 1 | 7800.98 | 2 | Ted | 0 | |||||
10 | sale | 2 | 7954.00 | 1 | Quentin | 0 | |||||
10 | sale | 3 | 6800.55 | 3 | Sharon | 2 |
You can see that the data from the non-repeating form (primary table) is interlaced with the data in the repeating form, creating a checkerboard pattern. In order to do analysis, the data must be normalized and then the tables rejoined. Normalization would result in two tables: 1) a primary table and 2) a sale table. The normalized tables would look like this:
Primary table
car_id | make | model | mpg | cyl | motor_trend_cars_complete |
---|---|---|---|---|---|
1 | AMC | Javelin | 15.2 | 8 | 1 |
2 | Cadillac | Fleetwood | 10.4 | 8 | 0 |
3 | Camaro | Z28 | 13.3 | 8 | 0 |
4 | Chrysler | Imperial | 14.7 | 8 | 0 |
5 | Datsun | 710 | 22.8 | 4 | 0 |
6 | Dodge | Challenger | 15.5 | 8 | 0 |
7 | Duster | 360 | 14.3 | 8 | 0 |
8 | Ferrari | Dino | 19.7 | 6 | 0 |
9 | Mazda | RX4 Wag | 21 | 6 | 0 |
10 | Merc | 230 | 22.8 | 4 | 0 |
Sale table
car_id | redcap_repeat_instrument | redcap_repeat_instance | price | color | customer | sale_complete |
---|---|---|---|---|---|---|
1 | sale | 1 | 12000.50 | 1 | Bob | 0 |
1 | sale | 2 | 13750.77 | 3 | Sue | 2 |
1 | sale | 3 | 15004.57 | 2 | Kim | 0 |
3 | sale | 1 | 7800.00 | 2 | Janice | 2 |
3 | sale | 2 | 8000.00 | 3 | Tim | 0 |
4 | sale | 1 | 7500.00 | 1 | Jim | 2 |
7 | sale | 1 | 8756.40 | 4 | Sarah | 1 |
7 | sale | 2 | 6800.88 | 2 | Pablo | 0 |
7 | sale | 3 | 8888.88 | 1 | Erica | 0 |
7 | sale | 4 | 970.00 | 4 | Juan | 0 |
10 | sale | 1 | 7800.98 | 2 | Ted | 0 |
10 | sale | 2 | 7954.00 | 1 | Quentin | 0 |
10 | sale | 3 | 6800.55 | 3 | Sharon | 2 |
Suppose you would like to do some analysis such as sale price by make of car or find
the most popular color for each model. To do so, you can join the tables together with
relational algebra. After inner joining the primary table to the sale table on car_id
and selecting only the fields you are interested in,
your resulting analytic dataset might look something like this:
car_id | make | model | price | color | customer |
---|---|---|---|---|---|
1 | AMC | Javelin | 12000.50 | 1 | Bob |
1 | AMC | Javelin | 13750.77 | 3 | Sue |
1 | AMC | Javelin | 15004.57 | 2 | Kim |
3 | Camaro | Z28 | 7800.00 | 2 | Janice |
3 | Camaro | Z28 | 8000.00 | 3 | Tim |
4 | Chrysler | Imperial | 7500.00 | 1 | Jim |
7 | Duster | 360 | 8756.40 | 4 | Sarah |
7 | Duster | 360 | 6800.88 | 2 | Pablo |
7 | Duster | 360 | 8888.88 | 1 | Erica |
7 | Duster | 360 | 970.00 | 4 | Juan |
10 | Merc | 230 | 7800.98 | 2 | Ted |
10 | Merc | 230 | 7954.00 | 1 | Quentin |
10 | Merc | 230 | 6800.55 | 3 | Sharon |
Such a join can be accomplished numerous ways. Just to name a few:
- SAS
- R
Currently, the R and SAS code is well-tested with mocked-up data.
- R
- SAS
I have made some effort to replicate the messiness of real-world data and have tried to include as many special cases and data types as possible. However, this code may not account for all contingencies or changes in the native REDCap export format. If you find a bug, please feel free to open an issue or pull request.
Currently, we have given some consideration to expand the capabilities into the following languages.
- Python
- VBA
If you have some talents in these or other languages, please feel free to open a pull request! We welcome your contributions!
First you must install the package. To do so, execute the following in your R console:
if (!require(devtools)) install.packages("devtools")
devtools::install_github("SpectrumHealthResearch/REDCapRITS/R")
After the package is installed, follow these instructions:
- Download the record dataset and metadata (data dictionary). This can
be accomplished by several methods:
- Using the API. Check with your REDCap administrator for details.
- Exporting the data from the web interface by selecting CSV / Microsoft Excel (raw data).
- Exporting the data from the web interface by selecting R Statistical Software. If you use this method, you may run the R script supplied by REDCap prior to splitting the data.
- Do NOT export from the web interface with the CSV / Microsoft Excel (labels) option. This will not work with REDCapRITS.
- Call the function, pointing it to your record dataset and metadata
data.frame
s or JSON character vectors. You may need to load the package vialibrary()
orrequire()
.
Here is an example usage in conjuction with an API call to your REDCap instance:
library(RCurl)
# Get the records
records <- postForm(
uri = api_url, # Supply your site-specific URI
token = api_token, # Supply your own API token
content = 'record',
format = 'json',
returnFormat = 'json'
)
# Get the metadata
metadata <- postForm(
uri = api_url, # Supply your site-specific URI
token = api_token, # Supply your own API token
content = 'metadata',
format = 'json'
)
# Convert exported JSON strings into a list of data.frames
REDCapRITS::REDCap_split(records, metadata)
And here is an example of usage when downloading a REDCap export of the raw data (not labelled!) manually from your REDCap web interface:
# Get the records
records <- read.csv("/path/to/data/ExampleProject_DATA_2018-06-03_1700.csv")
# Get the metadata
metadata <- read.csv("/path/to/data/ExampleProject_DataDictionary_2018-06-03.csv")
# Split the tables
REDCapRITS::REDCap_split(records, metadata)
REDCapRITS also works with the data export script (a.k.a., syntax file) supplied by REDCap. Here is an example of its usage:
# Run the data export script supplied by REDCap.
# This will create a data.frame of your records called 'data'
source("/path/to/data/ExampleProject_R_2018-06-03_1700.r", chdir = TRUE)
# Get the metadata
metadata <- read.csv("/path/to/metadata/ExampleProject_DataDictionary_2018-06-03.csv")
# Split the tables
REDCapRITS::REDCap_split(data, metadata)
- Download the data, SAS code to load the data, and the data dictionary from REDCap.
- Run the SAS code provided by REDCap to import the data.
- Run the RECapRITS macro definitions in the source editor or using
%include
. - Run the macro call
%REDCAP_READ_DATA_DICT()
to load the data dictionary into your SAS session, pointing to the file location of your REDCap data dictionary. - Run the macro call
%REDCAP_SPLIT()
. You will have an output dataset for your main table as well as for each repeating instrument.
Please follow the instructions from REDCap on importing the data into SAS. REDCap provides the data in a csv format as well as bat and sas files. The instructions are available when exporting the data from the REDCap web interface. If you do not use the pathway mapper (bat file) provided, you will need to go into the sas file provided by REDCap and alter the file path in the infile
statment (Line 2).
* Run the program to import the data file into a SAS dataset;
%INCLUDE "c:\path\to\data\ExampleProject_SAS_2018-06-04_0950.sas";
* Run the MACRO definitions from this repo;
%INCLUDE "c:\path\to\macro\REDCap_split.sas";
* Read in the data dictionary;
%REDCAP_READ_DATA_DICT(c:\path\to\data\ExampleProject_DataDictionary_2018-06-04.csv);
* Split the tables;
%REDCAP_SPLIT();
Suggestions and contributions are more than welcome! Please feel free to create an issue or pull request.
This code was written for REDCap electronic data capture tools [2]. Code for this project was tested on the REDCap instance hosted at Spectrum Health, Grand Rapids, MI. REDCap (Research Electronic Data Capture) is a secure, web-based application designed to support data capture for research studies, providing 1) an intuitive interface for validated data entry; 2) audit trails for tracking data manipulation and export procedures; 3) automated export procedures for seamless data downloads to common statistical packages; and 4) procedures for importing data from external sources.
[1] Henderson and Velleman (1981), Building multiple regression models interactively. Biometrics, 37, 391--411. Modified with fake data for the purpose of illustration
[2] Paul A. Harris, Robert Taylor, Robert Thielke, Jonathon Payne, Nathaniel Gonzalez, Jose G. Conde, Research electronic data capture (REDCap) – A metadata-driven methodology and workflow process for providing translational research informatics support, J Biomed Inform. 2009 Apr;42(2):377-81.