swan ellert, ms - checkmate poster
DESCRIPTION
Improving Investigator Adherence to Best Practice Guidelines in Electronic Data Capture Using an Automated Tool (CheckMate)TRANSCRIPT
Improving Investigator Adherence to Best Practice Guidelines in
Electronic Data Capture Using an Automated Tool (CheckMate)Swan Ellert MS, Linda Carlin PhD, Umit Kaya MS, Jessica Bondy MHA, Michael G. Kahn MD, PhD
Colorado Clinical & Translational Sciences Institute, University of Colorado Anschutz Medical Campus
Electronic data capture (EDC) best practice guidelines for
database design are intended to improve quality of
collected data by reducing risk of data entry errors. EDC
applications provide features to support many of these best
practices, such as minimizing use of free text entry,
validating text field formats, and setting expected
minimum/maximum values.
Background
Problem Description
Colorado Clinical & Translational Sciences Institute Aurora, CO cctsi.ucdenver.eduSupported by NIH/NCRR Colorado CTSI Grant Number 3UL1RR025780-02S3. Its contents are the authors’ sole responsibility and do not necessarily represent official NIH views.
CheckMate Features
CheckMate is a Microsoft Excel macro developed to identify
areas in a REDCap database data dictionary that do not
conform with applicable best practice guidelines. The
specific guidelines the tool currently identifies are shown in
the central panel. Most CheckMate recommendations are
made by searching for common words (e.g.. date, phone,
email) and matching these words to a guideline.
The percent of free form text did not
change across all studies.
Text validators were already applied
99% of the time across all 9 studies
that did not implement CheckMate
recommendations. For the 5 studies
that did implement CheckMate
recommendations, the text validator
percent went from 73% to 76% across
all studies.
Min/Max values on appropriate text
fields remained low (under 40%) across
all studies
PHI identification made the biggest
gain. However, PHI identification
remained under 50% across all
studies. Further analysis needs to be
done to determine whether word
searches correctly identify PHI.
Example CheckMate Data Dictionary Scan
and User Summary Output
The Colorado Clinical and Translational Science Institute
currently encourages researchers to use REDCap as their
data collection tool. This policy allows investigators, who
understand their study data needs best, to create forms
and determine what fields need to be assigned within the
forms. However, because researchers may not be familiar
with data management best practices, these practices can
often be ignored or misunderstood by investigators. We
created a tool to quickly review database designs for
adherence to best practices, and to provide
recommendations to investigators for improving their EDC
forms prior to data collection.
Create instructions to investigators about how to prepare a data
dictionary for upload to CheckMate (columns to remove, save format,
etc.)
Move “Best Practice Suggestion” column to far right beside “Best
Practice Followed”
Reorganize “Specific Recommendations” to mimic workflow, e.g.,
put text validation suggestions (4, 5) before min/max suggestions (2)
Suggest “Choices” option as well as text validator
Target “Text Validation Min” and “Text Validation Max” separately
Tighten words searches, to lower misidentified best practices (e.g.
search on “Name” identifies “Name” and “Name of Data Entry
Person”)
Check for “Choices” where multiple codes have the same meaning
(i.e., both 99 and 98 mean “Unknown” in the same study data
dictionary)
Future Directions
Evaluate whether other changes, not specifically suggested by
CheckMate, are made to the data dictionary as a result of reviewing
the CheckMate scan
Observe whether reviewing and/or implementing CheckMate
results limits the number of post-deployment changes
Review post-deployment changes to see whether they relate to a
problem identified but not implemented by the investigator in the
CheckMate scan
Validate specifically labeled fields against standard data
definitions. For example, verify that a field label of “Ethnicity” maps
to Cancer Biomedical Informatics Grid (caBig) permissible values.
Present CheckMate as opportunity to REDCap Consortium for use
by other consortium members or for future REDCap enhancements
CheckMate Improvements
Initial Findings
References:
1. CDISC CDASH Core and Domain Teams. Clinical Data Acquisition Standards Harmonization (CDASH); October 2008.
2. Harris PA, Taylor R, Thielke R, Payne J, Gonzalez N, Conde JG. Research electronic data capture (REDCap)--A metadata-driven
methodology and workflow process for providing translational research informatics support. Journal of Biomedical Informatics.
2009;42:377–381.
Best Practice
Suggestion Variable or Field Name Field Label
Choices OR
Calculations Text Validation Type Text Validation Min Text Validation Max Identifier? Best Practice Followed
studid
Study Identification
Number
1,2 mrn
Medical Record
Number number 4
1 last_name Last Name
1,2 dob Date of Birth date 4
sex Gender 0, Female | 1, Male
4,5 weight
Weight at time of
surgery
2 surgdate Date of surgery date 4
General Recommendations:
16% of the fields in this study are free form text. If responses can be categorized, consider using a dropdown field type to reduce risk of data entry error and make the data easier to analyze.
3 out of 8 forms in this study have more than 30 fields. Consider creating shorter forms for better data entry.
Specific Recommendations (not all recommendations listed below will apply to this study):
1. Possible PHI. Consider placing a 'Y' in column M [Identifier?]
2. For text fields validated as date, integer, or number, consider entering minimum and maximum expected values in columns K [Text Validation Min] and L [Text Validation Max] to decrease risk of data entry error.
3. Statistical results will vary when using dynamic expressions (e.g. 'today') in calculations. Consider using a fixed date, e.g. 'date of enrollment'.
4. When using text fields, consider validating the field to expect a specific data type by entering it in column J [Text Validation Type]. Available options are: date, time, integer, number, zipcode, phone and email.
5. When entering common lab/physical exam values, consider validating the text field to expect number or integer values in column J [Text Validation Type]
6. Consider making street, city, state and zip separate fields in the database
Fields in green meet best practice guidelines
CheckMate Scan Workflow
PI drafts DDPre-Deployment CheckMate Scan
CheckMate Scan recommendations
sent to PI
PI decides which recommendations
to acceptStudy goes live
Post-Deployment CheckMate Scan
0 10 20 30 40 50 60 70 80 90 100
% possible PHI fields
% Validated Free Form Text with Min/Max
% Free Form Text (Validated)
% Free Form Text
Post-Deployment Scan (5 studies with CheckMate scans implemented)
Pre-Deployment Scan (5 studies with CheckMate scans implemented)
Pre & Post-Deployment (9 studies with no CheckMate recommendations implemented)