Some think CDISC’s CDASH data capture standard is unnecessary.

They say
CDASH and SDTM are in fact very similar.
it’s very similar to SDTM, and the few differences create confusion and • 67% of CDASH v2.0 maps directly to SDTMIG variables, and CDASH v2.0 includes mapping
extra work. CDASH is similar to SDTM, but they solve different problems. • 86% of CDASH maps directly with standard mappings included (e.g., dates)
Used together they positively impact data capture, quality, usability, • 14% are different for a reason
repurposing, and traceability.
SDTM is optimized for tabulation, analysis dataset creation, & data submission.
We explore differences between CDASH and SDTM and CDASH is optimized for data capture, investigator site activities, & data quality.
why both standards are critical. Different requirements, different approaches, but with the same end in mind.

SDTM Why the Difference? CDASH

SDTM assumes that if there is no record then nothing Absence of evidence is not
Show me the data, not lack happened. This works but only if it was checked in data evidence of absence: must
of data capture, which requires a question and record (e.g., Were check that missing data is
there any AEs?) missing

Machine-readable data: SDTM machine-readable formats for variables such as Human-readable:

- ISO 8601 Dates/Times: 1 - Dates/Times: 2 or more
dates are good for data reusability but are not user-
variable, YYYY-MM- variables, DD-MMM-YYYY,
DDThh:mm:ss friendly for data capture. Sites recording data in HH:MM:SS
- Duration: P1M3D unfamiliar formats increases risk of errors - Duration: 1 month, 3 days

Variables must be in order by Domain-driven organization is critical for standard tools, Data structure harmonized
domain; non-standard
but data must make sense to the site. This can mean to with SDTM but variables can
variables are stored in
split domains across CRFs and CRFs across domains, and be arranged to make data
different datasets (e.g., FA,
not split custom and standard variables capture easier.
SUPP--)

RELREC is based on collected data, but data is not captured Links among records are
Collected relationships
explicit (e.g., this AE related to
between data are like that. Entering line numbers in the related datasets is
that CM), or implicit (e.g., AE
represented in RELREC, a simpler, requiring no derivations (e.g., adding AE line # to severity changes going into FA)
separate dataset related con med) in data collection

Normalized structures can store new tests without Findings data may be
Findings data must be in a changing dataset structures, but most EDC systems can’t horizontal, letting each test
normalized or vertical
structure; answers are
do this; also, different tests in a domain may need different have a different code list;
controlled terms (e.g., different answers for different SDTM CT is used for variable
already in variables
questions in a survey) names & CRF prompts

SDTM labels identify tabulation data. CDASH has question Metadata includes capture
Metadata centers on
texts and prompts designed to elicit clear responses on needs, e.g., question
tabulations, e.g., variable
CRFs. CRF instructions convey SDTM and CDASH text/prompt, CRF completion
labels and roles
assumptions in a data capture context instructions

Conclusions
To use SDTM instead of CDASH for data capture, take out derived variables, Whether Regulatory Affairs assembling a submission, FDA
records and datasets; add in data quality indicator variables; put all custom reviewers seeking safety signals, or Big Data miners
and FA variables into parent datasets; reformat variables that are not user- searching for as-yet unknown reasons, future users must be
friendly; reword variable labels to questions; and restructure vertical data confident that the data represents the “truth.”
to horizontal.
Using CDASH facilitates consistent, well-defined data across
This effectively produces CDASH. Except each organization will do it studies. Without that confidence, at best the data will
differently, resulting in reduced data quality and traceability produce vague associations; at worst, it may kill us.

Authors & Acknowledgements

Kit Howard, Director Education, CDISC, [email protected]
Shannon Labout, VP Education, CDISC, [email protected]
With thanks to the CDASH and SDS teams