FDA Points Out Few Whoops Moments Regarding CDISC SDTM Submission Data
It looks like the FDA wants CDISC SDTM data but is getting frustrated with sponsors not adhering to the guidelines and not checking their data. Their recent document 'CDER Common Data Standards Issues Document Version 1.0/May 2011' is basically a hand slapping finger-wagging confrontation indicating what not to do when submitting CDISC SDTM standard datasets.
Along with recommendations such as simply reading the manual, or in this case, reviewing and adhering to the standard where possible, there are some insights to the problems the FDA has seen with CDISC submission data. Some of these are subtle, but some are just plain embarrassing!
- Reviewers have encountered problems in submissions where legacy data was converted to CDISC SDTM inconsistently across studies.
- When submitting a begin date and end date for an event, the begin date should be prior to the end date. (!)
- FDA has received data that was not collected in SDTM and subsequently converted to SDTM. This is all well and fine, but the analysis datasets need to be traceable to the newly created CDISC SDTM datasets and not based on the original data collected.
- Dataset label names are important and should follow the standard; and, by the way, the ADaM and CDISC SDTM standard for the AE dataset should have different dataset labels!
- If the CDISC standards don’t work for your study, it is OK to create sponsor specific terminology and data definitions, but make sure you talk it over with your reviewer so they know what to expect.
- Spelling counts – when coding data, make sure the same terms or system organ class names terms are spelled consistently.
- Required variables are required and cannot be null. Don’t forget them or leave them blank.
- When coding data, the same dictionary should be used for all submitted domains. Although these tend to be coded at the time of the study lock, they all need to be recoded to provide consistency for the CDSIC SDTM submission.
- The FDA standard issue computers can’t handle big datasets, so split up big ones like labs (really!).
- Variables that are permissible need to be submitted if they are collected, or if they can be derived as with certain EPOCH designations or flags. Just because they are not required or expected doesn’t mean they can be omitted.
- USUBJID is a unique identifier for the entire CDISC SDTM submission. It looks like people don’t get that “unique” means unique with no trailing spaces or leading zeros. This section has the word exact and common error underlined so it looks like it is a problem.
Please let me know your thoughts on these CDISC SDTM findings. Is your company checking for these apparently common issues?