STRESN value not matching STRESC value when the values are the same in XPT

m Milan

on October 12, 2024

Hi Folks,

When I use the validator, I get a message that the number in my STRESN variable is not the same as STRESC variable. I have checked the .XPT and the values are exactly the same. Here is what the validator spits out

PPSTRESC=165.163206499472, PPSTRESN=165.1632065

I get this message only for 3 such observations rest of the observations match. I am using SDTM-IG 3.4 (FDA) configuration.

Thanks for your help.

Regards,

Milan

Forums: SDTM

j Jozef

on October 13, 2024

STRESN value not matching STRESC

Exactly why we should get rid of XPT as soon as possible. XPT is a completed outdated format, even discouraged by the US government (see https://www.loc.gov/preservation/digital/formats/fdd/fdd000464.shtml).
In XPT, --STRESN values are stored in binary "IBM double format", which was used by IBM mainframes in the sixties and seventies. Modern computers do not use this format anymore.
This format does not make a distinction between integers and floating point numbers.
--STRESC values however are stored in "US-ASCII" format (i.e. "text"). For details about how variable values are stored in XPT, see https://support.sas.com/content/dam/SAS/support/en/technical-papers/record-layout-of-a-sas-version-5-or-6-data-set-in-sas-transport-xport-format.pdf. This is the famous "TS140" document (as it was called in the past).

As the IBM notation is very different of what modern computers use (IEEE standard, see e.g. https://en.wikipedia.org/wiki/IEEE_754), it is essentially impossible to exactly compare --STRESN and --STRESC, and some "precision matching" needs to be performed by rounding. In your case, it looks as the validation engine doesn't seem to do this in an acceptable way.'
When you say "I have checked the .XPT ", you mean the presentation in the viewer, which is not what is in the real file, as the --STRESN value there is binary, which humans cannot read.

CDISC has published a modern, JSON-based format for submissions to the FDA, named Dataset-JSON 1.1 (see https://www.cdisc.org/dataset-json), with full support of the FDA. At the moment of writing, it is in public review.
My personal hope is that Dataset-JSON will be published in the FDA "standards catalog" (early?) next year, and as of that day, it can be used for FDA submissions. Many software tools (such as SAS, R, SDTM-ETL) already support the new format for generation of SDTM, SEND and ADaM datasets.

For the moment, you can document these validator messages as false positives in the Reviewers Guide.