On the other hand, we see that each time a new SDTM-IG is published, additional derived variables have been added - all on request of the FDA. So no wonder files become so large ...
Some time ago, a lot of new --DY (study day) variables were added and made "expected" and it is now also expected that the "the EPOCH variable is included for every clinical subject-level observation".
Now both --DY as EPOCH are derived. --DY (study day) can easily be calculated from the --DTC (date/time of collection or observation) and the RFSTDTC (reference start date/time) in the DM (demographics) domain. Similarly, EPOCH can easily be calculated from the --DTC and the records for that subject in the SE (Subject Elements) domain.
So why does the FDA then still insist that as well --DY as EPOCH is delivered for each observation although it blows up the size of the datasets? Can't the tools of the FDA reviewers calculate the --DY values and the EPOCH "on the fly"?
Some time ago, we developed the "Smart Dataset-XML Viewer", a free and open-source tool for inspecting SDTM, SEND and ADaM files in the new CDISC Dataset-XML format. The Viewer has a good number of great features for exactly doing what the FDA's tools cannot accomplish. It then was a big surprise to us when we were told that most reviewers choose not to use the viewer during the "FDA Dataset-XML pilot". Most of them preferred the (statistical analysis) tools that they always have been working with.
I recently added the newest feature to the "Smart Dataset-XML viewer" which is the display of the (on-the-fly) lookup of the EPOCH and the ELEMENT on each --DTC value. It took me just two evenings to implement that. What the tool is doing is for each --DTC value, it picks up the USUBJID and then compares the date/time with the SESTDTC (Start Date/Time of Element) and SEENDTC (End Date/Time of Element). When the --DTC value falls between the start and end date, the element code (ETCD) is retrieved as well as the value for EPOCH.
Here is a screenshot of a result (here for an observation in the LB domain):
with a detail snapshot:
showing that for the date/time of specimen collection "2014-01-16T13:17" (January 16th 2014 at 1:17 pm) the study day (LBDY) is 15 (15th day of the study for this subject) and the corresponding Element is "PBO" and the corresponding EPOCH is the "blinded treatment" (other epochs are "SCREENING" and "FOLLOW-UP). These values were NOT taken from the dataset, they were calculated "on-the-fly" from the RFSTDTC and the records for this subject in the SE dataset:
This shows that the obligation of the FDA to add --DY and EPOCH to each record is nonsense.
It can easily be done by viewing or analysis tools. The "on-the-fly" calculation even considerably improves data quality.
This nice little feature (again, programmed in two short evenings time) has some further implications. For that, let us have a look at the AE (adverse events). The originally captured dates are AESTDTC (start date/time of adverse event) and AEENDTC (end date/time of adverse event). That's it.
The FDA also requires to submit AESTDY (study day of start of adverse event) and AEENDY (study day of end of adverse event), unnecessarily blowing up the file size again, and even worse, introducing data redundancy. But what about the EPOCH?
Which EPOCH? That for the start of the adverse event? Of the end of it? Or maybe of the date/time of the collection of the AE? It is not possible to add 3 extra columns with EPOCH, or should we add new variables in SUPPAE, for example STAEPOCH, ENDEPOCH? The FDA doesn't tell us.
But of course, for each of them, the "Smart Dataset-XML Viewer" can calculate them "on-the-fly". Here are some screenshots:
First for AESTDTC:
and for AEDTC:
stating that the AE started on study day 3 (element PBO, epoch BLINDED TREATMENT) and ended on study day 199 (element FOLO, epoch FOLLOW-UP) and was captured on day 23 (element PBO, Epoch BLINDED TREATMENT).
So, with this feature, reviewers can easily find out in what study day and in which element and epoch the adverse event started, ended, and was captured in the EDC system, WITHOUT having the --DY and EPOCH variables in the dataset itself.
I will soon upload the updated "Smart Dataset-XML Viewer" programm and source code to the Sourceforge website from which anyone can download them.