Pfizer/BioNTech C4591001 Trial - Why do we have two "Vital Signs" files?

Why do we have twice the same dataset for the same submission number, and why are visits spawning or disappearing between files ..?

The “Vital Signs” observed during the trial are documented in a “VS” .xpt file. This file’s CDISC description is “a findings domain (VS) that contains measurements including but not limited to blood pressure, temperature, respiration, body surface area, body mass index, height and weight”.

The structure indicated is “one record per vital sign measurement per time point per visit per subject”1.

The problems are starting when we realize, while in an ongoing effort to clarify the files structure, that we don’t have one, but two Vital Signs files, corresponding to the first Biological License Application (BLA) submission. One has been dumped in the Public Health Medical Professionals for Transparency (PHMPT) drop of December 20222, while the second has followed in the drop of January 20233.

Differences between files

1. The file names are quite close, the difference being “0688438-0698919” in one case, “0623311-0633789” in the second case, prior the “125742_S1_M5” section. We don’t know exactly to what these numbers correspond, as they are documented nowhere, but we suspect it’s related to “submission sections” - I.E. quick references allowing the reviewers to find quickly a source.

2. Extracting data from these files, the only obvious difference is that while they concern the same number of subjects (46 693), they don’t contain the same total rows of data:

• a. 419 178 for the file dropped in December 22
• b. 419 293 for the file dropped in January 23.

3. 50 visits - with “VISITNUM” = “NA”, corresponding to the offset in total rows, are appearing in the January file, but aren’t appearing in the December file. These visits are all concerning reports of Fevers occurring within a week after vaccination (more than one date, test or category can be related to a visit, resulting in a total of events reported impacted higher than 50: 115).

4. The order of the rows has changed between files. We shouldn’t care much, except that it shows poor methods on the exports.

5. The data itself is identical - if we consider that “USUBJID → VISITNUM → VSDTC → VSTEST → VSCAT” ensures the “unique character” of an entry.

Overall, it doesn’t look like much. More precisely, it looks like the sponsors made an error in the December file - and rectified it in a later export.

Nevertheless, as it can impact re-analysis, it’s worth noting that the January version of the Vital Signs file should be preferred over the December one.

How can a visit number be “NA” is another unclear detail. Let us you know if you have any insights on these matters.