| 55min | Review of the 11 proposed solutions | Éanna Kiely | - Erin Notes:
- Feedback From Trish Gleason (BMS): Hi Erin, We’ve met with our statisticians and reviewed all of the options. BMS is currently already providing Option 2 to the FDA. The majority of all other options were not well received. Option 7 was the only one we documented as being somewhat acceptable. Note: This would impact more than just the LB domain, correct? Lab tests will span multiple domains: IS, MB, where there is potential for US and SI units to differ. Has that been considered? Concerns with 3 and 5 would make the lab structure too big.
- Team agrees options 4 and 8 should be removed from consideration (SCAT, UCUM).
- Original result would still go into the ORRES value.
- Using UCUM does not solve the problem. UCUM has both SI and conventional units but does not specify which to use. Can UCUM be used for storing conversions in the define xml.
- Creating one comprehensive dataset would be of more value (would require changes to the model) than submitting multiple datasets.
- Change the LB domain such that you only capture the ORRES, and then put conversions/derivations for the units somewhere else. Do you put this in another dataset or is this derived data and should be in ADaM?
- Make better use of the Lab transmission model: It has blocks for SI units, conventional units, and reported units. Doesn’t solve how you represent it in the SDTM though.
- We need to be careful because LOINC codes change based on whether the units are SI or conventional. Mass-based vs substance-based LOINC codes are differentiated with different LOINC codes. From FDA recs doc: LOINC code specified in LB.LOINC applies to LB.ORRES (original result) rather than to LB.STRES(C/N) (standardized result). Such an approach is consistent with the recommendation that LB.LOINC should be taken from original laboratory result (i.e., the data transfer itself), and not be derived/converted into another form.
- Team discusses option 9. There seems to be resistance about using define.xml from statisticians in general as they want everything in the dataset. Craig asks whether a conversion factor would be appropriate for define.xml. Alan says that this could become very large (with local labs). This might require a new version of define.xml. The team thinks the define.xml file may be a good place to store conversion factors. Take 9 off the table but define.xml can
- Concerns with reporting data that wasn’t provided by the lab. ETL tools can do the conversion but you may not get the same answer as the lab and the lab can say they didn’t provide it. The TCG is very clear that SDTM and SEND datasets should not include imputed data. The lab team in general agrees that conversions seem to be imputed data. However local lab data gets converted all the time and that data still is in the dataset. FDA wants STRESC to all be one unit so they seem to be contradicting themselves.
- CDISC may need to create a controlled space for conversion factors one day.
- Eanna Notes:
|