I cannot find in any of the discussions the current recommendation for drug_exposure_end_date when you have no information at all other than the start date. I assume, because it is required, you set end date to the start date, and you leave the days_supply and verbatim_end_date null. Is that correct? Thanks.
On hearing no replies, I will assert that you do indeed set the drug_exposure_end_date to the start date when you have no information whatsoever about the length of the drug administration.
George – are you not making an assertion that you do not know to be true by setting the end_date equal to the start_date? Isn’t the alternative to change the data model to not require an end_date so that you can leave it null when you have no information regarding the real end_date. What analyses/tools/religions would this change to the data model break?
We (PEDSnet) do not require drug_exposure_end_date(?:time), though this is mostly by virtue of not having updated our DDLs in the past month or two; it was not required in OMOP 5.1.
On the merits, I tend to agree with Michael; if it’s not present in (or deriveable from) the data, it’d be nice to preserve that. It seems reasonable to assert a same-day end for drug_era and dose_era if all else fails, since the concept of interval is inherent in those tables, but I can’t think of a reason other than analytic simplification to do so in drug_exposure. What am I missing?
If I understand correctly, you can preserve the missing drug_exposure_end_date by leaving verbatim_end_date empty. Then it is up to the ETL document to specify how the drug_exposure_end_date is inferred.
See discussions on the forum and the Github issue.
Most often drug_exposure_end_date is defined as drug_exposure_start_date + days_supply, so we do not use the reported (if any) end date in our analyses. There have been some cases that reported end dates did not match drug’s supplied days, so we decided to only include days_supply in the CDM and put drug_exposure_end_date=NULL.
Friends:
Hang on here. This is a settled case. Sorry, @hripcsa, for not answering earlier. Plus, you didn’t actually ask about leaving it empty, but what should go in there. So, here is the deal:
- drug_exposure_end_date is now mandatory. Sorry, @abedtash_hamed.
- You should fill it with whatever value you decide is the best. That is something you can only decide knowing the source data. If your source data don’t contain a reliable end date, come up with the heuristic.
- If you have to guess, I would suggest 30 days for prescriptions, and 1 day (same day) for provider-administered drugs.
- The verbatim_end_date is for those folks, who want to know when you guessed, so they have a choice to ignore your guess if it is empty.
1 Like
Would add that, at times, you know if it is mail order and then you might
assume 90 days.
OK, thanks. I remember the discussion, but had missed the change in constraint.
This is the hard case. OT1H, I can certainly understand that the people closest to the source have the best view of the data. OTOH, with a wide-open spec, standard analytics effectively can’t use this field at all, because there’s no common semantics.
What, not drug_end_date_source_value? 
Two points:
- Well, we need to nail that. If you want to do analytics on drug exposure, you shouldn’t start having to do all sorts of guesswork on whether you trust the data or not. Maybe it’s true, and not all analytics are reasonable in a world like that. But remember: OMOP CDM is for aggregate research, not individual patient management. If there are one or two wrong spots - who cares.
- We want to fix that exact problem in THEMIS: Which use cases, what convention. Please bring your use cases.