OHDSI Home | Forums | Wiki | Github

Drug_exposure_end_date for negative days_supply?

Hi all!

I have a question about populating drug_exposure_end_date field when days_supply is negative.

Following the specification: drug_exposure_end_date = drug_exposure_start_date + days_supply -1.

If there is any standard approach or recommendation about how to calculate drug_exposure_end_date with negative days_supply?

P.s. in our case, negative days_supply indicates voided or adjusted claim.

I’d say that days supply should always be >= 0, tho I understand your use case, just not sure that’s the way to represent it. Maybe this is a question for the THEMIS group?

And, what does voided mean, exactly? That the exposure never happened? If you see a drug_exposure record with a negative days supply, should it be ignored for purposes of identifying people exposed?

Should be >0. What’s a days_supply=0? Empty bottle?

Focus Group 3. Please bring it up with @Asha_Mahesh.

Touche! But maybe injections that don’t have pills but instead have an effective duration. But in that case maybe you derive the days supply from the effective start and end date. Don’t know. But thanks for checking my math. :+1:

If in the original claim was a coding error, then a new record was created with all negative values to void previous one. Sometimes, we have just a source record with negative values without original claim record. I suppose, in such cases a patient had a drug prescription and maybe for some researchers it would be useful to know this fact.

@Chris_Knoll, @Christian_Reich do you think we can ignore such records before THEMIS group’s decision?


If voided means it didn’t happen it didn’t happen. CDM Axiome #2: If nothing happened, we have no record. So, kill them. If these researchers want to study how data get first put in and then voided, because somebody made a mistake, they can go to the raw database.

I will add negative days supply to our THEMIS issues list. I have seen zeros. May be our ETL is kicking negatives out. need to check with Erica.

I know that Christian feels we should drop negative drug records, but I think THEMIS hasn’t 100% decided yet - so stay tuned. Personally I rather keep them in and handle them at analysis time (because as an ETLer I like to keep as much data as possible) but I understand the limitations to that.

If anyone is interested the way Janssen will handle this until a THEMIS decides is DRUG_EXPOSURE_END_DATE = DRUG_EXPOSURE_START_DATE if DAYS_SUPPLY is negative.

So I am on the fence about this one. I see the argument to keep all data but then we need to make sure that all analyses moving forward incorporate some type of check either for a negative DAYS_SUPPLY or negative cost. In that way I can also see the argument to remove them, though we would need to talk with the data vendor to be sure we are removing them correctly.

I know we see these negatives in Truven though I am not sure about our other databases.

I feel strongly that if a drug exposure did not happen then it should not appear in the drug_exposure table. We shouldn’t have to interpret source-specific rules (ie negative means voided in some cases, 99999 means voided in other cases).

To have the best of both worlds, I’d recommend an etl rule that placed voided prescriptions in the observation table.


What’s the use case for having a voided record? What analytical questions are we answer?

I was thinking that if you wanted to accommodate the following cases:
1: refine drug exposure duration by left-joining from the drug exposure table to the observation period table, and if you find an observaton of type ‘discontinuation of drug’ with value_as_concept = the concept from the drug_exposure table and observation was found within the exposure’s start_end date, you could use the observation of discontinuation as the era end date). i might argue, however, that this is ETL logic, back to my point that if it didn’t happen, it shouldn’t be in the drug_exposure table.

2: Be able to exclude patients from a study who were deemed to have a disqualifying event that would cancel their exposure to the drug. In this case, it’s not about trying to refine the actual exposure, but rather disqualify patents from inclusion.

The first case is a sort of ‘partial voiding’ of a drug exposure. through use of the observation table, where there is an exposure record + a cancellation record. But maybe this should be handled in ETLing. The second case is purely an annotation that a drug was refulsed/voided/canceled which maybe be useful for candidate selection in a study (it may serve as proxy for a disqualifying criteria).

So might I.

And a voided=retracted record is a disqualifying event?

Why bother with all this? Let’s get rid of them. Nobody knows what they mean.

I’m imagining the circumstance where someone tries a new drug because it’s cheaper or newer, and immediately has an allergic reaction. Do they get their money back? does some sort of claim-breadcrumb-trail get created to show they sent back a partial filled bottle? I don’t know anything about these sorts of details of claims reporting, but if it is the case, I’d say it would be a useful thing to exclude patients from the study of a drug that showed some type of record that they were somehow not using the drug as intended. I’m not saying ti’s a solid piece of information, but something that maybe useful in sensitivity analysis.

But either way, this information doesn’t belong in drug_exposure, and quite debatable if it even belongs in the patient record at all.

I would only say it is worth a closer look before we just blanket remove negatives. Qianli and I look at it a million years ago and felt we couldn’t make the best decision on how to handle the negatives and left in the CDM for the analyst. I know most of the vendors try to remove these records before they give us data however they still exist. In a situation where we see a +30 and -30 for a Rx if we just remove the -30 we are just as bad as leaving it in. Maybe @clairblacketer and I could review the raw a bit and make a recommendation (unless others want to join in on the fun).

Based on conversation had at the THEMIS F2F, this was the recommendation of what to do with negative days supply for a drug exposure.

When the native data suggests a drug exposure has a days supply less than 0, drop the record as unknown if a person has received the drug or not.

Add this statement under the DRUG_EXPOSURES.CONVENSIONS section of the WIKI https://github.com/OHDSI/CommonDataModel/wiki/DRUG_EXPOSURE#conventions

Internally I was asked to review how often drug exposures with negative days supply is happening for one of our large commercial claims database. 0.04% of the records have a with a days supply less than 0 and the top NDC associated negative days supply is 00000000000. There are 40K distinct NDCs that have negative days supply and 38K of those have <100 exposures.

When reviewing one of our large US EHR data sets, only 14 rows entered the CDM with negative days supply for 6 people.

None of our other CDMs have negative days supply.

Looks like this THEMIS recommendation should be set for release in July. Final words?

When the native data suggests a drug exposure has a days supply less than 0, drop the record as unknown if a person has received the drug or not.

Add this statement under the DRUG_EXPOSURES.CONVENSIONS section of the WIKI

Silence is agreement.