OHDSI Home | Forums | Wiki | Github

ATC DDD to impute drug_exposure_end_date

cdm

(Alexandra Orlova) #1

If source data provides the drug and total quantity only (no daily dosage, no frequency, no days supplied, no end date), then we need to impute drug_exposure_end_date.

We can use 30 days as a default value, but it’s not precise enough.
Together with @zhuk and @Alexdavv we are trying to apply the ATC DDD approach (at least for oral solid forms) and found some pros and cons. Please find a couple of examples attached:

  • ‘proportion, %’ column contains % of records for a combination of drug and total quantity in the source data
  • ‘days_supply’ column is calculated value: drug strength * quantity / ddd
    ATC DDD.xlsx (17.2 KB)

Our questions are the following:

  1. Has anybody tried to calculate the duration of taking the drug using ATC DDD?
  2. How did you deal with cases when DDD is not working (how did you automate that in your ETL?)?
  3. Did you build any heuristic based on quantity typical values (28/56/84/100/60/90)?
  4. Did you take into account the total quantity and ‘DDD / drug strength’ ratio? For example, if quantity > 30, then use DDD, if ‘DDD / drug strength’ > 2, then use DDD, otherwise use total quantity as duration days.

(Oleg Zhuk) #2

Hello! We know, that Symposium took all your attentition (and breath away) this week, but please take a look at our proposal and share your valuable opinion!
Thank you very much
Tagging @Christian_Reich @Vojtech_Huser @MPhilofsky @ericaVoss @MaximMoinat @DTorok @abedtash_hamed @aostropolets


(Anna Ostropolets) #3

I doubt such estimates of drug exposure are reliable. Based on my experience, real-world ATC code assignment mainly accounts for ingredient, sometimes for dose form and almost never for DDD. I’d suggest assessing plausibility of your calculations prior to implementing them (e.g. if inpatient administrations align with length of stay, if mode is 30/60/90 days etc.). As in point #3 I’d expect 7/30/60/90 days to be common values.

Also, you mentioned having some insights from such an estimation. Would be curious to hear more.


(Hamed Abedtash) #4

Hi @Alexandra_Orlova, your calculations based on DDD is a typical work in pharmacoepi studies when dose value is missing; however, I wouldn’t recommend the approach to “reverse engineer” days of supply because DDD is not a true prescribed/administered dose but is “the assumed average maintenance dose per day for a drug used for its main indication in adults”, so it doesn’t present how much of the drug was prescribed by the physician. Also, per the CDM’s convention, “days supply” is the verbatim value as it was prescribed per the claim record or physician order.

Thus, you are welcome to use this approach when analyzing data (if applicable) but not at the time of data transformation as it will introduce significant bias in your data.


(Alexandra Orlova) #5

Hi Hamed,
Thank you for your answer!

One clarification: we are not going to impute days_supply field. We need to impute drug_exposure_end_date because it’s a required field (for OMOP CDM v5.2 and later).


(Christian Reich) #6

That is correct.

That’s what I would do. Check it against other plausible guestimates, like Anna’s length of stay or timing of following prescription. It doesn’t have to be perfect, it has to be better than the standard fixed guess of 30 days.


(Alexandra Orlova) #7

@aostropolets, @abedtash_hamed, @Christian_Reich, thank you all for suggestions!
We checked source data and found out that ATC DDD method for oral solid drugs gives 28, 56, 30, 7, and some other values as most common.
Also, we checked that against the following prescription method and got similar results (28, 29, 27, 56, 7, 30).

So we will try to apply ATC DDD method for oral solid drugs.


(Christian Reich) #8

Can you publish on Github when ready? We need scripts like that.


t