OHDSI Home | Forums | Wiki | Github

Days_Supply (yet again)

Working with EHR prescriptions. Issue arose about what is the correct value for days_supply when prescription has refills (this posting does not apply to fulfillment records). In Should days_supply and refills be included to calculate observation_period_end date?, @Klaus states: BTW: days_supply refers to a single prescription, it must not contain days from refills! Otherwise we would calculate a wrong daily dosage! If you want to record refills, use the drug_exposure_end_date. I would like to argue this is wrong for prescriptions.

Consider three example prescriptions, all of which we have seen in real records:

  1. Drug A once a day, 6 month supply, no refills
  2. Drug A once a day, 3 month supply, refill x 1
  3. Drug A once a day, 1 month supply, refill x 5 (edited to correct error in original post)

The controversy is: For these drug_exposure records, is “Days_Supply” 180, 90, and 30 respectively? Or is Days_Supply = 180 for all three? My reading of @Klaus’s posting above is that he advocates 180, 90, 30. I am advocating 180 for all since it seems clear to me that the intent of all three prescriptions is to provide Drug A for 180 days. They just “get there” in different ways which will be reflected in the dispensing records (if you are lucky enough to have them which we don’t). In all three cases, you will not expect to see a new prescription record for 180 days (assuming full compliance). If you use 180/90/30 days_supply, you could falsely generate disjoint drug_eras from the shorter prescriptions, which is (unfortunately) what we are generating.

Let the vehement conversations begin…

1 Like

I am going to have to vehemently agree with you (except that scenario 3 should have 5 refills). If all you get is the prescription as written, then I would code them all as 180 days. We don’t know any more about the first prescription than we do about the refills, so it makes sense to treat them the same. You could code them as discrete records (i.e, the first would be one record, the second would be 2 records, and the third would be 6 records). But that doesn’t seem helpful unless you want to know about the days supplied for the prescriptions/refills (e.g., 30 days vs. 180 days).

1 Like

Of course @Mark_Danese is right about my mistake regarding Scenario 3 and 5 refills. I will edit my original post to erase my mistake!

Friends:

This depends on whether you see prescription written, or dispensed off of a written prescription. In the former, you have to specify the whole pile of pills, in the second case you do that for each refill. @mgkahn: you probably have the former case.

Does that answer the question?

The controversy is: For these drug_exposure records, is “Days_Supply” 180, 90, and 30 respectively?

Your 3 scenarios more elaborated (in prescription realm)

OPTION A

  1. start=Jan 1 2000, days_supply=180, refills=0, end=S+6M, quantity=?
  2. start=Jan 1 2000, days_supply=180, refills=1, end=S+3M
  3. start=Jan 1 2000, days_supply=180, refills=5, end=S+1M

OPTION B (per Klaus)

  1. start=Jan 1 2000, days_supply=180, refills=0, end=S+6M, quantity=?
  2. start=Jan 1 2000, days_supply=90, refills=1, end=S+6M
  3. start=Jan 1 2000, days_supply=30, refills=5, end=S+6M

The key question is about quantity = ???
Am I communicating the total number of pills for next 6 months (accross all refils) or quantity of pills for a given “refill session”. The sig is clear - ‘take one pill a day’.
So in dispensing realm, it is intuitive what quantity is. I would say that in prescription realm - the doctor thinks about start and end date, via refills or one big shot and sig. The physician may not really “calculate quantity”. Sig is most clinically inutitive for him/her.

For quantity, the specs say 'The quantity of drug as recorded in the original prescription or dispensing record.'

Quantity field should indicate the total number of pills (or boxes, milliliters, milligrams, international units - depending on the class of drug_concept_id) applied during current DRUG_EXPOSURE event, not considering refills field in CDM.

Refills field indicates just the current number of total refills number. That is described in p. 6.8.1 Conventions of CDM spec:
The content of the refills field determines the current number of refills, not the number of remaining refills. For example, for a drug prescription with 2 refills, the content of this field for the 3 Drug Exposure events are null, 1 and 2.

@Christian_Reich correct me if I’m wrong, your second case will look like:

Drug A once a day, 6 month supply, no refills
1 row in drug_exposure
start=Jan 1 2000, days_supply=180, refill=null, end=S+ 180 days, quantity=180 (number of pills per day * days_supply)

Drug A once a day, 3 month supply, refill x 1
2 rows in drug_exposure
start=Jan 1 2000, days_supply=90, refill=null, end=S+90 days, quantity=90
start=April 1 2000, days_supply=90, refill=1, end=S+90 days, quantity=90

Drug A once a day, 1 month supply, refill x 5 (edited to correct error in original post)
5 rows in drug_exposure
start=Jan 1 2000, days_supply=30, refill=null, end=S+30 days
start=Feb 1 2000, days_supply=30, refill=1, end=S+30 days
start=March 1 2000, days_supply=30, refill=2, end=S+30 days
start=April 1 2000, days_supply=30, refill=3, end=S+30 days
start=May 1 2000, days_supply=30, refills=4, end=S+30 days

all in one row?

Option A correctly describes my position for Days_Supply but not for end_date where I would put S+6M for all three scenarios. The logic is all three prescriptions are intended to provide medications for the entire 6 month period. It is semantically irrelevant that they get their via three different dispensing methods – the Rx is “valid” for enough medications to last 6 months. Thus, with the information available in the Rx, end date for all three scenarios = S+6M.

I think we need to rethink the semantics for refills as currently written in the CDM data dictionary. While the description makes sense for fulfillment records, it doesn’t make sense (to me) for prescriptions. I have thoughts about this but want to noodle more before offering ideas in public.

@aostropolets – I also considered your approach but ultimately convinced myself that a single record for the one prescription rather than N records for each N refill in the prescription more accurately represented the intended semantics of the prescription (to have medications available for an entire 6 months). I don’t have a strong rationale other than it “feels better” (a horrific reason) or perhaps that the one-row approach makes less assumptions than the N-row approach. But I really could live a very happy OHDSI life either way.

1 Like

Friends:

The intention of the CDM is to capture the exposure. Not the intent of exposure. The actual exposure, because that’s what will have a biochemical effect on the patient. Obviously, we don’t know the actual exposure. But we should say as much as we can. So:

  • For dispensing, we should have a record for each trip to the pharmacy, whether it is the initial prescription fill or a refill. And the record should state the days_supply for that one amount of the drug.
  • For prescriptions, we only know the intention, so that’s what we got.

In short, I agree with @aostropolets (as usual here).

Not sure what you mean, @mgkahn. @aostropolets says 180 days, which sounds like your 6 months. You Say ‘Tomato’, I say ‘Tomato’…

@Christian_Reich:
Regarding “The intention of the CDM is to capture the exposure. Not the intent of exposure.” If you take this statement to its logical conclusion, then we should not be including prescriptions at all. An even more extreme position (that I know you are not advocating but is consistent) would be that we also cannot trust a medication fill event since we have no indication that the patient actually consumed the medication. I think both statements are not tenable positions. I think we leave it to the investigator to determine if prescriptions are sufficient “evidence” to use or not. They can be easily be included/excluded in a query. It should NOT (added in editing) be for ETL to determine its utility in an analysis. I can think of many CER/HSR studies where prescriptions are very informative. You are showing your pharma-drug surveillance up-bringing.

Regarding @aostropolets suggested CDM representation where I came down favoring a different approach, in noodling on this for an additional 24 hours, I think @aostropolets recommended CDM representation is better since it does capture the intended distribution of medications over time (see, I am not always at odds!).

Bottom line, I am perfectly fine having intended exposure as long as it is clearly marked as being that. Let the science folks own the decision to use or not.

1 Like

Given Anna’s elaboration and artificial rows (at start of month (NOTE not equal to +DOS(of n-1 refill) we need more specific guidance for how ETL will create refill hypothethical rows.

I would also like to see values for drug_type_concept_id.

Given scenario 2 (with 1 refill) and option B for days_suply I would like to extend Anna’s data rows with actual trips to pharmacy. A site may have both and these could be studied (medication adherence study).

  • drug_type_cid=prescription start=Jan 1 2000, days_supply=90, refill=null (or 0??), end=S+90 days, quantity=90

  • drug_type_cid=dispensation start=Jan 4 2000, days_supply=90, refill=null, end=S+90 days, quantity=90

  • drug_type_cid=ETL-derived-prescription, start=prior+DOS (NOT April 1 2000), days_supply=90, refill=1, end=S+90 days, quantity=90

  • drug_type_cid=dispensation , start=April 15 (February gave me extra pills), days_supply=90, refill=1, end=S+90 days, quantity=90

  • Q1: Do we have CID for ETL-derived-prescription.

  • Q2: Where in the specs it tells me to do do start+DOS(of n-1 refill). In other words for ETL-injected row - I need to always compute prescription date + DOS (which is 90) - which gives me March 31st. (and do that for all refills)
    (btw 1 minus 1 is not null - it would be zero) for the ETL-derived-prescription row

drug_type CIDS are here: http://athena.ohdsi.org/search-terms/terms?conceptClass=Drug+Type&page=1&pageSize=15&query=

so what I call prescription would be http://athena.ohdsi.org/search-terms/terms/38000177

btw. pcornet has refils=0 and MEDICATION_EXPOSURE and “MEDICATION_INTENT”. I am with Michel on his “feeling” (share his feeling intuition). We can make OMOP more EHRish and less claim/dispens-ish . We need a “screenshot” from Epic and Cerner of how a 3 month order (with one refill) looks like from MDs point of view and what is s/he calculating in mind (care planing) and what the EHR pre-calculates for him/her. (e.g., (if sig says take 3 pills a day or take estrogen every other day and next week every 3rd day). The medication_end_date + sig is most intuitive.

Uploading…

Q2: Where in the specs it tells me to do do start+DOS(of n-1 refill)
(btw 1 minus 1 is not null - it would be zero) for the ETL-derived-prescription row

look: 2 pills for two days. You started April 1st and took the second one the next day. End_date = April 1 + 2 - 1 = April 2


This could be that the duration of drug_supply was reached (in which case drug_exposure_end_date = drug_exposure_start_date + days_supply -1)

null. The same page in spec

Prescription written it is

From Data quality stand point - if we can not distinguish imputed hypothetical ETL-generated prescription rows from real events than it will raise a data quality question - how come there is an event to the patient without a visit to any facility. (not even phone contact). It would be nice to have a different drug_type_cid for the artificially created rows.

It also violates the mantra of “record events from patient point of view”. An those hypothetical “prescription written” events did not occur.

I started an idea of synthetic demo data called Kratos data
I used the scenario of prescription and dispensation as scenario for demo patient 1

What’s that, @Vojtech_Huser? An anticipated visit to the pharmacy for a refill? Why do we need that?

Use case 1: Very useful data if one is studying drug adherence – the dates for expected medication refills as specified by the prescription (assuming full compliance_ versus observed compliance viewed from medication fulfillment records (when one has them). The discrepancy is a useful measure of compliance/adherence.

Use case 2: In the absence of fulfillment records, the analyst can decide if the sequence of anticipated refills based on the intent of the prescription is sufficient or not for whatever drug exposure question they are attempting to answer in a data set that has no (or incomplete) fulfillment records.

Kratos is in development stage. To demostrate demo data annotation, I improved the documentation. See here

sandbox/readme.md at master ¡ OHDSI/sandbox ¡ GitHub

t