OHDSI Home | Forums | Wiki | Github

Treatment Pathways: Combination drugs (Posting 1)

First of two postings. This is a repeat of a post elsewhere (Extension of Treatment Pathways study in Europe - #9 by Rijnbeek) that got no love… :wink:

We are scratching our heads about what to do with combination (multi-ingredient) medications or medications that are started on exactly the same day. We understand that only the first exposure to an ingredient is illustrated in the sunburst plot. But what should be done when that first exposure includes more than one ingredient? Was there a previous thread that I could not find that discussed how the community resolved this

1 Like

Posting 2: Our understanding of the current treatment pathway code: If a patient is exposed to Drug A with concept_id =3 and Drug B with concept_id = 2 on the same day, Drug B will be listed as the treatment and Drug A will be ignored. If the patient is then exposed to Drug C, then Drug D and then Drugs A/B a second time, Drug A will be listed and Drug B will be ignored because it was listed in the first exposure. The pathway would list: Drug B --> Drug C --> Drug D --> Drug A. But this is not correct since the patient was actually exposed to Drug A before exposed to Drug C and Drug D.

We are seeing very long treatment pathways as described above so this is not a theoretical question. How would the community like us to alter the code for dealing with combo drugs?

@mgkahn:

Why can’t you just change the code so that combinations are treated like individual drugs? So you have A, B, and A/B? And then it doesn’t matter if it is a fixed combo (one pill) or prescriptions of 2 drugs (two pills).

@Christian_Reich: This is the approach that we are currently developing – reconstitute separate ingredients into a pseudo-drug called “A/B” as you say. Pinging the community to see if others addressed this issue differently. Else we will continue to pursue the pseudo-drug approach.

I answered this but I apparently did not hit reply. Sorry. The intended effect was that combination treatments were treated as two ingredients on the same day, and that both went into the pathway. They were sorted by concept_id so that we did not artificially expand the number of pathways (didn’t have some patients as A-B and others as B-A).

You’re example would be B A C D. (Note 2<3)

If the code does not execute this, then that is a bug.

So, if you have a combo A-B then C, the algorithm reports a treatment pathway A-B-C, with 0 days between the A and the B?

Responding to @hripcsa and adding @Dekermanjian who will correct any errors since he’s the person digging thru the code. A combo drug or two independent drugs started on the same day appears as a pathway sequence based on concept_id value. So my original example where Drug B concept_Id is less than Drug A concept_id shows up as the following treatment pathway: B–> A --> C --> D and the second A/B exposure is suppressed since there are exposures to B and A earlier in the pathway. In the starburst representation, Drug B is the innermost ring, Drug A is the second ring, Drug C is the third ring and Drug D is the fourth ring.

Was this the intended behavior?

Yes, that is the intended behavior. It can bias first-line drugs, obviously. But 50-50 is not quite right either. Really would need a way to allow both to be first, or neither to be first. After going through all the ways to solve it, we decided to keep it simple

Q1: Do you see any fatal flaw that we are not considering in making a pseudo-drug A/B and putting that as the first “ingredient” when there is either a combo medication or two simultaneous first-line meds? We would use the same concept_id sort order trick to ensure that we always get A/B and not B/A.

Q2: What about the decision to not replicate a second exposure to an ingredient later in the pathway. What was the logic behind that decision?

Q1: Pseudo-drug A/B is okay but you need to decide whether to worry about A being A+A/B and so on in the displays. I think our decision was just simplicity.

Q2: Again simplicity. How often is a second exposure just a lapse of a prescription and not a true on-off-on? What if someone starts on A/B and comes off A. Is that A/B to B, or just A/B.

Q1 and Q2: Also, say one patient starts on A and B on the same day, takes them for 100 days and stops. Another patient starts A on day one, B on day two, and takes both for 100 days. Under your algorithm, they will look very different, but in reality, it was just that the pharmacy didn’t have the second drug in house for a day.

Thank you for posting this interesting topic @mgkahn
The European guideline for Hypertension 2018 will recommend to start ‘combination drug’ for treating hypertensive patients as the first line (But they will not revise the definition of hypertension, opposite to the American guideline).
So treatment pattern of combination drugs gets more importance (especially in hypertension).

I also prefer @Christian_Reich’s solution

But we need to define ‘what is combination?’ as @hripcsa concerned
If treatment pathway only represent drug patterns in chronic diseases, I want to suggest 30-day rule (or 7-day rule).

In this rule,

  • Only when A and B drug are concurrently used more than 30 days, it can be considered as combination. (Three or four combination is possible)
  • If we remove drug era shorter than 30 days, we can remove many noises. If the patient starts A and B on the same day, and then stopped B 10 days after initiation, B drug would be ignored (The diagram shows that the patient use only A). If the patient stopped B 45 days after initiation, the single A drug use would be the second line of treatment.

And, How about using Sankey diagram? This diagram seems to be able to represent the reverse order of drug usage pattern and more flexible

1 Like

My two cents: in LEGEND we use a slightly different definition of a combination treatment. We define it as:

  1. Initiation of A and B on the same day (so no exposure to A or B before the day they both start).
  2. Initiation of B while still on A, with at least one prescribing/dispensing of A after/on the date of initiation of B to rule out switching.

For those interested in how it is coded:

My recommendation is the following:

  1. If you expect combinations then you need to treat the intake of multiple drugs on one day as a pseudo-drug (as others proposed).
  2. A patient is on multiple drugs not only when multiple drugs are initiated on a single day. An add-on of another drug later on constitutes a combination therapy as well.
  3. You will observe a lot of noise. One major reason for this noise is generated by refills. You can’t expect a patient initiates a refill exactly at the time it is necessary. A compliant patient will organize the medication before it is used up. This will cause overlaps and you have to consider these overlaps most of the time. Furthermore, you need rules for gap filling (minor gaps between same drug episodes).
  4. In order to calculate these mono-episodes accordingly, you need the DRUG_EXPOSURE.verbatim_end_date information or calculate these with appropriate methods.
  5. You will still have some noise caused by extraneous add-ons and drop-offs. These can be addressed by different approaches. A few examples:
  • Choose only the “most important” combination within time intervals per patient. This pre-dominant combination is the longest combination in the given time interval.
  • Suppress combination episodes of less than x days.
  • Suppress combination episodes with less than x patients.
  • Consider the top x combinations with respect to the number of days.
  • Consider the top x combinations with respect to the number of patients.
  • While the first option is usually not combined with other options, the last four approaches are often combined.

Hope this helps.

1 Like
t