OHDSI Home | Forums | Wiki | Github

Limits of useful precoordination


#1

I am working on creating a vocabulary for a set of cancer treatment regimens (similar to HemOnc vocabulary)

My question is just to get a sense of the wisdom of defining these regimens at a significantly higher granularity than HemOnc, whereby we would want to define not only the drugs that belong to a given regimen, but also their dosage and schedule.

The goal of getting more specific is in measuring unwarranted variation from the specified protocol (e.g. underdosing, changes in schedule), so there is a specific reason for wanting to spell out precisely the how and when, not just the what of each protocol, but since it introduces more complexity than I could find in similar concepts and vocabularies, I was hoping there would be someone who had achieved (or come up against a road-block to achieving) similar goals to give some guidance before I head too far down that path however.

The draft definition for the treatment protocol looks something like this (chemotherapy example - only partially transcribed - NB ~ placeholder for fields not relevant for this protocol just to give an idea of the full set of fields).

The idea is that a protocol may be made up of one or more regimens. Regimens typically defined as treatment phases (Induction / Consolidation / Maintenance) across a protocol.

  • Protocol (Autologous conditioning protocol germ cell tumour TICE treatment overview)
    • Regimen (Autologous conditioning protocol germ cell tumour TICE (PACLItaxel and IFOSFamide) (part 1)
      • Has intent (~)
      • Has type (~)
      • Has indication (Relapsed and refractory metastatic germ cell tumours)
      • Has contra-indication (~)
      • Has cycles (2)
        • Has frequency (14 days)
        • Has duration (~)
        • Has treatment Session (session name tbd)
          • Is scheduled on day (1)
          • Has treatment (treatment name tbd)
            • Drug (Loratadine)

              Dose (10 mg)
              Route (Oral)
              Form (~)

          • Has treatment (treatment name tbd)
            • Drug (Ranitidine)

              Dose (50 mg)
              Route (IV)

          • Has treatment (treatment name tbd)
            • Drug (Dexamethasone)

              Dose (20 mg)
              Route (IV)


(Christian Reich) #2

@gkennos:

Sounds like a nice idea. It’s not going to be easy to establish all that information and keeping it fresh, though. Couple thoughts:

  • I’d generally not focus on things like intent, indication, contra-indication. That is information for guidelines, which in desperate situations will be overwritten. It is also hard to check them against reality in the data. Also, it is under pressure to change all the time, and you’d be chasing those debates.
  • Your nested logic of Regimen - Cycle - Session - Treatment makes sense. Not sure why Drug is its own level, since each Treatment contains one Drug.
  • You may need a duration for each session or treatment.
  • Dose may not be an absolute value, but something in relationship to body surface.
  • You may need more than one Route or Form.
  • Obviously I’d recommend Concept IDs rather than strings for most of these things. Certainly Drugs, Routes, Forms.

But why is the whole post called “Limits of useful precoordination”? I am not seeing any precoordination. This is a reference table.


(Jeremy Warner) #3

Hi @gkennos before going too much further down this road you should know that we already have all of these concepts, and more, in the full HemOnc vocabulary. Please contact me to discuss further, we’re happy to give you a copy of the full vocabulary for academic or non-commercial uses!


#4

Hi @Christian_Reich, Thanks for the feedback.

  • I’d generally not focus on things like intent, indication, contra-indication. That is information for guidelines, which in desperate situations will be overwritten. It is also hard to check them against reality in the data. Also, it is under pressure to change all the time, and you’d be chasing those debates.

I am trying to establish a new vocabulary that actually reflects the guidelines themselves, because we are looking to measure variance in treatment against the relevant guidelines.

We are automating the consumption of the guideline documents themselves so that changes can be detected and reflected periodically - the point of this is really to capture reality vs. prescribed intention

I called it precoordination because I guess the question is really how much of this information should be encoded in the vocabulary itself and how much of it could be recorded elsewhere. I was hoping to precoordinate as much as possible so that we could detect variances in a variety of groupings (e.g. for all regimens containing drug x, how much was underdosing a factor? for all regimens indicated for condition y, how much variance was seen in timing of treatments against the prescribed schedule?). Since it’s to act as a reference rather than an observation, I expected that full precoordination of concepts into a top level regimen concept defintion was the way to go, but was worried that was overcomplicating it?

@Jeremy_Warner has kindly offered to meet with us so will discuss the HemOnc experience offline, but in case anyone comes back to this thread with similar goals, I’ll try summarise at a high level and post back after we have spoken, because I wasn’t able to find similar guidance.

  • Your nested logic of Regimen - Cycle - Session - Treatment makes sense. Not sure why Drug is its own level, since each Treatment contains one Drug.

Good point - at the moment only looking at chemotherapy regimens so I was keeping that extra level thinking it may be necessary for radiation but that’s TBC so will remove if it isn’t required.

  • You may need a duration for each session or treatment.

All sessions are defined in our guidelines by start day - this will also need to be reviewed when we expand to the radonc set though.

  • Dose may not be an absolute value, but something in relationship to body surface.
  • You may need more than one Route or Form.

Yes, this wasn’t the case in this example, but there are definitely instances where we will have to capture a relative value / alternative forms.

  • Obviously I’d recommend Concept IDs rather than strings for most of these things. Certainly Drugs, Routes, Forms.

Yes, we will be defining our own concept ids for all levels and mapping to rxnorm or other relevant vocabularies


(Christian Reich) #5

That makes sense and is definitely a good use case. And a ton of work. Good luck. Let us know if you need help from the Working Group.


t