combination sequential and substitutes operators on drug collections

[ahwagner]

The CIViC documentation lacks clear definitions of what the drug interaction types (combination, sequential, and substitutes) mean and when it is appropriate to use one over another.

Starting the discussion here.

[arpaddanos]

My understanding is

• combination == A and B at the same time
• sequential == First A and then B without overlap
• substitutes == A or B could be used.

An example of substitutes could be in civic:AID5 (attempt at a Curie notation there). It talks about the approval for the first generation TKIs erlotinib and gefitinib. They can be used interchangeably, one or the other in EGFR mutation NSCLC

[arpaddanos]

The documentation should be updated once there is consensus and other more detailed questions from @ahwagner are resolved

[ahwagner]

@arpaddanos your definitions mirror my understanding. Beyond this, though, is the need for formal definitions and guidance on when the use of each. The motivation for this comes from the modeling I am doing for MetaKB, where I have created three distinct TherapeuticCollection subclasses:

  CombinationTherapeutic:
    inherits: TherapeuticCollection
    description: >-
      A set of therapeutics that are taken together as a treatment.

  SequentialTherapeutic:
    inherits: TherapeuticCollection
    description: >-
      A list of therapeutics that are taken sequentially as a treatment.

  SubstitutionTherapeutic:
    inherits: TherapeuticCollection
    description: >-
      A set of therapeutics that are considered valid alternative treatments.

NOTE: I use Therapeutic here instead of Drug as the term encompasses concepts such as Radiation Therapy; but the principle of these classes is meant to reflect the "drugs + interaction type" concept from CIViC.

One of the challenges I have with this structure is the notion of SequentialTherapeutic. Is the order meaningful in the CIViC data model? In general? More concretely, if a study describes the role of Drug A then Drug B, and another study describes the role of Drug B then Drug A, is it important to the interpretation of the variant to distinguish those statements computationally? Or is the goal simply to index the collection of therapeutics associated with some sequential application of those therapeutics (a regimen)?

[ahwagner]

Another question is about the use of Substitution. If this means Drug A or Drug B could be used, how is that different from two CIViC evidence items, one about Drug A, and one about Drug B? Or is it not different, but used only as a convenient shortcut in lieu of creating two independent evidence items in such cases? Are these drugs supposed to reflect two of a collection of drugs with a shared class, e.g. Crizotinib and Alectinib as substitute representatives of the broader class of ALK Kinase inhibitors? Or is it a combination of these things?

[acoffman]

One of the challenges I have with this structure is the notion of SequentialTherapeutic. Is the order meaningful in the CIViC data model? In general? More concretely, if a study describes the role of Drug A then Drug B, and another study describes the role of Drug B then Drug A, is it important to the interpretation of the variant to distinguish those statements computationally? Or is the goal simply to index the collection of therapeutics associated with some sequential application of those therapeutics (a regimen)?

@ahwagner I can't speak to the importance of capturing that information, but in terms of goals or data model I can say we are not currently capturing the sequence order, merely the collection, and I don't know that we've ever really discussed doing so.

Also, longer term, I believe we are also intending to switch from Drug -> Therapy so our terminology will be in alignment there.

[kkrysiak]

Agreed with @acoffman above for sequential being stored in an unordered list and change of drug term coming. Also, with @arpaddanos about the basic definitions of these 3 drug interaction types.

A few things:

• We unfortunately also have drug combinations / therapeutic regimens that are more complex without being labeled with an interaction type because these are supported by NCIt and often more complex (getting at what you've described above). For example CHOP Regimen (C9549).
• Substitutions are most often used for drug classes like EGFR inhibitors being testing in a clinical setting or on cell lines. In some cases these could be separated into multiple EIDs but other times the statistics are really powered by a cohort of patients in which both have occurred. Example: https://civicdb.org/evidence/2115/summary