Re: 7 days left to comment (was Re: representing potential drug-drug interaction knowledge and evidence - Last commenting period for the W3C Community Group report

Dear Colin and other interested list members,

We worked through all of the comments and have produced a new draft version of the Report: https://w3id.org/hclscg/pddi

Nearly all comments besides Colin's were grammatical and spelling. Multiple people thought a figure would help provide an overview of the Report's recommendations. That is added now (https://w3id.org/hclscg/pddi#recommendations-summary)

I have replied to the remaining comments below. I think we have addressed everything as best as we can. If anyone has an objection to the Community Group proceeding to publish the note please let me know by next Tuesday 3/2.

Thank you very much,
-Rich

---
The one area that I might take issue with the model itself is in your definition of <Contextual information/modifying factors>. In particular, the inclusion of product formulation (dose form) in this category. A medicinal product's dose form is one of its defining characteristics and it is therefore not a feature that can be changed without changing the medicinal product entity itself. And if you change the medicinal product entity then you have a different PDDI (because the <Drugs Involved> have changed). The report illustrates this nicely with the example PDDI between warfarin and NSAIDs, in particular the focus on warfarin and diclofenac. On the one-hand  there is <topical diclofenac> for which the Operational Classification yields "No special precautions", and on the other-hand there is <non-topical diclofenac> (let's assume for a minute that that means oral/parenteral diclofenac) for which the Operation Classification indicates that evasive action may be necessary. I would argue that the description of the clinical consequences ("Increased risk of bleeding including gastrointestinal bleeding, intracranial hemorrhage, and cerebral hemorrhage") is simply not accurate for <warfarin + topical diclofenac> and that's because <topical diclofenac> is a fundamentally different medicinal product to <oral/parenteral diclofenac> with a fundamentally different drug interaction profile.

This is quite different from considering a PDDI that is modified because a patient's renal function is significantly impaired. Here, the PDDI and its associated clinical consequence are essentially similar irrespective of the patient's renal function, but the impact of that PDDI may be very different because of the patient's inability to clear a metabolite, for instance.

Staying with the NSAID example, drug class-level PDDIs are fine in principle, but all members of that drug class must have uniform drug interaction characteristics. That is not something that can be said of NSAIDs, at least not as the report defines them in A.1.4.12. For instance, the NSAID value set includes <1236089 | Benzydamine hydrochloride 3mg oral lozenge> which, according to the UK Summary of Product Characteristics produces peak plasma levels that are "not sufficient to produce pharmacological systemic effects". So, like the topical diclofenac discussed earlier, benzydamine lozenges have a very different drug interaction profile to say, parenteral ketorolac or oral celecoxib. As such they cannot all be in the same grouper interactant (namely "NSAIDs").

Another difficulty with the NSAID value set is that it includes ingredients e.g.  <3355 | Diclofenac> as well as clinical drugs e.g. <1442116 | Diclofenac 18mg oral capsule>. Just as an ingredient cannot treat a disease, so too ingredients are not capable of participating in PDDIs — this is a feature of formulated medicinal products designed to be administered to patients. With this in mind, it would be better to avoid including ingredient concepts in your NSAID value set.

In Section 3.1, describing the interactant as "NOT topical diclofenac" creates an ambiguity since this entity includes oral, parenteral, rectal, and ophthalmic forms of diclofenac. The latter has a different drug interaction profile to the other dose forms which deliver active ingredient to the systemic circulation and, like topical diclofenac, needs to be considered as a separate entity to the oral/parenteral/rectal formulations.


REPLY: You make some valuable and interesting points. Let me start by stating that we did not notice the issue with  "1236089 - Benzydamine Hydrochloride 3 MG Oral Lozenge" and I have now removed that entity from the value sat for NSAIDs. There are three comments that I would make in reply to your points:

- about the issue with describing the interactant as "NOT topical diclofenac" : I was a part of the clinical sub-team calls that discussed this approach and remember them discussing the oral, parenteral, rectal, and ophthalmic forms specifically. After looking at the pharmacology evidence, their decision was to treat them all as potentially bioavailable to a much greater degree than topical diclofenac. The logic used in the Report reflects this decision. While it is entirely reasonable that there are reasonable, evidence-based arguments, for doing this differently, these are exemplars and not intended for direct integration into a clinical use. I added the following sentence with emphasis to the first paragraph of Section 3 where the examples are discussed: "Note that while the examples shown here are realistic, they are not intended for direct integration into a clinical applications."

- about including ingredients in the some value sets : I think I understand your point and agree that in the case of PDDI alerts presented during CPOE, it is operationally more precise to avoid using ingredents in the value sets. However, we could not recommend that our intended audience (persons or organizations who publish PDDI information) avoid using ingredients in value sets for a couple of reasons. One reason is that both drug classes and ingredients are very common in the domain of discourse but specific clinical drug forms are not. The greater issue we tried to address was that drug class mentions are highly ambiguous since no two drug classification systems assign the same members to classes with the same or similar names. The examples we show are emphasizing that it is better for stakeholders to enumerate the ingredients and clinical drug forms that they believe belongs to a drug class that to just state some abiguous class name (as is usual current practice). The second reason is that the topic of appropriate value set creation is more detailed than we had the ability to fit into the Report. A related HL7 project is addressing that for the PDDI use case as part of an implementation guide. For example, the terminology page of the draft HL7 implementation guide (http://build.fhir.org/ig/HL7/PDDI-CDS/terminology.html) describes how to compose drug value sets using either intensional or extensional definitions, and do so in a way that allows for internationalization. This work is mentioned in A.1.4 of the report.

- about definition of <Contextual information/modifying factors> : As I read your comment, I think you might be suggesting that Clinical Consequence statements should be as specific as possible to the medicinal products involved in the potential interaction. Using the warfarin - NSAIDs example, we would perhaps create a "warfarin - systemic NSAIDs" information artifact that was only about the potential interaction between warfarin and systemic forms of clinical NSAID drugs. I would say that the examples in the Report do not necessarily disagree with this. Rather, they reflect the scientific discussion that occurred about the PDDIs. Section A.1.1.2 shows how the two pairs were selected along with 12 others. The team started with the pairs (drug-drug or drug-class or class-class) and then worked through the evidence to "fill in" the min info model. So, the first question for warfarin and NSAIDS was "what drug entities do we mean by the mention of <drug> or <class>?". Then, "is there evidence that a <clinical consequence> could occur between members ot the two sets of enumerated entities?". Then, "what  <Contextual information/modifying factors>?". After answering those questions, we could have decided to modify the PDDI to be  "warfarin - systemic NSAIDs". However, an advantage of this approach we took is that is a bit more comprehensive from a drug information perspective. For example, a CDS system that follows e.g., the warfarin - NSAIDs example can track why it is that an interuptive alert should NOT trigger for topical diclofenac. This could be useful for auditing.


Section A.1.4.2 is an example of where RxNorm is used to group together both single-ingredient medicines and multi-ingredient medicines within a single-ingredient grouper (in this case, "aldosterone antagonists"). This gives the impression that, for instance, <198224 | Hydrochlorothiazide 25mg / Spironolactone 25mg oral tablet> is a type of aldosterone antagonist which, as a combination product, it cannot be. A better alternative would be to use SNOMED CT which has made significant changes in the last couple of years to its handling of multi-ingredient medicinal products; for example, <SCTID: 346312000 | Product containing aldosterone receptor antagonist>.

REPLY: Thank you for pointing this out. I am not able to change the value set at this time but did change the label to "aldosterone antagonists and products containing an aldosterone receptor antagonist"


Based on the content of Section A.1.4.4 I suspect you are referencing ICD-10-CM rather than the unmodified ICD-10. If that is indeed the case, it would be worth specifying that in the section title.

REPLY: Done - thank you.

Now, I'm not an expert in anatomy but it looks to me like the title of Section A.1.4.4 is inaccurate because there are several disorders listed that are not cerebral hemorrhages. For instance, subdural, subarachnoid, and cerebellar hemorrhages are siblings of cerebral hemorrhage not subtypes of it. Also, <S06.8 | Other specified intracranial injuries> (the last item in this listing) includes some descendant concepts that are not hemorrhagic in nature.

REPLY: These are removed now - thank you.

In Section 3.2 the clinical consequence uses the term "chronic myeloid leukemia" to refer to a disorder. However, in Section A.1.4.5 where the corresponding SNOMED CT codes are listed, the last item in the list (<SCTID: 128826001 | Atypical chronic myeloid leukemia, BCR/ABL negative>) represents a morphological abnormality. As such, this would not typically be found in a patient's problem list.

REPLY: Fixed - thank you.


Finally, I think it would be more accurate to label Section A.1.4.7 as "Upper gastrointestinal bleeding…".

REPLY: Thank you - fixed.




On 3/21/19 9:06 AM, Macfarlane, Colin R. (ELS-LOW) wrote:

Dear Rich,



It is great to finally have a model in the drug interaction domain that is grounded in the clinical requirements of day-to-day practice. The comments I have are not about the model per se (which I think is spot on) but about how you represent the model in your report and about some of the value sets used to illustrate the examples.



Firstly, it would be nice to see some graphical representation of the model; this would certainly help reader-interpretation of the narrative text and ensure that, as readers, we are envisaging the model in the same way as the development group. For instance, this would help determine whether <Seriousness> is an attribute of <Clinical consequences> or an attribute of the <PDDI> directly.



The one area that I might take issue with the model itself is in your definition of <Contextual information/modifying factors>. In particular, the inclusion of product formulation (dose form) in this category. A medicinal product's dose form is one of its defining characteristics and it is therefore not a feature that can be changed without changing the medicinal product entity itself. And if you change the medicinal product entity then you have a different PDDI (because the <Drugs Involved> have changed). The report illustrates this nicely with the example PDDI between warfarin and NSAIDs, in particular the focus on warfarin and diclofenac. On the one-hand  there is <topical diclofenac> for which the Operational Classification yields "No special precautions", and on the other-hand there is <non-topical diclofenac> (let's assume for a minute that that means oral/parenteral diclofenac) for which the Operation Classification indicates that evasive action may be necessary. I would argue that the description of the clinical consequences ("Increased risk of bleeding including gastrointestinal bleeding, intracranial hemorrhage, and cerebral hemorrhage") is simply not accurate for <warfarin + topical diclofenac> and that's because <topical diclofenac> is a fundamentally different medicinal product to <oral/parenteral diclofenac> with a fundamentally different drug interaction profile.



This is quite different from considering a PDDI that is modified because a patient's renal function is significantly impaired. Here, the PDDI and its associated clinical consequence are essentially similar irrespective of the patient's renal function, but the impact of that PDDI may be very different because of the patient's inability to clear a metabolite, for instance.



Staying with the NSAID example, drug class-level PDDIs are fine in principle, but all members of that drug class must have uniform drug interaction characteristics. That is not something that can be said of NSAIDs, at least not as the report defines them in A.1.4.12. For instance, the NSAID value set includes <1236089 | Benzydamine hydrochloride 3mg oral lozenge> which, according to the UK Summary of Product Characteristics produces peak plasma levels that are "not sufficient to produce pharmacological systemic effects". So, like the topical diclofenac discussed earlier, benzydamine lozenges have a very different drug interaction profile to say, parenteral ketorolac or oral celecoxib. As such they cannot all be in the same grouper interactant (namely "NSAIDs").



Another difficulty with the NSAID value set is that it includes ingredients e.g.  <3355 | Diclofenac> as well as clinical drugs e.g. <1442116 | Diclofenac 18mg oral capsule>. Just as an ingredient cannot treat a disease, so too ingredients are not capable of participating in PDDIs — this is a feature of formulated medicinal products designed to be administered to patients. With this in mind, it would be better to avoid including ingredient concepts in your NSAID value set.



In Section 3.1, describing the interactant as "NOT topical diclofenac" creates an ambiguity since this entity includes oral, parenteral, rectal, and ophthalmic forms of diclofenac. The latter has a different drug interaction profile to the other dose forms which deliver active ingredient to the systemic circulation and, like topical diclofenac, needs to be considered as a separate entity to the oral/parenteral/rectal formulations.



In Section 4.2, first paragraph, there is a typo: "pharmacists"



Section A.1.4.2 is an example of where RxNorm is used to group together both single-ingredient medicines and multi-ingredient medicines within a single-ingredient grouper (in this case, "aldosterone antagonists"). This gives the impression that, for instance, <198224 | Hydrochlorothiazide 25mg / Spironolactone 25mg oral tablet> is a type of aldosterone antagonist which, as a combination product, it cannot be. A better alternative would be to use SNOMED CT which has made significant changes in the last couple of years to its handling of multi-ingredient medicinal products; for example, <SCTID: 346312000 | Product containing aldosterone receptor antagonist>.



Based on the content of Section A.1.4.4 I suspect you are referencing ICD-10-CM rather than the unmodified ICD-10. If that is indeed the case, it would be worth specifying that in the section title.



Now, I'm not an expert in anatomy but it looks to me like the title of Section A.1.4.4 is inaccurate because there are several disorders listed that are not cerebral hemorrhages. For instance, subdural, subarachnoid, and cerebellar hemorrhages are siblings of cerebral hemorrhage not subtypes of it. Also, <S06.8 | Other specified intracranial injuries> (the last item in this listing) includes some descendant concepts that are not hemorrhagic in nature.



In Section 3.2 the clinical consequence uses the term "chronic myeloid leukemia" to refer to a disorder. However, in Section A.1.4.5 where the corresponding SNOMED CT codes are listed, the last item in the list (<SCTID: 128826001 | Atypical chronic myeloid leukemia, BCR/ABL negative>) represents a morphological abnormality. As such, this would not typically be found in a patient's problem list.



Finally, I think it would be more accurate to label Section A.1.4.7 as "Upper gastrointestinal bleeding…".



Once again, this report is a great piece of work and I think will be extremely valuable in the future.



Kind regards

Colin Macfarlane

Colin Macfarlane
Medicines Optimisation Team Lead
ELSEVIER | Clinical Solutions



--
Richard D Boyce, PhD
Associate Professor of Biomedical Informatics and Clinical and Translational Science in the Clinical and Translational
Science Institute
Director of the Informatics Core for the Center of Excellence for Natural Product- Drug Interaction Research (NaPDI)
Faculty, Center for Pharmaceutical Policy and Prescribing
Faculty, Geriatric Pharmaceutical Outcomes and Gero-Informatics Research and Training Program
University of Pittsburgh
rdb20@pitt.edu<mailto:rdb20@pitt.edu>
Office: 412-648-9219
Twitter: @bhaapgh