address

bibliography

title

King's College London

bib.bib

CONSULT guideline server: TMRweb

Introduction

This document summarises how to interact with the CONSULT guidelines service, TMRweb, which packages TMR [@Zamborlini2016] as a web service, allowing for the representation of, and identification of interactions between, clinical guidelines, specifically drug administration, in an interoperable manner.

Scenario

{#HT width="0.8\linewidth"}

Figure 1{reference-type="ref" reference="HT"} shows two hypertension guidelines that are alternatives, in that they express the same information. This is also a type of alternative known as a repairable transition, because the actions undertaken by one guideline can be reversed by performing the actions associated with the other. In text form, these guidelines might read 'To reduce a patient's blood pressure, administer Thiazide. Similarly, to avoid exacerbating a patient's blood pressure, avoid the administration of Ibuprofen.'

In the example detailed in this document, our aim is to represent these guidelines using the semantic format used by TMR, and to then use a computational implementation of TMR to identify the repairable transition shown, features which are both offered by TMRweb. This then serves as an example for how to use TMRweb to create new guideline sets, and identify interactions between the constituent guidelines.

In Figure 1{reference-type="ref" reference="HT"}, the second guideline, Reduce Blood Pressure, will be referred to using the ID Diuretic; the first, Avoid High Blood Pressure, using the ID Diuretic2; and the guidelines as a collection using the ID HT.

Implementation

TMRweb is a RESTful web service, which accepts HTTP POST requests. Therefore, to construct our semantic representation, and to then interrogate this implementation, requests will be issued to this service, using the CURL command¹. Examples are given throughout the document, and should always be accompanied by the following header information:

  --header `Content-Type: application/x-www-form-urlencoded' \
  --header `Authorization: Basic M01FdlFrdW46WVdoQ1BCSDk='

Example representation process {#example}

To represent the set of guidelines shown in Figure 1{reference-type="ref" reference="HT"} and have TMRweb successfully leverage TMR to identify the interactions between them, we first need to define the world knowledge that supports their definition and allows TMR to operate: the existence of drugs; the concept of a patient having a certain medical state, such as a blood pressure level; the concept of altering a person's medical state, such as their blood pressure; and the effects of taking a drug.

Drugs

To define a drug, we might start by representing a general category of drugs. Here, we create a dummy category Thiazide for Diuretic:

curl --request POST \
  --header `Content-Type: application/x-www-form-urlencoded' \
  --header `Authorization: Basic M01FdlFrdW46WVdoQ1BCSDk=' \
  --url https://kclhi.org/tmrweb/drug/category/add \
  --data `drug_category_id=Thiazide'

When we define a category, we can also specify which drugs are in this category (in this case only Thiazide), and general properties of the drugs in this category, but this falls outside the scope of this example.

 subsumed_drug_ids=Aspirin%2C%20Ibuprofren&grouping_criteria_ids=SitNonSteroidalDrug%2C%20TrAntinflammatory`

Given this category, we can now define the actual Diuretic drug, and state that it is part of this category. We can also state individual relationship to other drugs (in this case not one we have priorly represented):

curl --request POST \
  --url https://kclhi.org/tmrweb/drug/individual/add \
  --data`drug_id=Thiazide&drug_category_id=Thiazide&subsumed_drug_id=Bendroflumethiazide'

We follow a similar approach for Diuretic2, which pertains to the drug Ibuprofen and in this instance does not have a category:

curl --request POST \
  --url https://kclhi.org/tmrweb/drug/individual/add \
  --data drug_id=Ibuprofen

We have now defined the drugs shown in Figure 1{reference-type="ref" reference="HT"}.

Situation

We next define the situations a patient might find themselves in, in respect of their vitals, which in the case of HT is varying levels of blood pressure (Figure 1{reference-type="ref" reference="HT"}). Specifically we define the state in which a patient has normal blood pressure, and the state in which a patient has high blood pressure. We can also add additional information such as the clinical codes that might be used to reference such states in an EHR:

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/situation/add \
  --data `situation_id=NormalBP&situation_label=Blood%20pressure%20is%20normal'

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/situation/add \
  --data `situation_id=HighBP&situation_label=Blood%20pressure%20is%20high&umlsCodes=C0020538%2C%20C3843080'

Transition

We next model transitions between the situations specified previously, specifically moving between different blood pressure levels, as shown in Figure 1{reference-type="ref" reference="HT"}.

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/add \
  --data `transition_id=IncreaseBP&prior_situation_id=NormalBP&post_situation_id=HighBP'

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/add \
  --data `transition_id=DecreaseBP&prior_situation_id=HighBP&post_situation_id=NormalBP'

Beliefs

Finally, we combine our drug information (Section 4.1{reference-type="ref" reference="drugs"}) and transition information (Section 4.3{reference-type="ref" reference="transition"}) to construct beliefs about the effects of administering a drug, for Diuretic and Diuretic2 (Figure 1{reference-type="ref" reference="HT"}):

curl --request POST \
  --url https://kclhi.org/tmrweb/belief/add \
  --data `belief_id=ThiazideBP&drug_cause_id=Thiazide&transition_effect_id=DecreaseBP&strength=L1&frequency=always&author=martin'

curl --request POST \
  --url https://kclhi.org/tmrweb/belief/add \
  --data `belief_id=IbuprofenBP&drug_cause_id=Ibuprofen&transition_effect_id=IncreaseBP&strength=L1&frequency=always&author=martin'

We can add additional information, such as a code to indicate the strength of the belief, and how often this belief applies to the referenced transition.

Guidelines

With all of our background information specified, we can now construct our actual guidelines, combined in a guideline group, which brings all of this information together.

First, we supply some details for our guideline group²:

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/create \
  --data `guideline_group_id=HT&description=CIG%20for%20hypertension'

Then, we specify information for Diuretic, including the remaining information from Figure 1{reference-type="ref" reference="HT"}, such as the guideline name, and the nature of the associated recommendation (should or should not):

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/add \
  --data `guideline_group_id=HT&guideline_id=Diuretic&drug_id=Thiazide&belief_id=ThiazideBP&label=Reduce%20blood%20pressure&should_or_shouldnot=should&author=martin'

Diuretic2 is also defined accordingly:

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/add \
  --data `guideline_group_id=HT&guideline_id=Diuretic2&drug_id=Ibuprofen&belief_id=IbuprofenBP&label=Avoid%20high%20blood%20pressure&should_or_shouldnot=should-not&author=martin'

Example interrogation process (single interaction)

Now that we have our guideline set represented, we can interrogate it in order to find any interactions in the following way:

curl --request POST \
  --url https://kclhi.org/tmrweb/guidelines/interactions \
  --data guideline_group_id=HT

This gives us the following response:

[interaction(http://anonymous.org/data/ReparableTransitionRecHT-Diuretic2RecHT-Diuretic,Reparable Transition,[http://anonymous.org/data/RecHT-Diuretic,http://anonymous.org/data/RecHT-Diuretic2],[])]

The first element of this interaction object is a unique name for the identified interaction, the next the type of interaction, the third the fully qualified IDs, and finally any additional external information. We can see that, as expected, the TMR reasoning engine within TMRweb has identified the interaction shown within Figure 1{reference-type="ref" reference="HT"}.

Using this information, we can learn more about the guidelines involved in the interaction, and thus re-acquire the information shown in Figure 1{reference-type="ref" reference="HT"}, from TMRweb. For example, we can first ask which drug each guideline in the interaction relates to:

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/drug/get \
  --data `guideline_id=http%3A%2F%2Fanonymous.org%2Fdata%2FRecHT-Diuretic&guideline_group_id=HT'

http://anonymous.org/data/DrugCatThiazide

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/drug/get \
  --data `guideline_id=http%3A%2F%2Fanonymous.org%2Fdata%2FRecHT-Diuretic2&guideline_group_id=HT'

http://anonymous.org/data/DrugTIbuprofen

We can then ask what the effects of these drugs are, and thus learn why they are considered alternative actions:

curl --request POST \
  --url https://kclhi.org/tmrweb/drug/effect/get \
  --data `drug_full_id=http%3A%2F%2Fanonymous.org%2Fdata%2FDrugCatThiazide'

http://anonymous.org/data/ActAdministerThiazide causes http://anonymous.org/data/TrDecreaseBP

curl --request POST \
  --url https://consultin.hscr.kcl.ac.uk/tmrweb/drug/effect/get \
  --data `drug_full_id=http%3A%2F%2Fanonymous.org%2Fdata%2FDrugTIbuprofen'

http://anonymous.org/data/ActAdministerIbuprofen causes http://anonymous.org/data/TrIncreaseBP

Example interrogation process (multiple interactions)

{#HT-OA width="0.8\linewidth"}

In order to illustrate the identification of multiple guideline interactions, we add a new guideline, Painkiller, shown in Figure 2{reference-type="ref" reference="HT-OA"} along with Diuretic and Diuretic2. With the addition of this drug, we now also have a contradiction interaction: two guidelines in the same set that recommend, and do not recommend, the administration of the same drug, respectively. To model and identify this through TMRweb, first we model the new transitions and beliefs associated with this guideline (Ibuprofen is already modelled):

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/situation/add \
  --data `situation_id=PatientHasNoPain&situation_label=Patient%20has%20no%20pain'

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/situation/add \
  --data `situation_id=PatientHasPain&situation_label=Patient%20has%20pain&umlsCodes=C0030193'

curl --request POST \
  --url https://kclhi.org/tmrweb/transition/add \
  --data `transition_id=Painkiller&prior_situation_id=PatientHasPain&post_situation_id=PatientHasNoPain'

curl --request POST \
  --url https://kclhi.org/tmrweb/belief/add \
  --data `belief_id=IbuprofenPain&drug_cause_id=Ibuprofen&transition_effect_id=Painkiller&strength=L1&frequency=always&author=martin'

Next, we again model Diuretic and Diuretic2 as part of a new guideline group along with our new guideline Painkiller:

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/create \
  --data `guideline_group_id=HT-OA&description=CIG%20for%20hypertension%20and%20osteoarthritis'

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/add \
  --data `guideline_group_id=HT-OA&guideline_id=Diuretic&drug_id=Thiazide&belief_id=ThiazideBP&label=Reduce%20blood%20pressure&should_or_shouldnot=should&author=martin'

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/add \
  --data `guideline_group_id=HT-OA&guideline_id=Diuretic2&drug_id=Ibuprofen&belief_id=IbuprofenBP&label=Avoid%20high%20blood%20pressure&should_or_shouldnot=should-not&author=martin'

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/add \
  --data `guideline_group_id=HT-OA&guideline_id=Painkiller&drug_id=Ibuprofen&belief_id=IbuprofenPain&label=Reduce%20pain&should_or_shouldnot=should&author=martin'

We can now interrogate these guidelines in order to identify interactions:

curl --request POST \
  --url https://kclhi.org/tmrweb/guidelines/interactions \
  --data `guideline_group_id=HT-OA'

[interaction(http://anonymous.org/data/ContradictionRecHT-OA-PainkillerRecHT-OA-Diuretic2,Contradictory Norms,[http://anonymous.org/data/RecHT-OA-Diuretic2,http://anonymous.org/data/RecHT-OA-Painkiller],[]),
interaction(http://anonymous.org/data/ReparableTransitionRecHT-OA-Diuretic2RecHT-OA-Diuretic,Reparable Transition,[http://anonymous.org/data/RecHT-OA-Diuretic,http://anonymous.org/data/RecHT-OA-Diuretic2],[])]

This time, we have two interaction objects. This second we have already seen, but the first now identifies the contradiction shown in Figure 2{reference-type="ref" reference="HT-OA"}. From this information, we can deduce further information such as the fact that Painkiller is, through the fact that Diuretic2 and Diuretic are alternatives (repairable transitions), also contradictory with Diuretic.

If we identify the drugs involved in this inferred contradiction, then we can further infer that Thiazide (Diuretic) and Ibuprofen (Painkiller) should not be prescribed together.

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/drug \
  --data `guideline_id=http%3A%2F%2Fanonymous.org%2Fdata%2FRecHT-Diuretic&guideline_group_id=HT-OA'

http://anonymous.org/data/DrugCatThiazide

curl --request POST \
  --url https://kclhi.org/tmrweb/guideline/drug \
  --data `guideline_id=http%3A%2F%2Fanonymous.org%2Fdata%2FRecHT-OA-Painkiller&guideline_group_id=HT-OA'

http://anonymous.org/data/DrugTIbuprofen

Footnotes

Postman (https://www.getpostman.com/), neatly wraps CURL commands. ↩
For now, we pre-configure TMRweb with each guideline group ID, outside of the REST interface. ↩

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

README.md

README.md

Introduction

Scenario

Implementation

Example representation process {#example}

Drugs

Situation

Transition

Beliefs

Guidelines

Example interrogation process (single interaction)

Example interrogation process (multiple interactions)

Files

README.md

Latest commit

History

README.md

File metadata and controls

Introduction

Scenario

Implementation

Example representation process {#example}

Drugs

Situation

Transition

Beliefs

Guidelines

Example interrogation process (single interaction)

Example interrogation process (multiple interactions)

Footnotes