How to define transitions in the ISM

Clinical (archive)
1

Hi all,

I’m testing the AE for a new workshop, and designed a simple state machine for and order so my students can use it as basic for more complex state machines.

I have: NEW (maps to ISM PLANNED), ASSIGNED (maps to ISM PLANNED), STARTED (maps to ISM ACTIVE) and FINISHED (maps to ISM COMPLETED).

What the AE is not allowing is to specify the ISM_TRANSITION.transition : DV_CODED_TEXT.

The problem is if I have two states mapped to ASSIGNED, how a software knows which one is the state to activate if the transition “initiate” is not define. Also I want to specify that from new should happen a “plan_step” transition to change the state to ASSIGNED. Seems we are missing important metadata in the archetype.

How do clinical modelers solve those problems?

Will test LinkEHR to see how they define the ISM and the valid transitions.

Thanks,
Pablo.

2

Hi Pablo!

I’ll try to answer your question about how clinical modellers solve this problem. Have a look at the ACTION.medication archetype (http://openehr.org/ckm/#showArchetype_1013.1.123). This archetype has 11 separate steps for the ACTIVE state. In each medication management context, one or more of these will be relevant, and often in a way or order that’s not possible to predict. We therefore “solve” the problem by leaving it to the business logic of the application. This may be frustrating for the implementers (I don’t know, is it?), but it makes our work manageable. Designing ACTION archetypes is complex in the first place, and I’m not sure we’d get any published if we needed to map out all possible combinations and orders of pathway steps too.

Regards,

Silje

3

Hi Silje,

I got the issue with complex workflows.

With the current solution you’ll need to provide more metadata to the developers so they can implement the correct workflows, like possible or impossible transitions from one state to another, because constraints are not in the archetype.

On the other hand, simple workflows can be completely specified in the archetype without providing extra medadata separately from the archetype, since both states and possible transitions can be specified there, like the little toy state machine on my previous message. The issue is the AE doesn’t allow to express constraints for the ISM_TRANSITION.transition (DV_CODED_TEXT) attribute (a constraint that can explicitly state a list of valid transitions to reach that state, I think “transition” is about inbound transitions not outbound, but that is a separate issue). I’ll test if this can be done using LinkEHR.

Also for complex flows, it would be good to provide the possible transitions, even if the list of possibilities is big, this is just to make the archetype contain all the metadata needed for implementation, without the need of providing that externally to the archetype. I know this requires more work in the archetype, but it might be less work in total, since the problem will need to be solved as you said, in the business logic. IMO this approach does not add more constraints to the archetype, just more information, and made the implicit freedom of transitions explicit.

Maybe this should be considered case by case, and modeling tools should allow to constraint the transition, but leave that to the modeler. I think a good approach is to constraint what can be constrained, for instance on the medication archetype there are a lot of transitions between active states, but maybe there are less transitions between other states, and those can be in the archetype. This would remove a little friction at development time.

It would be nice to know how other modelers do this and how other implementers deal with non defined transitions in ACTION archetypes.

Best,
Pablo.

4

Hi Pablo,

Every archetype ideally needs to be designed for the maximal dataset and universal use case. The ACTION archetypes are no different.

But you have picked up on a major gap in our tooling at present – the modellers need the ability to be able to constrain the ACTION archetypes in templates for each use case:

  • to show what data points are relevant for each pathway step, and
  • which steps are relevant to our use case.

It is also not currently possible for modellers to record the proposed workflow or transitions in any template at present. This is another major gap and, in practice, is usually managed on a project by project basis a negotiated by the parties involved – verbally, word docs etc.

This is a relatively unexplored area where we need more tooling and/or standardised processes to communicate the requirements of the clinicians and intent of the modellers to the software engineers implementing systems.

No silver bullet here, yet. But open to collaborate with anyone who has suggestions…

Regards

Heather

5

Hi Heather, thanks for the insight.

It seems this is a well known issue for clinical modelers.

I certainly agree with the criteria of the maximal dataset, IMO what makes a maximal dataset depends on the modelers interpretation of each specific use case. Of course less constraints allow a greater set of use cases to be considered, but also increases the work that needs to be done to fill the blanks between a generic archetype and a specific State Machine to be implemented. That negotiation that you mention is what I described as “extra metadata needs to be given for implementation”.

In terms of the gap in modeling tools, I agree, technically archetype editors and template designers (Ocean and others) should be able to constraint the valid or invalid transitions. Then if modelers use or not those constraints, should depend on their criteria, not on limitations of modeling tools. On the other hand, this issue of modeling tools not supporting these constraints might be because in the RM, ISM_TRANSITION is not LOCATABLE (all classes that can be archetyped), but inherits from PATHABLE (RM 1.0.2 & 1.0.3). Considering this, it is a little inconsistent that the AE allows to create constraints for ACTION.ism_transition, but only for the ISM_TRANSITION.current_state and ISM_TRANSITION.careflow_step. but not ISM_TRANSITION.transition.

Maybe a solution form the RM is to make ISM_TRANSITION inherit from LOCATABLE, then update modeling tools to support it.

I will mention this to the SEC.

Best,
Pablo.

6

Hi all

Somewhat late reply due to vacation…

We have come across that same problem and for us it actually was a show stopper for which we had to invent a work around.

First a remark about the tools:

We saw that ArchetypeEditor did not add the transition. So we tried to add I manually to the adl-file. We found however that AE ignores it and after saving again from AE it is gone. Further we tried to use the modified adl in a template using Template Designer, but it was ignored and no trace of it in the resulting opt.

These are very serious limitations in the tools and forces a work around that we should very much like to abandon (see below). It raises the question how the community should go forward to make sure there are appropriate tools. Who owns the tools? Who pays for their maintenance?

The ISM is potentially a very powerful asset of openEHR. Missing the transition property mutilates it to very limited value.

Then a remark to Silje’s reply:

“Solving” the problem in the business logic is only possible when recoding after the fact. Given that the current state is so and so and the new state is so and so we can deduce (in most cases) the transition.

Our problem:

Our problem is the opposite of solving after the fact. We want to present to the user only the transitions valid at any moment in time. Given the ISM and completely defined ISM_TRANSITIONs this would be possible and easy. But not so without the transition! Without that information it is not possible to distinguish the transitions having the same current state.

To see the problem, assume a simple state machine with one of each of these transitions: active_step, suspend and resume. Let the current state be SUSPENDED (last recorded action was suspend). In this state we only want to give the user the option to resume. However, without the transition property in the ISM_TRANSITION we cannot distinguish resume from active_step. Both have ACTIVE as their current step and careflow_step is only descriptive and not usable. The only option is to give the user all choices and assume he does the right thing. Not a good option. After all, resuming a suspended drug and administering the drug are quite different things and we do not want an erroneous administering to take place as result of our system suggesting it!

Our work around:

Fortunately each ISM_TRANSITION has a unique id. Based on this id we add the missing transition, from our own local configuration, to the archetypes we use after having loaded them. This information is transient and only lives in our memory instances of the archetype. But at least we have it available so that we can make a full state machine evaluation and find only the relevant transitions to present to the user.

Some questions:

What if the user inadvertently administers a drug that has been suspended? In that case he surely needs to have this transition anyway, doesn’t he? Well, yes, but not as a suggestion from our system! This situation must be handled separately from guiding the user through the states. In fact, it could be argued that this be recorded as an ad hoc action.

With regards,
Ivar Yrke

Senior systemutvikler

DIPS AS
Telephone +47 75 59 24 06

Mobile +47 90 78 89 33

7

Hello,

In principle, it is possible to edit the transition part of Action archetypes in LinkEHR by using the Archetype Tree view.

Nothing fancy but should do the trick (at least it should be better than editing them by hand :slight_smile:

If there is interest we could improve the openEHR specific editor to support the definition of all ISM_TRANSITIONS constraints in a single dialog

Regards

8

Thanks Ivar, that was a great explanation (much better than mine) and describes the same issues I’m having with the legacy modeling tools, and how that affects development.

I’m also looking into what Diego suggest, migrating the legacy modeling tools to LinkEHR.

(attachments)

imagen.png

9

Hi all

Somewhat late reply due to vacation…

We have come across that same problem and for us it actually was a show stopper for which we had to invent a work around.

First a remark about the tools:

We saw that ArchetypeEditor did not add the transition. So we tried to add I manually to the adl-file. We found however that AE ignores it and after saving again from AE it is gone. Further we tried to use the modified adl in a template using Template Designer, but it was ignored and no trace of it in the resulting opt.

These are very serious limitations in the tools and forces a work around that we should very much like to abandon (see below). It raises the question how the community should go forward to make sure there are appropriate tools. Who owns the tools? Who pays for their maintenance?

The ISM is potentially a very powerful asset of openEHR. Missing the transition property mutilates it to very limited value.

Then a remark to Silje’s reply:

“Solving” the problem in the business logic is only possible when recoding after the fact. Given that the current state is so and so and the new state is so and so we can deduce (in most cases) the transition.

Our problem:

Our problem is the opposite of solving after the fact. We want to present to the user only the transitions valid at any moment in time. Given the ISM and completely defined ISM_TRANSITIONs this would be possible and easy. But not so without the transition! Without that information it is not possible to distinguish the transitions having the same current state.

To see the problem, assume a simple state machine with one of each of these transitions: active_step, suspend and resume. Let the current state be SUSPENDED (last recorded action was suspend). In this state we only want to give the user the option to resume. However, without the transition property in the ISM_TRANSITION we cannot distinguish resume from active_step. Both have ACTIVE as their current step and careflow_step is only descriptive and not usable. The only option is to give the user all choices and assume he does the right thing. Not a good option. After all, resuming a suspended drug and administering the drug are quite different things and we do not want an erroneous administering to take place as result of our system suggesting it!

Our work around:

Fortunately each ISM_TRANSITION has a unique id. Based on this id we add the missing transition, from our own local configuration, to the archetypes we use after having loaded them. This information is transient and only lives in our memory instances of the archetype. But at least we have it available so that we can make a full state machine evaluation and find only the relevant transitions to present to the user.

Some questions:

What if the user inadvertently administers a drug that has been suspended? In that case he surely needs to have this transition anyway, doesn’t he? Well, yes, but not as a suggestion from our system! This situation must be handled separately from guiding the user through the states. In fact, it could be argued that this be recorded as an ad hoc action.

With regards,
Ivar Yrke

Senior systemutvikler

DIPS AS
Telephone +47 75 59 24 06

Mobile +47 90 78 89 33

10

Hi Peter, thanks for your reply. It adds several relevant facts and background to the problem description.

We did in fact have a copy of the AE with transistions, but we could not figure out how to use it. We do not need to go beyond the RM, we only need to fully specify according to RM. If memory serves me right I think that implementation did not help us with that. In fact, I think the whole implementation/visualization if pathway in AE is part of the problem/limitation. It kind of contains a left-to-right idea, which isn’t really reflection the dynamics of the ISM.

I actually did look into the code. After some struggling into the VB code (which isn’t my strong side) I eventually found that the problem also went into the underlying java-classes (which is not my strong side either). I concluded this was not an easy fix, which I had hoped, and basically gave up.

But you are absolutely right, the rest of the tool stack is just as important. This problem really needs support from the community and is desperately needed for serious use of the ISM.

11

Hi Ivar,

12

@Peter thanks for the feedback!

As Diego mentioned, I think currently the only tool that support full specification of constraints for the ISM is LinkEHR, I need to test it! And since they added OPT support, transitions might get exported as well.

I also have the VB code, but I’m a little allergic to VB :slight_smile:

Best,
Pablo.

13

Yeah, in LinkEHR you are able to constraint any part of the RM, and since the export OPT feature doesn’t really remove any constraint probably it should work for you.
In fact, I’m currently working on giving support to the creation of ISM_TRANSITIONS in LinkEHR openEHR specific editor to ease the creation of these constraints, I’ll have news about it very soon.

14

Pablo, it would be great to hear how to get on with LinkEHR.

To me it seems very «technical», exposing all the nitty-gritty details and requiring knowledge of such from the user. So I can’t see how we can force everyone to use that tool rather than AE, resulting in loss of information. So even though there might be a tool that works all the way through we really have a tool issue.

Vennlig hilsen

Ivar Yrke

Senior systemutvikler

DIPS AS
Telefon +47 75 59 24 06

Mobil +47 90 78 89 33

15

That’s why specific rm editors are also provided with LinkEHR, to hide that complexity. I’m adding support to these new classes to the openEHR editor as we speak :slight_smile:

16

The nice thing is artifacts are standard, so you can use AE, and when you find a limitation, just load the ADL on LinkEHR to add the tiny bits, also can export OPT from there. Until we have a perfect tool, we need to use what is there (and free). Also use and feedback make tools to evolve, and currently the more active dev effort is on LinkEHR. My idea is to completely switch all ADL and template design to LinkEHR in the short term.

17

Excellent if we can get there. My concern is that someone opens my meticulously crafted (using LinkEHR) ACTION with transitions in AE, to fix say a translation error, removing all transition info when they save. Hopefully these new tools become good enough to be accepted by all users.

Vennlig hilsen

Ivar Yrke

Senior systemutvikler

DIPS AS
Telefon +47 75 59 24 06

Mobil +47 90 78 89 33

18

Well, LinkEHR has been around for a while now :slight_smile:
We have implemented an “export to archetype editor” functionality that limits the ADL syntax to what Archetype Editor & Template designer can understand, so I’m pretty sure you won’t have problems in that area.

In regards of OPT, exported OPTs are indistinguishable from the ones created with TD. For example, if you import a template into LinkEHR and export it right away OPT is exactly the same.

Regards

19

Hi Ivar,

I believe that should be handled by an internal management process, defining who can update an archetype and which tool should be used for that.

Also my comment was considering the current situation where we are “in transition” from a tool set to another.

As a parallel though, and trying to avoid what happened with the AE/TD, I think we need to find a sustainable model for modeling tools, so we are sure someone is maintaining them, and we can reach out to someone for bug reports and change requests. One way would be through donations, but that is too variable. Another way would be to buy support, or event getting some support from the Foundation, to be able to keep up to date modeling tools (this could be proposed from the Software Program). I like this idea because the Foundation can put a % of the memberships into the Software Program, to allow free access from the community to modeling tools and other tools that might be useful for implementation. These are just ideas to avoid future problems that we are having right now.

20

this is not a legal thing to do! If Assigned is a ‘careflow step’, its execution in the real world has to result in the ISM state machine being advanced to one defined state. So there is a problem from the outset with this discussion… - thomas