2017-10-12 IA Small Group Call

Date

Attendees

Scribe

Goals

  • Review Keith's email to Joey and Stephanie that poses questions relating to ability to create guidelines for ANF modeling of CDS KNART clinical statements and discuss these same questions during today's call.

Discussion items

TimeItemWhoNotes
60minReview responses to Keith's email inquiry re: ANF modelingGroup
  • Keith:  He started an email string to Joey and both he and Stephanie provided their comments.  In the email, Keith posed an area of question/confusion into a set of examples. Discussion took place regarding that email string, which has a use case of dot blot hemorrhage.  (Reference email dated 10/11/17, 1:13 MT, subj: "Re: Finding vs Observable Entity" addressed to Joey Coyle and cc: Stephanie Klepacki.)
  • Question #1:  Independent of some aspects of constraints on the measurement, there is the phenomenon of a dot blot hemorrhage that can be observed and measured by some means, qualitative and/or quantitative.  Agree?  If not, what is the substance of the disagreement?
    • Joey:  Took some exception to the clinical statement, particularly with documenting the presence of it vs. the characteristics of it.
    • Stephanie:  Agrees with the statement as is.
    • Sarita:  Agrees with Joey.  She would have agreed with the statement as is, but Joey's comments made me reconsider.
    • John:  Need to know who is going to be used the data and how.  He has found that discussions centered on "what something is" to be generally less productive than discussions that focus on or include centrally, "How will this be used, who will use it, and for what?"  For everything we want to talk about (i.e., represent in a clinical record), we want to talk about for a reason.  Having said that, when the scope is very large (human biomedicine vs. dispensing drugs), then it is perhaps less clear how to determine the boundaries of what is helpful and necessary to include in the representation and what is not.
  • Question #2:  Categorizing the phenomenon of a dot blot hemorrhage into one of two disjoint sets (observable entity vs. finding) is problematic at best...Agree?  If not, what is the substance of the disagreement?
    • Joey:  He does not enough data - no proof of a use case where it is problematic.  Is it a concept of the presence, and if yes, it should go in one place.  But for the example you gave yesterday, others might document it in a different manner.  Keith:  Let's say we are creating a decision support rule, and the patient has 1-5 dot blot hemorrhages and we're vague as to what it means in either eye.  For this rule, it would be routine referral to Ophthalmology but it would be a STAT referral if quantity is >5.  We can meet the criterion of 1-5 based on the presence of dot blots.  Is that a finding or an observation?  Joey:  The words used by SNOMED for observable entity and finding are meaningless but those two things you mentioned are (did not capture what he said).
    • Stephanie:  Observable entity, which she believes is synonymous with the term we're using of "Phenomenon."
    • Sarita:  Requested clarification on what "phenomenon measurement" means.  Keith:  We are trying to use the definitions of observation of phenomenon and measurement (see below).  Part of what he wants to point out is that we must reach agreement on these things.
    • John:  Is leery about making distinctions more than necessary.  In other words, the terminology and/or model should only "say what is necessary."  Every distinction has a maintenance cost, and as the model or terminology gets more complex, the marginal cost of each distinction grows (i.e., additional content or elements have to be shoehorned into the existing construct.
    • Definitions
      • Observation of Phenomenon is the active acquisition of information from a primary source.  In living beings, observation employs the senses.  In science, observation can also involve the recording of data via the use of instruments.  The term may also refer to any data collected during the scientific activity.  The human mind, and modern scientific instruments can extensively process "observations" before they are consciously surfaced to the observer.  This unconscious or automated pre-processing of data makes answering the question as to where in the data processing chain "observing" ends and "drawing conclusions" begins difficult.  For our purposes, we do not try to draw a line between "observing" and "drawing conclusions" because for our analysis purposes, the distinction is immaterial.
      • Measurement consists of using observation to compare the phenomenon being observed to a standard (not a normal range).  Measurement asserts something.  These standards can be qualitative, that is, only the absence or presence of a property is noted, or quantitative if a numerical value is attached to the observed phenomenon by counting or measuring.  The standard of comparison can be an artifact, process, or definition which can be duplicated or shared by all observers, if not by direct measurement then by counting the number of aspects or properties of the object that are comparable to the standard.  Measurement reduces an observation to a number which can be recorded, and two observations which result in the same number are equal within the resolution of the process.
  • Keith:  For some of these, we'll need to stop because we don't have clarity, such as for #1 and #2.  Given that we have a phenomenon of dot blot hemorrhage, now we want to apply measurement to it in question #3.
  • Question #3:  There are many aspects of measurement of dot blot hemorrhage that are necessary to know how to interpret the scope and significance of the measurement.  For example, the instrument used and whether cycloplegic administration occurred.  Keith:  When you have a less dilated eye, the results can be more suspect.  Cycloplegic administration could be a precondition, which may be an action performance (i.e., a separate but related clinical statement).  Stephanie:  For this clinical statement of dot blot hemorrhage, she thought all of the information in question #3 would be part of the clinical statement itself and not another type of separate, clinical statement, such as action performance.  Keith:  That is what we are working to define and document.  There are some things that still need defining, such as a precondition.  You could have observations with actions and there could be a certain amount of recursion but it could provide clarity as well.  What are your thoughts?  Are you getting a stronger sense of where we need to go?
    • Joey:  This is a slippery slope of including the world into one model.  Keith:  We can't do that.  Joey:  I could state everything about the patient at the moment I took the measurement but some things are very important, such as taking a BP during a stress test vs. taking it under normal conditions.  (Missed part of what he said but it related to preconditions.)  Keith:  We want to provide general patterns.  Is it important for someone to record the administration of a cycloplegic for a dot blot hemorrhage?  The discussion about whether we record everything about the patient as a precondition is not our task. Rather, our task is to say if you want to record a precondition, here is how you do it. We want to keep this as 4 general patterns (i.e. models).  The question is can we represent them.  Joey: That helps.  It is doable, and it doesn't mandate that they record it.  Keith:  For BP, there is the issue of ensuring the patient doesn't have a full bladder when it is taken and how do we represent it.  If someone wants to record an action as a precondition, then how is it done?  For the issue of having the result as part of the topic, if you do that, you can't use the topic for an action request/performance.  The principle of containment over inheritance pushes the issue of keeping them separate because for one class, you would have a topic that includes the result and another that does not contain the result.
    • Stephanie:  Her understanding has not improved.  The confusion arises when we have a clinical statement within a clinical statement, such as a phenomenon measurement that has an action request.  Keith: There are different ways to reference clinical statements that are related.
    • Sarita:  She learns best by understanding the big picture.  The way things are outlined in Kirsten's topic/circumstance document are at a high level for the various types of action requests and what part of them would be included in the topic vs. circumstances and it shows the various types of circumstances needed for the various types of orders (such as route of administration for a medication order).  We are working on the medication, laboratory, and radiology orders from the Chest Pain/CAD order set KNARTs.  Claude put them into the action request models and we are using the circumstances that have been identified, and it seems to be working well.  Can we take this for dot blot hemorrhage and see if they will hold up?  For action requests, we did one issue for precondition.  We can define what is missing or what needs revision.  That is the best way for her to work through this. Keith:  When you go through these and make decisions, do you describe the principles you used to make them?  Sarita:  Yes, she had said we need to write the heuristics as to what is determining goes into these fields.  The need to write up the heuristics for how each field in the model is completed was something that Kirsten, Sarita, and others from the Deloitte team discussed yesterday and will work on this going forward.  Keith:  Getting those heuristics documented and comparing to the principles is an important aspect of where we need to go.
    • John:  The metaphor he sees for this is that we are trying to devise a signature, a set of arguments, for a procedure.  He doesn't know that we can find a single set of arguments that will match everything we want to measure because they each have their own things we care about.  It seems very challenging to him.  Keith:  We seem to be having anxiety at different levels.  In the end, he agrees with what you are saying but it's not related to what is asking for.  For these 4 models, we are trying to provide a general pattern that for ~95% of the things we need to do, we can do them.  For the remaining 5%, we'll see if the current model works and if not, figure out how to revise it.  For example, saying there are certain kinds of preconditions that themselves are phenomenon measurements is part of a general pattern for certain things, except for something like cycloplegic administration.  There are certain preconditions that are actions and certain ones that measurements.  John:  It sounds like we may have a model.  The representation of a measurement should include "findings" and "performed procedures" as preconditions; what else could be a precondition?
    • Keith:  Do we review the use cases and try to come up with a general set of patterns?  That is different from what we were doing for constraints.  Part of the struggle is he is trying to help us pull together a productive process for a model.  Can you productively work on refining a model like this or is this a hopeless cause and he just needs to go away and propose a model and bring it back to the group?  He wants this group to come up with these things and document and contribute vs. just reacting. The group needs to construct.  Sarita:  She would need to have full time availability to do that and it's difficult just to give a fraction of time to think about something this big, it's daunting.  We are trying to use the 4 models we have and test them against the initial KNART content.  That is a good start and will help us refine the models. But again, it's difficult to do this with a limited time constraint.  Keith:  Agrees.  He would like to propose that Joey make time next week to work on this.  There are 2 small group meetings and Stephanie will handle the logistics.  He would like for Joey to work on:
      • Better articulating the model and use the dialogue in the email string yesterday between Keith, Joey, and Stephanie
      • Using the nitroglycerin use case and identify those parts that are themselves clinical statements and identify what else is needed
    • Joey:  Yes, he can work on this.  He will also resume work on what he started previously.  There are 2 disjoint sets and they are really questions and answers.  If we look at:  1) dot blot hemorrhages and 2) the number of those per quadrant, then clearly one of those is a question (i.e., the first one).  Number of dot blot hemorrhages is an answer and would be documented as a finding.  Keith:  He would like for Joey to write this up next week and we can review as a group. But, it needs to be better articulated so he can understand it better.

Action items

  • Joey Coyle:  For next week, work on better articulating the model and use the dialogue in the email string yesterday; use the nitroglycerin use case and identify those parts that are themselves clinical statements and identify what else is needed.
  • Stephanie Klepacki:  Schedule calls for next week.