Categorising EHR Content

Technical (archive)
1

Aniket,

I have refreshed myself on openEHR's clinical model
terminology, but I still think you miss my point.
openEHR has three types of EHR_Entry, namely
OBSERVATION, EVALUATION, INSTRUCTION.

I am using "event" in the natural language context,
rather than the hijacked Event Summary context. My
"event" refers to a non-care event ( predominantly ).
Such "events" do not fall under any of the 3 EHR_Entry
subclasses. Some examples:

a. heart attack
b. bali bombing
c. job redundancy
d. rape
e. snake-bite
f. car accident
g. surgical error

All of these lead to a change of state! I don't see how
any of these could legitimately be classified as any of
OBSERVATION, EVALUATION, INSTRUCTION. Yet,
data describing these "events" should be captured in
the EHR. The data could and should exist independent
of any healthcare action. In some cases, there may not be
any healthcare action following the event. I can conceive
of situations where subjects might wish to record the event
(consider rape) in their EHR, prior to any visit to a
health provider.

The only way current models appear to deal with this is
by lumping them into either OBSERVATION, or, even as
you suggest, INSTRUCTION (represented by ACTION_SPECIFICATION).

I think the problem arises because there is no "higher"
level representation of an event ( either definition ).
This is my point. At the high level we should separate
out state, activities/events that change state, risks.

* state can be categorised as OBSERVATIONS or EVALUATIONS.
* change of state can be categorised as _health_care_actions
  (openEHR's INSTRUCTIONS?) or non_health_care_events.
* risks need their own class.

Clinicians often use OBSERVATIONS and EVALUATIONS to deduce
the event. This is often called diagnosis. A subject might
attend a GP with severe pain and swelling in the hand. The
GP makes some OBSERVATIONS, undertakes a test (issues
an INSTRUCTION) and deduces that an event "funnel-web
spider bite" has occurred. However, sometimes we ( subject,
clinician, other person ) know the event a-priori.
Either way, the event(my definition) is a first class
object in its own right, and should be represented as
such in the EHR.

Perhaps the data pertaining to a non-care event could be
recorded as "OBSERVATIONS", but some of this data,
may not explicitely be OBSERVATIONS of the subject_of_care.
The data qualifies the event, not the subject_of_care.

I hope I have explained myself a little more clearly.

eric

2

Eric

You are into the territory that Computing and Health care have been swimming
in for many years - how to model health care - rather than health care
recording.

All of these are events - but in the record they will cause recordings that
are observations, instructions and evaluations.
a. heart attack
Might start with the patient observation of chest pain...an obseravtion of
ECG... an instruction to order a blood test.. an evaluation of a
differential diagnosis.. the observation of the result of the test.. a
diagnosis.

b. bali bombing
Observation .. was in Kuta and hit by debri ... evaluation .. Very
distressed and requires counselling .. Instruction - referral to counsellor
who is working with such clients.

c. job redundancy

Observation .. made redundant... evaluation , this is a problem that is
worth noting in persistent data.

ETC...

d. rape
e. snake-bite
f. car accident
g. surgical error

We are modelling the health record not health care - attempts to model the
former have been going on for decades and their work is all over the web.

Cheers, Sam

3

Sam,

OK. By extrapolation, then, any documentation of a healthcare event
or non-healthcare event that has occurred in the past, is recorded
as an observation? Any healthcare event that has not
yet occurred, but envisaged, is recorded as an instruction?

When a patient is discharged from hospital, all actions that were
taken on the patient are recorded as observations. e.g. appendectomy ?

This means the semantic knowledge of state vs. change_of_state is
buried pretty deep in the record.

eric

4

Sam Heard wrote:

Eric

You are into the territory that Computing and Health care have been swimming
in for many years - how to model health care - rather than health care
recording.

exactly right. The models we have developed describe in a regular way the concept of "recording" - whcih they have to, because there is no other way for information to be committed to any medium. Thus, a model of recording has to have phenomenologically primacy in any list of models which apply to the information in question. Models of concepts like "real world event", "accident" etc will appear as archetypes.

All of these are events - but in the record they will cause recordings that
are observations, instructions and evaluations.
a. heart attack
Might start with the patient observation of chest pain...an obseravtion of
ECG... an instruction to order a blood test.. an evaluation of a
differential diagnosis.. the observation of the result of the test.. a
diagnosis.

right - in general, there is no way for anyone to say that "X had a major MI" other than via the symptoms reported by X and/or the outwardly observable signs.

b. bali bombing
Observation .. was in Kuta and hit by debri ... evaluation .. Very
distressed and requires counselling .. Instruction - referral to counsellor
who is working with such clients.

yep. And consider: while it would in theory be possible to put something in the EHR indicating the fact of the Bali bombing, this is in fact of now use to patient care - we have to the know the patient's point of view, not just the independently reported fact from the ABC reporter. Were they in the nightclub? Around the corner? Heard the blast (ear damage)? etc Again - we need the patient's account (or that of other relevent person, e.g. patient's friend, or other bystander who knows what happened to te patient) - and this is recorded as OBSERVATIONs whose content include statements by the patient and/or others, and clinical observations.

c. job redundancy

Observation .. made redundant... evaluation , this is a problem that is
worth noting in persistent data.

- similar argument - we need the patient's experience of this, not a news report from The Australian.

It is worth remembering that Acts or Events can be quite easily be modelled using archetypes, and this is the view of information that the GUI user will see. The constructs of OBSERVATION, EVALUATION and INSTRUCTION are very broad categories, and are derived from a philosophical conceptualisation of recording information, i.e. "knowing", also the epistemological categories of knowledge (OBSERVATION = empirical; EVALUATION = a priori ideas; INSTRUCTION = knowledge of how or what to do)

- thomas beale

5

Eric Browne wrote:

Sam,

OK. By extrapolation, then, any documentation of a healthcare event
or non-healthcare event that has occurred in the past, is recorded
as an observation? Any healthcare event that has not
yet occurred, but envisaged, is recorded as an instruction?

When a patient is discharged from hospital, all actions that were
taken on the patient are recorded as observations. e.g. appendectomy ?

This means the semantic knowledge of state vs. change_of_state is
buried pretty deep in the record.

I don't think so - it all depends on how one categorises these things in the archetypes. The categories of OBSERVATION, EVALUATION and INSTRUCTION are not what one sees on the screen - what one sees is due to archetypes. If you decide to create archetypes for OBSERVATION whose purpose is to record certain kinds of events, and certain kinds of states, this can be done. It could even be done in such a way that "event" and "state" were recognised as basic categories.

Plans in general will be recorded as EVALUATIONs, but scheduled admiinstrations, interventions etc will be recorded as INSTRUCTIONs (which are not just the idea of a general plan, but have the intention of being exectuted, and carry sufficient info to be executed by a human or machine)

- thomas beale

6

The events which will be significant for the patient
should be and will be narrated by the patient which
happens in our day to day practice and are recorded in
the Social History or Past history.
I think the events can be recorded in the EHR as
observations under these particular headings.
Comments
Aniket

7

Tom & Sam,

Thanks for taking the time to explain the openEHR use of OBSERVATION,
EVALUATION and INSTRUCTION and how these do not limit the ability
to express state and events in a variety of clinical models. When
one moves from thinking in the healthcare space to thinking in the
recording space it is easy to misinterpret terminology, particularly
in simplistically mapping state to OBSERVATION and healthcare actions
to INSTRUCTION.

I would, however, still like to crusade for the importance of the
notion of non-care event, and its usefulness in future care. I
appreciate it can easily be catered for in archetypes, but would like
to re-stress its importance.

To start with your statement, Tom, regarding the usefulness of recording
the Bali-bombing in a subject's EHR:

yep. And consider: while it would in theory be possible to put something
in the EHR indicating the fact of the Bali bombing, this is in fact of
no use to patient care - we have to the know the patient's point of
view, not just the independently reported fact from the ABC reporter.
Were they in the nightclub? Around the corner? Heard the blast (ear
damage)?

The very knowledge of the event totally transforms the care that is
provided, as you, yourself indicate. I doubt that a person admitted to an
ICU with burns to his/her foot would normally be tested for hearing loss!

The normal course of healthcare is one of deducing the event. From thence
forth, domain knowledge, harnessed from many similar events to other
subjects, is used to guide the course of analysis and treatment. The
analysis and treatment is, of course, modulated by the individual's
symptoms, as you suggest. In fact, this process occurs so frequently that
the first part of it is given a special name - diagnosis. It's just that
diagnosis is usually limited to a subset of the event space (i.e. those
change_of_state events that are taught in medical schools).

Again, consider a 25 year old female who presents at a clinic having
missed 2 successive periods. The GP, having considerable knowledge
of a generalised pregnancy event, suspects, tests for, and diagnoses
pregnancy. The event, and domain knowledge thereof, is more important
than the recording of the observation "missed 2 successive periods".
Now one could view pregnancy as an aggregation of observations. One could
view pregnancy as an evaluation from observations. One could view
pregnancy as an event, about which special data should be stored (
subject's weight, estimated date of conception, HbA1c, etc. ) I think
that there is value in the last of these views, independent of the first
two.

From an epidemiological point of view, it is useful to store non-care

events. In their absence, one could trawl through a population's
set of EHR's and discover a correlation between first degree burns,
hearing loss and trauma. But I am not convinced this would lead to
a clinical guideline for dealing with bomb victims.

I seem to have drawn the discussion away from the topic of this list.
Perhaps I should redirect further discussion to openehr-clinical instead?

Thanks again for your explanations.
regards,
eric

8

Eric Browne wrote:

Tom & Sam,

To start with your statement, Tom, regarding the usefulness of recording
the Bali-bombing in a subject's EHR:

yep. And consider: while it would in theory be possible to put something
in the EHR indicating the fact of the Bali bombing, this is in fact of
no use to patient care - we have to the know the patient's point of
view, not just the independently reported fact from the ABC reporter.
Were they in the nightclub? Around the corner? Heard the blast (ear
damage)?

The very knowledge of the event totally transforms the care that is
provided, as you, yourself indicate. I doubt that a person admitted to an
ICU with burns to his/her foot would normally be tested for hearing loss!

The normal course of healthcare is one of deducing the event. From thence
forth, domain knowledge, harnessed from many similar events to other
subjects, is used to guide the course of analysis and treatment. The
analysis and treatment is, of course, modulated by the individual's
symptoms, as you suggest. In fact, this process occurs so frequently that
the first part of it is given a special name - diagnosis. It's just that
diagnosis is usually limited to a subset of the event space (i.e. those
change_of_state events that are taught in medical schools).

I think you are really campaiging for the general importance of social, human and other real-world events in the health status of the individual, which I cannot disagree with of course. As a non-clinician, I cannot say anything about how such things should be recorded, but I can say that (as far as we know) there is nothing to stop openEHR models of the EHR getting in the way of radically or even paradigmatically different ways of doig medicine (but we still think the activity of "recording" will continue to occur).

Again, consider a 25 year old female who presents at a clinic having
missed 2 successive periods. The GP, having considerable knowledge
of a generalised pregnancy event, suspects, tests for, and diagnoses
pregnancy. The event, and domain knowledge thereof, is more important
than the recording of the observation "missed 2 successive periods".
Now one could view pregnancy as an aggregation of observations. One could
view pregnancy as an evaluation from observations. One could view
pregnancy as an event, about which special data should be stored (
subject's weight, estimated date of conception, HbA1c, etc. ) I think
that there is value in the last of these views, independent of the first
two.

Well, once pregnancy is diagnosed, the most likely date of conception is recorded. If the fact of pregnancy were marked in some way as an IMPORTANT EVENT, any application could use such a marker to make the event clear in the history of things on the screen. I don't see why this could not be done, and maybe it needs some investigation.

From an epidemiological point of view, it is useful to store non-care

events. In their absence, one could trawl through a population's
set of EHR's and discover a correlation between first degree burns,
hearing loss and trauma. But I am not convinced this would lead to
a clinical guideline for dealing with bomb victims.

That's probably true - but on the other hand, even with the fact of Bali bomb reocrded as an event in the EHR, there is no way to guarantee which later symptoms and/or problems are due to it - this is only going to be possible in population studies, in which many instances are used to infer general patterns. But I think human management of the inferencing process is still needed.

I seem to have drawn the discussion away from the topic of this list.
Perhaps I should redirect further discussion to openehr-clinical instead?

that is probbaly the correct group, but don't worry, this discussion is interesting and may indeed lead to some new ideas.

- thomas beale

9

Eric

I have archetyped this to some extent based on health care pathways and
external payments/insurance etc in such circumstances.

I have called it Accident/Injury/Poisoning - this sounds like your
non-health care event. It has information about the insurance company (if
relevant). It does not, at present, have any features not dealt with in
openEHR entry.observation class.

Pregnancy is also modelled as an observation - it is does not include the
antenatal visits - as these are recordings made during pregnancy. The fact
that a (persistent) pregnancy recording might contain links to these (event)
recordings is optional. One day, archetypes might be accepted nationally
that make this mandatory?

The pregnancy observation includes (as I have archetyped it)
Date of LMP
Date of EDD (can be revised as required)
Active (is this pregnancy still active)

(to cope with multiple births)
[Name of child] \ date and time of birth
[Name of child] \ location of birth
[Name of child] \ mode of birth
[Name of child] \ Birth weight
(Then a whole lot on complications....)
Fertility procedure

This may not prove to be the best approach - but it is clearly persistent
information.

Cheers, Sam