Stream: cimi
Topic: Mapping Clinical Finding
Mark Kramer (Jul 27 2017 at 20:20):
How is CIMI planning to map ClinicalFinding? FHIR Observation is supposed to be only simple, atomic results, not complex assessments. More complex assessments are either ClinicalImpression or perhaps DiagnosticReport. If you have one CIMI class mapping to several FHIR classes, there could be a problem.
Grahame Grieve (Jul 27 2017 at 20:34):
it would make life more challenging
Grahame Grieve (Jul 27 2017 at 20:34):
but I'm not sure that this is quite correct:
Grahame Grieve (Jul 27 2017 at 20:34):
"More complex assessments are either ClinicalImpression or perhaps DiagnosticReport"
Grahame Grieve (Jul 27 2017 at 20:35):
i think it depends on what 'assessment' means - it's an ambiguous word that is used for both a process of observation, and a a decision making exercise
Bryn Rhodes (Jul 27 2017 at 22:31):
I'll add that we have use cases that use the "component" element of Observation to model the results of an assessment. There's overlap in that use case with QuestionnaireResponse, but using Observation disconnects the results from what we see as the more "workflow" representation in QuestionnaireReponse.
Grahame Grieve (Jul 27 2017 at 22:43):
right. that disconnect is both good and bad depending on the context in which you are using the data
Mark Kramer (Jul 27 2017 at 23:28):
I have always wondered about the wisdom of Observation.component. It forces slicing. I could see a workable alternative where (for example) Systolic and Diastolic appear as extensions. Was that pattern investigated? What were the pros and cons?
Grahame Grieve (Jul 27 2017 at 23:31):
well, you're going to get slicing either way. and extensions and deeply complex extensions aren't a happy place - extensions aren't a perfect solution by any means, but the more complex they are, the less perfect they are. Initially Observation.component has come and gone as we argued about it
Grahame Grieve (Jul 27 2017 at 23:33):
There's kind of 2 real needs for Observation.component:
- complex tuples as the observation. systolic and diastolic are kind of a special case because people argue about how seperable they are
- machines that generate a set of inter-related data as a cluster
Mark Kramer (Jul 27 2017 at 23:35):
Yes, one could argue BP is a panel, and Systolic and Diastolic are individual observations. But under the assumption they are not, you are stuck slicing. But somehow, multiple extensions doesn't seem like slicing to me.
Also, thanks for the history. There's certainly a need for representing "clusters" but that is neutral in terms of whether it's components or extensions.
Grahame Grieve (Jul 27 2017 at 23:39):
well, it depends how complex the clusters are... own reference ranges, interpretations, error codes. More complexity is not neutral
Grahame Grieve (Jul 27 2017 at 23:39):
indeed, they could be a panel, that's what we did at one stage. but then you have a separability problem.
Mark Kramer (Jul 27 2017 at 23:42):
While I have your attention, what do you think about using Observation for something like an assessment. The example I'm fighting with is a clinician's judgment on disease progression. It is a clinical assessment that takes into account multiple pieces of evidence, and results in a conclusion that either the disease is progressing, or not (oversimplifying here). Could Observation be used?
Grahame Grieve (Jul 27 2017 at 23:46):
PC defined ClinicalImpression for that
Grahame Grieve (Jul 27 2017 at 23:46):
well, no. I need to be carefu
Grahame Grieve (Jul 27 2017 at 23:47):
ClinicalImpression is for the decision making process - what did I consider, what evidence did I look at, why did I eliminate things... now what is our plan
Grahame Grieve (Jul 27 2017 at 23:47):
that's not the same as the decision about whether a disease is progressing or not - I'd take that up with PC
Last updated: Apr 12 2022 at 19:14 UTC
