Create observation abstract class

[wdduncan]

The model has many types of observation classes. E.g.:

• AlcoholExposureObservation
• CancerGradeObservation
• HistologicalCompositionObservation
  etc.

In scenarios like this, there is often an abstract parent class that the children inherit from. Wondering why this isn't the case here ...

From a more pragmatic standpoint, having such a parent class may allow for easier querying of all observations.

[bfurner]

generally "inheritance" is useful in situations where there is need to extend attributes/methods of a base class, and "profiling" is used in situations where there is need to constrain attributes/methods of a base class.

Observation and the various Observation implementations (e.g. CancerGradeObservation) are trying to address a profiling, or constraining, use case. the "base" Observation entity defines a number of different "value" attributes to hold values of different data types (e.g. value_integer, value_string, value_codeable_concept, etc.) but a given implementation (e.g. CancerGradeObservation) only needs to implement the "value" attribute appropriate to its use case. in the case of CancerGradeObservation, this is value_codeable_concept which is typed as a CodeableConcept. in the case of DimensionalObservation, this is value_quantity which is typed as a Quantity. theoretically we could have a base class that defines a single 'value' attribute that can take more than one data type and then inherit from this base class and specify the particular data type at the child level.

happy to talk through this in an upcoming DMH meeting especially to see what is supported by LinkML

[wdduncan]

@bfurner Not sure I follow what you mean by "profiling". Do you have link to a design pattern? From what I could tell there are a common set of attributes that observation of attributes that are shared.

Regarding values please see comment #150.

[wdduncan]

@bfurner I just saw that there is a class Observation. Sorry ...

However, the documentation does not show any linkage between a specific kind of observation (e.g., alcohol exposure observation) and Observation (which is why I missed it).

Is there supposed to be one? By profiling maybe you mean something like composition, but I'm not sure. Or perhaps you mean something like an interface? If so, then having observation as a mixin would be useful.

[bfurner]

@wdduncan see the following from FHIR: https://www.hl7.org/fhir/profiling.html

there currently is no linkage between the generic Observation and the "profiled" subtypes as this functionality isn't supported at this point in LinkML as far as i know.

in FHIR, there is a generic Observation (https://www.hl7.org/fhir/observation.html) that defines a value[x] that can take different value data types that get defined in individual profiles. for example, BodyWeight profiles Observation. https://www.hl7.org/fhir/bodyweight.html

[cmungall]

I've put this on the agenda for the LinkML call, I don't want to give any recommendations here before being fully appraised by others