Decision Variables Metadata

A missing element of the metadata surrounding decision variables is sources/origins and types. It answers the questions,”What is the nature of the variable?” and “What is the origin of the decision variable data value?” The challenge with this line of inquiry is that there may be multiple valid sources and none is equivalent nor perhaps useful in all cases. For example, “patient presenting with back pain” is fundamentally different from “patient is less than 5 years old” if one wishes to identify where to obtain the information. Along these lines, the decision variable may be vague on purpose, such as “teenage patients” or “elderly adults” where the specific age is not to be obtained, but rather the patient is to be identified as part of a group. To fire a rule, the decision variable must be specified in concrete terms and so further specification of the value is needed.

Sources/Origins

Observations recorded in Clinical notes

Data obtained from the EHR – defines patient State – age, temperature, blood pressure, family history, etc…

As we know types are useful from an ontological perspective.

Types

Temporal – Initial laboratory evaluation of patients presenting with HF

Condition – Patient presents with …

Event – admission to the hospital

Over the next few weeks, we will attempt to put more depth into this preliminary assessment. Once we have defined sets of sources and types, we can apply our machine learning techniques to see if we can assign sources and types dynamically from decision variables instead of defining them manually.

4/5/2020

Our development of decision variable types and sources will remain as a work in progress for now. It is a broad topic that has less definite return compared with other activities so we feel it is best to move on to other topics and circle back at a later date.

Our focus will thus turn to state vectors that can be used to represent guideline recommendations. These will be valuable for recommendation systems that offer additional insight to users based on guideline content. We have worked on these in the past, but further refinement should allow us to achieve greater success in implementing guidelines.

4/18/20