EHR Code Sets Recap

ND320 AIHCND C01 L02 A12 Lesson Overview V2

EHR Code Sets Recap Key Points

Awesome! You made it to the end of code sets in EHR.

In this lesson, you learned about key components of EHR data, Code Sets. As a quick reminder, EHR Code sets allow for the comparison of data across various EHR systems.
You first learned about diagnosis codes. In particular, ICD10-CM. You gained knowledge of what this code set is, why it's important, how to read them. You also applied that knowledge by grouping data using EHR codes as well as build an embedding with visualizations. Hopefully, ICD10- CM codes will be easy to work with from now on.

Then you reduced cardinality of a dataset using procedure codes after learning more about ICD10-PCS, CPT and HCPCS. You should be able to break down medication codes too! You should also be able to apply your knowledge to deal with the challenges of working with medication code sets.

Finally, you put all of this code knowledge to work by grouping and categorizing these code sets with CCS. I know that there was a lot of information in this lesson, but now it's time to move on to learning some magic to transform EHR Data! Into what? You'll have to head to the next lesson to find out!

Key Terms

Key Terms	Definition
Medical Encounter	An interaction between a patient and healthcare provider(s) for the purpose of providing healthcare service(s) or assessing the health status of a patient.
ICD10:	International Classification of Diseases 10
WHO:	World Health Organization
ICD10-CM	International Classification of Diseases 10 - Clinical Modification
Primary Diagnosis Code	The code that takes up the most resources to treat.
Principal Diagnosis Code	The diagnosis that is found after hospitalization to be the one that is chiefly responsible.
Secondary Diagnosis Codes:	The other diagnosis codes listed on an encounter.
Medical Procedure	A course of action to achieve an intended result for a patient in healthcare.
Procedure Codes	The categorization of these medical codes during an encounter.
NDC	National Drug Code
Crosswalk	A connection between two different code sets or versions of drugs in the same code set.
RXNorm	Groups medications together.
CCS	Clinical Classifications Software used to map diagnosis or procedure codes from ICD code sets.
MS-DRG	Medicare Severity-Diagnosis Related Group
SNOMED CT	Systematized Nomenclature of Medicine—Clinical Terms

Additional Resources:

Advanced EHR Data Representation: Graph Convolutional Transformer

We have learned about the complexity of information found within code sets such as diagnosis, procedure, and medication codes. However, there is even more benefit in using these code sets together to understand connections between codes within a given encounter and trend across the longitudinal view of a patient’s medical history. You can also compare at a population level to extract useful trends, insights, and even prognostic capabilities.

One paper that I recommend reviewing is the paper that came out of work from Google Health and DeepMind where they create a Graph Convolutional Transformer or GCT.

GCT leverages the now commonly used Transformer Architecture with graph embedding representations of code sets such as diagnosis, procedure, lab, and medication codes.

The advantage of this approach over the standard bag of features approaches is that it retains connections between features such as

diagnosis code for a symptom
the connection to a medication code prescribed
a procedure code that was performed.

It does this by utilizing a conditional probability matrix that you can see in the illustration in figure 3 from the GCT paper. This is definitely worth your time to read through even though it was a bit outside of the scope of this course and should help clarify how these codes are used together in real-world healthcare data.

You can also find the Github Code for the implementation of this paper and can request access to the EHR data if you want to try it yourself.

Links: