Modeling CGM metrics (TIR, GMI, Mean Glucose) without using Laboratory Analyte CLUSTERs

Clinical
1

Hi everyone,

My colleague @david.hernandez and I are currently working on the modelling for a remote glucose monitoring initiative. We need to represent several calculated metrics derived from Continuous Glucose Monitoring (CGM) data, such as:

  • Time in Range (TIR), TAR, and TBR (expressed as percentages).

  • Glucose Management Indicator (GMI).

  • Glycemic Variability (Coefficient of Variation and SD).

  • Mean Glucose.

In many existing templates, I see these metrics modeled using the CLUSTER.laboratory_test_analyte.v1. However, we feel this is semantically inaccurate since these are sensor-derived calculations and not “analytes” in the clinical laboratory sense (they are not processed from a physical sample).

I am looking for best practices or examples on how to model these “summary” metrics. Specifically:

  1. Do you recommend using a specialized CLUSTER for this purpose? Or perhaps a specific OBSERVATION for CGM Summaries?

  2. How do you handle the math functions? For metrics like “Mean Glucose”, would you use an interval event with the mean math function, or represent it as a single point-in-time EVALUATION / OBSERVATION that captures the result of a period’s calculation?

If we use an Interval Event at the OBSERVATION level to indicate that the data covers a 14-day period, we run into the following issues:

  • Analyte Name: Which name should we use for this element, since it is mandatory?

  • Math Function: The mathematical function (mean, variation, etc.) would be defined in the math_function attribute of the OBSERVATION Event, not within the CLUSTER itself.

In conclusion, is there any recommended ‘Summary’ or ‘Device-derived statistics’ CLUSTER that you would suggest as a best practice? Or perhaps a specific pattern to handle these metrics without forcing them into a laboratory-centric structure?

Thanks in advance for your insights!

3 Likes
2

This is a really good question.

It certainly merits a specific archetype . I would go for a Cluster as I think it then sits more comfortably inside the typical lab test hierarchy and accept that what we are recording is a summary of raw device data, and therefore somewhat flattened i.e do not try to adhere too closely to interval events etc.

I expect dissent!

2 Likes
3

I agree (crazy, I know! :winking_face_with_tongue:)

If we’re getting these data from continuous glucose monitoring, ie we get the mean etc every time we get a new point in time glucose level from the device, it doesn’t make sense to record it as an explicit interval event with the mean math function. If we were calculating the mean glucose outside of the device based on a series of data from the device, I’d be more inclined to use the mean interval event method.

3 Likes