Skip to main content

Sep 18 2026

Idea here is to understand the data. Ask yourself

  • Who/What produced the data?
  • Why?
  • Under what conditions?
  • With what transformations?

Understand the ‘rhythms’ and what missingness means1. The data is not separate from the task you’re going to do.

  • EHR? You’re only observed when you’re in the system. It’s all geared towards billing because having a humane health system that prioritizes people over profits is communist (my observations, not professor’s.)
  • Self-Reported? It’s irregular.
  • Wearables? Pretty dense but quite a few gaps.

Random​

  • HL7 (Health Level Seven) has a “seven” that maps to the OSI model (1987).
  • Columbia P&S was one of the first to adopt an EHR
    • DBMI was a formal department in 1994 and is the second oldest (oldest is University of Utah all the way back in 1972!)
  • SNOMED is for patient care documentation vs ICD10 is for billing (and epi.)
    • Former is this elaborate clinical terminology
    • Latter is a statistical classification system

This is a duhburger but always ask “Which system” when you see a code (“Code Identity” = System + Code). Not everything can be captured by the coding systems (discussed five during the lectures.) Notes can preserve uncertainty and can be a treasure trove of information.

Aligning syntax and aligning meaning are two different things!

LOINC is fascinating. You’ll see a code like 4548-4. That last digit is a checksum (! Mod10/Luhn). That entire thing points at a longer “Fully-Specified Name”:

4548-4 <---> Hemoglobin A1c/Hemoglobin.total:MFr:Pt:Bld:Qn:

FHIR is about “exchangeable units”: you don’t get an entire patient with a FHIR call! It specifies “Resources”

In your app, don’t just say “You have hypertension”. Think of the data provenance questions above. What is a better phrasing? “Your records show a condition coded as ‘Hypertensive Disorder’ with onset recorded on Jun 17 2018.”

Questions​

  • What makes ICD10 a “statistical” classification system?
  • When does an institution decide to ‘upgrade’ from ICD10 → 11?
  • What if there’s stuff missing in the codes?
  • Is FHIR basically “a RESTful specifiction for health data”? Looks like it.
    • How orthodox is it? How are allowed deviations specified (if at all)?
  • Does FHIR actually capture a diagnosis?

Footnotes​

  1. The ‘best’ thing to start with is to do nothing until you have more evidence. ↩