There’s a Second Language Underneath Your ICD-10 Codes. Almost No Coder Speaks It.
Why certified coders are quietly moving into informatics — and getting paid more
You spent months learning ICD-10-CM. You learned CPT® conventions, modifier logic, sequencing rules, the difference between a principal diagnosis and a principal procedure. You passed an exam that most people fail on the first attempt.
Then you started working, and you noticed something nobody mentioned in training.
The problem list in the EHR doesn’t behave like free text. It behaves like a database. Diagnoses appear in it with a suspicious uniformity — the same phrasing, the same structure, chart after chart, physician after physician.
That’s not the physician’s writing. That’s SNOMED CT.
And the fact that almost no working coder recognizes it is, right now, one of the largest unclaimed career opportunities in this profession.
Two vocabularies doing two different jobs
Here’s the distinction that unlocks everything else.
ICD-10-CM was built to classify. It takes the enormous variation of human illness and sorts it into manageable groups — for billing, for statistics, for reimbursement. Classification requires collapsing detail. Twenty clinically different presentations may share one code, and for a payer deciding what to pay, that’s exactly correct.
SNOMED CT was built to describe. It’s a clinical terminology — over 350,000 concepts, each with a defined meaning and defined relationships to other concepts. It doesn’t group things in order to pay for them. It represents them in order to be precise.
Take a real example.
A physician documents: acute exacerbation of COPD with acute respiratory failure requiring non-invasive ventilation.
In ICD-10-CM, that becomes a handful of codes designed to land the encounter in the right DRG.
In SNOMED CT, every clinical element exists separately and is machine-readable — the exacerbation, the underlying disease, the respiratory failure, the ventilatory support, and crucially, how they relate to each other. SNOMED knows that a concept is a kind of something else, that it involves a particular body site, that it was caused by a specific agent.
One is a classification of the encounter.
The other is a representation of the patient.
You already work with the first. You’ve been working near the second for years without being told.
Three things changed while you weren’t looking
For most of this profession’s history, SNOMED lived somewhere coders never went — deep inside the EHR, underneath problem lists and clinical documentation, invisible.
Three developments dragged it up to your desk.
The record became structured. Modern EHRs increasingly capture problem lists, allergies, and clinical findings as coded SNOMED concepts rather than as prose. When you read a structured problem list today, you are frequently reading SNOMED output. Nobody told you, because whoever configured the system assumed you didn’t need to know.
Automated coding runs on it. This is the one that should get your attention.
Computer-assisted coding and autonomous coding engines don’t read a chart the way you do. Many of them work by extracting clinical concepts, mapping those concepts to SNOMED, and then mapping SNOMED to ICD-10-CM.
So when the machine produces a wrong code, the failure often isn’t in the ICD assignment at all. It’s upstream — in concept extraction, or in the map.
Think about what that means. A coder who doesn’t know the SNOMED layer exists cannot diagnose that error. They can only observe that the machine was wrong, correct that one chart, and move on — while the identical fault silently reproduces across ten thousand other charts.
A coder who does understand the layer finds the systemic cause. That’s not a different skill level. That’s a different job title.
Research now uses your data directly. And this is the part almost nobody in coding has absorbed yet.
The research problem hiding in your work
Clinical research used to depend on data collected specifically for research — trial forms, registry abstraction, chart review by trained staff.
Get Medesun Medical Coding Academy’s stories in your inbox
Join Medium for free to get updates from this writer.
Increasingly, it doesn’t. Real-world evidence studies, comparative effectiveness research, pharmacovigilance, population health analytics, and AI training sets now draw straight from routinely collected clinical data.
The record is the dataset.
That creates a problem every coder should understand: administrative codes were never designed to answer clinical questions.
A researcher asking which patients had heart failure with reduced ejection fraction, or a specific tumor histology at a specific stage, or a particular adverse drug reaction, is asking something billing codes often can’t answer precisely. The classification collapsed exactly the detail the research question depends on.
SNOMED can carry that detail. Its hierarchy means a query for bacterial pneumonia automatically retrieves every subtype beneath it. Its relationships let a cohort be defined by clinical meaning rather than by billing category.
But here’s the part that should make you sit up:
SNOMED is only as accurate as the documentation and the mapping beneath it.
If a problem list carries a diagnosis nobody has evaluated in four years, SNOMED records it faithfully. If the documentation is ambiguous, the extracted concept is ambiguous. If a map is applied without judgment, the error propagates in both directions — into the claim, and into the research dataset.
The same discipline that protects a claim protects a study.
That’s your discipline. You already have it. You just haven’t been shown where else it applies.
Why you specifically are positioned for this
You sit at the exact point where clinical language becomes structured data. Very few people in any healthcare organization occupy that position, and almost none of them have your training.
The clinician knows medicine but rarely thinks about how their words become data.
The data scientist knows the dataset but has usually never read a discharge summary, and has no idea why a problem list can’t be trusted at face value.
You know both.
That’s not a small advantage. It’s the entire basis of a second career.
What to actually do about it
Start reading problem lists as SNOMED output, not as truth. A concept sitting in a structured problem list is a claim about the patient — possibly entered years ago, possibly by a template. Your MEAT discipline applies at full strength.
Treat every map as a judgment, never a lookup. SNOMED-to-ICD maps exist and are carefully maintained, but a map cannot know the clinical context of an encounter. Where it offers options, a human decides. Make sure that human is you.
Audit automation at the concept layer. Don’t just ask whether the ICD code is right. Ask whether the underlying concept was extracted correctly. A pattern of wrong codes almost always traces to one bad concept or one bad map — and that’s a systemic finding worth a hundred single-chart corrections.
Learn the structure, not the content. Nobody memorizes 350,000 concepts. You need to understand what SNOMED is for, how hierarchies and relationships work, where maps are fragile, and why your accuracy now travels far beyond the claim.
Start with the SNOMED International browser. It’s free. Spend an hour looking up conditions you code every week and studying how they’re represented. That hour will reframe how you see your own work.
The honest part
I’m not going to tell you this is easy money or that a weekend course makes you a terminologist. It doesn’t.
But the roles are real — terminology specialist, clinical informaticist, data quality lead, coding informatics consultant — and the supply of people who genuinely understand both clinical terminology and administrative classification is very small. Small enough that organizations regularly can’t fill these positions.
Meanwhile, entry-level coding is getting more crowded and more automated every year.
There’s a rule worth remembering here: when a machine takes over the easy 60% of a job, the people who thrive aren’t the ones who compete on the remaining 40%. They’re the ones who move to the layer the machine depends on.
For coders, that layer has a name.
Every good coder eventually learns that a code is never just a code.
SNOMED CT is where that stops being a philosophy and becomes literal. Your work is no longer only the bill.
It’s the evidence.
