ICD-O — coding oncology

In one line

ICD-O — the International Classification of Diseases for Oncology — codes a tumour along two independent axes at once: where it is (topography) and what it is (morphology). Cancer is the one place where a single diagnosis code is structurally not enough.

Why one code fails for cancer

Ask ICD-10 about a cancer and it gives you roughly "malignant neoplasm of the stomach." That's adequate for a mortality statistic and useless for cancer science, because it silently merges cases that behave nothing alike.

The problem: site and cell type vary independently. A lymphoma can arise in the stomach. An adenocarcinoma can arise almost anywhere. Knowing only the organ tells you little about prognosis or treatment; knowing only the cell type tells you little about presentation or surgery. You need both, and you need them separable, because researchers ask questions along each axis on its own: all gastric tumours, or all lymphomas regardless of site.

Collapse both facts into one code and you can never take them apart again.

The two axes

Topography (C codes) — anatomical site. Deliberately aligned with ICD-10's neoplasm chapter, so C16.0 is the gastric cardia in both.

Morphology (M codes) — the histology, plus behaviour and grade:

M-8140/3
  │      └── behaviour  — /0 benign  /1 uncertain  /2 in situ  /3 malignant, primary
  └───────── histology  — 8140 = adenocarcinoma

The behaviour digit is the one to respect. /2 versus /3 — carcinoma in situ versus invasive — is not a shade of meaning. It's the difference between a lesion that has not breached the basement membrane and one that has. Two characters apart in the code; entire worlds apart clinically.

So a stomach adenocarcinoma is C16.9 + M-8140/3: two codes, one tumour, each answerable alone.

Where it's actually used

ICD-O is the working language of cancer registries, and it's maintained under WHO/IARC. Registries are the reason we know anything durable about cancer incidence, survival and trends — and they only work because every registrar codes the same tumour the same way.

This is the thing to grasp about ICD-O's purpose: it's not a billing instrument. Nobody adjudicates a claim on an M code. It exists so that a gastric adenocarcinoma registered in Chennai in 2026 is comparable with one registered in Osaka in 1998. Its whole value is longitudinal and international comparability — which is precisely why registries are so strict about coding rules that look pedantic from outside.

How it relates to the neighbours

• ICD-10/ICD-10-CM — one code, for statistics and claims. ICD-O topography deliberately mirrors its neoplasm chapter so cases can be rolled up.
• ICD-11 — brings much of this thinking into the main classification through post-coordination, letting site and histology be combined natively. A conceptual convergence, not a replacement of registry practice.
• SNOMED CT — can express the same clinical reality via post-coordinated expressions. Registries nonetheless run on ICD-O, because decades of comparable data are already in it and continuity beats elegance.

The lesson worth taking away

ICD-O is the clearest proof that the shape of a code system follows the question it was built to answer. Cancer research needed to slice by site and by cell type independently, so the classification grew two axes. When you meet a code system that looks strangely structured, ask what question it was designed for — the structure is almost always the answer.