---
license: cc-by-4.0
language:
- de
task_categories:
- text-classification
- text-retrieval
tags:
- medical
- icd-10
- icd-10-gm
- clinical-coding
- german
size_categories:
- 10K<n<100K
configs:
- config_name: default
  data_files:
  - split: train
    path: icd10.jsonl
---

# ICD-10-GM Codes (German)

Complete hierarchy of **ICD-10-GM** diagnosis codes with their German
descriptions, scraped from the official German National Health Portal
[gesund.bund.de](https://gesund.bund.de/icd-code-suche).

**11072 codes** across the full classification.

## Fields

| field | description |
|-------|-------------|
| `code` | ICD-10-GM code, e.g. `A22.7` |
| `description` | official German label, e.g. `Milzbrandsepsis` |
| `body_text` | plain-language German explanation of the code, written for laypeople by the "Was hab' ich?" gGmbH; empty string when the portal has no explanation for that code |
| `text_variations` | alternative spellings — the dot-less compact form, e.g. `["A227"]` (empty list when identical to `code`) |
| `parent_code` | direct parent code in the hierarchy (`null` for top-level categories) |
| `parent_slug` | parent URL slug (`null` for top-level categories) |
| `slug` | URL slug on gesund.bund.de, e.g. `a22-7` |
| `depth` | hierarchy depth (1 = three-char category such as `A22`) |
| `url` | canonical source page |

Distribution: depth 1: 1757, depth 2: 9315. **8537 of 11072** codes carry a plain-language `body_text`.

In `icd10.csv`, `text_variations` is `|`-joined; in `icd10.jsonl` it is a JSON list.

## Example

```json
{"code": "A22.7", "description": "Milzbrandsepsis",
 "body_text": "Sie haben sich mit Milzbrand-Erregern angesteckt.\n\nMilzbrand-Erreger sind bestimmte Bakterien...",
 "text_variations": ["A227"], "parent_code": "A22", "parent_slug": "a22",
 "slug": "a22-7", "depth": 2, "url": "https://gesund.bund.de/icd-code-suche/a22-7"}
```

## Hierarchy

The classification is a tree: three-character categories (e.g. `A22`) are
parents of their subcodes (`A22.0`, `A22.7`, …), which may themselves have
deeper subcodes. Walk it via `parent_code`, or reconstruct ranges from `code`.

## Source & method

Built by crawling the publicly served code pages on gesund.bund.de (the
National Health Portal run by the German Federal Ministry of Health). Each
page lists its direct children and carries the code + description in its
heading; the tree was walked from the 26 letter indices down to every leaf.
A documented JSON search API also exists at
`https://search.gesund.bund.de/v1/gematik?q=<code>&collection=icd`.

Descriptions are the official German ICD-10-GM labels as published on the
portal. Provided for research; verify against the official BfArM/DIMDI
classification for clinical or billing use.


## Companion file: top ICD codes by specialty

`top_icd_by_specialty_2024q1.json` — the most frequent ICD-10 categories per
ambulatory care specialty in **KV Nordrhein, Q1/2024** (source: KV Nordrhein
"Die 100 häufigsten ICD-10-Schlüssel und Kurztexte"), joined to the canonical
German descriptions and `body_text` of this dataset on the 3-character ICD
category.

Shape: `{quarter, n_specialties: 21, specialties: [{care, codes: [...]}]}`, where
each code has `rank`, `icd`, `anteil_pct` (share of cases in %), `kurztext` (KV
short label), `description` (official ICD-10-GM label), and `body_text`.

Note: in the source, *Internisten (fachärztlich tätig)* and *Nervenheilkunde*
are split into several unlabeled subspecialty lists; only the first list of each
is kept here, so those two entries represent one subspecialty rather than a
specialty-wide aggregate.