---
license: apache-2.0
task-categories:
- question-answering
language:
- en
tags:
- medical
- reinforcement-learning
- agent
size_categories:
- 10K<n<100K
source_datasets:
- ofir408/MedConceptsQA
annotations_creators:
- machine-generated
pretty_name: MedConceptsQA 15K Sample for MedARC
dataset_info:
- config_name: all
  features:
  - name: question_id
    dtype: int64
  - name: answer
    dtype: string
  - name: answer_id
    dtype: string
  - name: option1
    dtype: string
  - name: option2
    dtype: string
  - name: option3
    dtype: string
  - name: option4
    dtype: string
  - name: question
    dtype: string
  - name: vocab
    dtype: string
  - name: level
    dtype: string
  splits:
  - name: dev
    num_bytes: 9728
    num_examples: 12
  - name: test
    num_bytes: 10412499
    num_examples: 15000
  download_size: 3781436
  dataset_size: 10422227
- config_name: default
  features:
  - name: question_id
    dtype: int64
  - name: answer
    dtype: string
  - name: answer_id
    dtype: string
  - name: option1
    dtype: string
  - name: option2
    dtype: string
  - name: option3
    dtype: string
  - name: option4
    dtype: string
  - name: question
    dtype: string
  - name: vocab
    dtype: string
  - name: level
    dtype: string
  splits:
  - name: icd10cm_easy
    num_bytes: 3708173
    num_examples: 5000
  - name: icd10cm_medium
    num_bytes: 3280944
    num_examples: 5000
  - name: icd10cm_hard
    num_bytes: 3452547
    num_examples: 5000
  download_size: 3933246
  dataset_size: 10441664
- config_name: icd10cm_easy
  features:
  - name: question_id
    dtype: int64
  - name: answer
    dtype: string
  - name: answer_id
    dtype: string
  - name: option1
    dtype: string
  - name: option2
    dtype: string
  - name: option3
    dtype: string
  - name: option4
    dtype: string
  - name: question
    dtype: string
  - name: vocab
    dtype: string
  - name: level
    dtype: string
  splits:
  - name: dev
    num_bytes: 2555
    num_examples: 4
  - name: test
    num_bytes: 3699636
    num_examples: 5000
  download_size: 1445808
  dataset_size: 3702191
- config_name: icd10cm_hard
  features:
  - name: question_id
    dtype: int64
  - name: answer
    dtype: string
  - name: answer_id
    dtype: string
  - name: option1
    dtype: string
  - name: option2
    dtype: string
  - name: option3
    dtype: string
  - name: option4
    dtype: string
  - name: question
    dtype: string
  - name: vocab
    dtype: string
  - name: level
    dtype: string
  splits:
  - name: dev
    num_bytes: 4319
    num_examples: 4
  - name: test
    num_bytes: 3417134
    num_examples: 5000
  download_size: 1233636
  dataset_size: 3421453
- config_name: icd10cm_medium
  features:
  - name: question_id
    dtype: int64
  - name: answer
    dtype: string
  - name: answer_id
    dtype: string
  - name: option1
    dtype: string
  - name: option2
    dtype: string
  - name: option3
    dtype: string
  - name: option4
    dtype: string
  - name: question
    dtype: string
  - name: vocab
    dtype: string
  - name: level
    dtype: string
  splits:
  - name: dev
    num_bytes: 2854
    num_examples: 4
  - name: test
    num_bytes: 3295729
    num_examples: 5000
  download_size: 1290247
  dataset_size: 3298583
configs:
- config_name: all
  data_files:
  - split: dev
    path: all/dev-*
  - split: test
    path: all/test-*
- config_name: icd10cm_easy
  data_files:
  - split: dev
    path: icd10cm_easy/dev-*
  - split: test
    path: icd10cm_easy/test-*
- config_name: icd10cm_hard
  data_files:
  - split: dev
    path: icd10cm_hard/dev-*
  - split: test
    path: icd10cm_hard/test-*
- config_name: icd10cm_medium
  data_files:
  - split: dev
    path: icd10cm_medium/dev-*
  - split: test
    path: icd10cm_medium/test-*
task_categories:
- text-classification
- question-answering
---

# Dataset Card for MedConceptsQA 15K Sample for MedARC

This is a hierarchically stratified sample of the ICD10-CM coding system from
MedConceptsQA.
This dataset was sampled such that maximum coverage across all levels 
of the ICD-10 hierarchy to be as representative as possible while constraining
the dataset to a reasonable size for evaluation and potential reinforcement
learning.

Users can generate their own sampling of MedConceptsQA via 
[this generator script](https://github.com/sameed-khan/medconceptsqa-sample-generator).

- [Original MedConceptsQA Paper](https://www.sciencedirect.com/science/article/pii/S0010482524011740)
- [MedConceptsQA Github Repo](https://github.com/nadavlab/MedConceptsQA)

## Hierarchy Coverage

Full details can be found in `medarc_15k_report.txt`.
The below reports the percentage of chapters, categories, subcategories and 'full_codes'
(the most granular level of the classification) that are covered by this sampling.

```
Total samples: 15,000

ICD10CM:

  Available nodes in dataset:
    Chapters     : 26
    Categories   : 1914
    Subcategories: 46380
    Full codes   : 95513

  AGGREGATE COVERAGE (across all difficulties):
    Chapters     : 26 / 26 (100.0%)
    Categories   : 1914 / 1914 (100.0%)
    Subcategories: 11672 / 46380 (25.2%)
    Full codes   : 13704 / 95513 (14.3%)

  Coverage by difficulty:

  EASY:
    Chapters     : 26 / 26 (100.0%)
    Categories   : 1913 / 1914 (99.9%)
    Subcategories: 4553 / 46380 (9.8%)
    Full codes   : 5000 / 95513 (5.2%)

  MEDIUM:
    Chapters     : 25 / 26 (96.2%)
    Categories   : 1899 / 1914 (99.2%)
    Subcategories: 4510 / 46380 (9.7%)
    Full codes   : 5000 / 95513 (5.2%)

  HARD:
    Chapters     : 25 / 26 (96.2%)
    Categories   : 1708 / 1914 (89.2%)
    Subcategories: 4528 / 46380 (9.8%)
    Full codes   : 5000 / 95513 (5.2%)
```