|
--- |
|
task_categories: |
|
- question-answering |
|
language: |
|
- en |
|
size_categories: |
|
- 10K<n<100K |
|
--- |
|
|
|
The JAMA_GxE_20240810_ft dataset is derived from the JAMA Clinical Challenge, featuring real-world clinical cases designed to enhance physicians' decision-making skills. This dataset is structured as a JSONL file, adhering to OpenAI's fine-tuning guidelines. |
|
|
|
Key Characteristics: |
|
- Training set includes cases before 2022, while the testing set comprises cases from 2022 onwards |
|
- The dataset is augmented, with each original clinical case having 8 patient profile variations: |
|
- Gender: male, female, neutral |
|
- Ethnicity: White, Black, Asian, Hispanic, Arab |
|
- Covers various medical specialties |
|
- Contains approximately 10,000 cases in the training set and 5,000 cases in the testing set |
|
|
|
Each entry includes: |
|
- Detailed patient case (limited to 250 words) |
|
- Specific clinical question |
|
- Four potential courses of action |
|
- Correct answer index |
|
- Discussion section (500-600 words) |
|
- Medical specialty classification |
|
- Link to the original case on the JAMA Network website |
|
|
|
This dataset is valuable for training and fine-tuning language models in clinical decision-making, analyzing the impact of gender and ethnicity on medical diagnoses and treatments, and enhancing AI systems' ability to assist in clinical reasoning across diverse patient populations. |
|
|
|
Metadata UI for Hugging Face: |
|
|
|
Dataset Card for JAMA_GxE_20240810_ft |
|
|
|
Dataset Summary: A collection of augmented clinical cases from the JAMA Clinical Challenge, designed for training and evaluating clinical decision-making models with consideration for gender and ethnicity factors. |
|
|
|
Supported Tasks: |
|
- text-classification |
|
- question-answering |
|
|
|
Languages: |
|
- English |
|
|
|
Dataset Structure: |
|
- Data Instances: JSON objects containing case details, options, correct answer, discussion, and metadata |
|
- Data Fields: |
|
- case: string |
|
- options: list of strings |
|
- correct_answer: string |
|
- discussion: string |
|
- field: string |
|
- link: string |
|
- gender: string |
|
- ethnicity: string |
|
- Data Splits: |
|
- Train: ~10,000 cases (before 2022) |
|
- Test: ~5,000 cases (2022 onwards) |
|
|
|
Dataset Creation: |
|
- Source: JAMA Clinical Challenge |
|
- Annotations: Original clinical cases augmented with gender and ethnicity variations |
|
|
|
Considerations for Using the Data: |
|
- Social Impact of Dataset: Enables analysis of potential biases in clinical decision-making |
|
- Discussion of Biases: Dataset is intentionally augmented to study gender and ethnicity effects |
|
- Other Known Limitations: Limited to cases from a single source (JAMA) |
