Medical Textbook Patient Presentations and Diagnoses

MedText contains over 1000 textbook-quality patient presentations and corresponding diagnoses or treatments. The dataset covers the 100 most common diseases and 30 most common injuries, with multiple examples per condition. Author BI55 published this shuffled version on Hugging Face in July 2023 to prevent model training issues.

Use Cases

• Generate synthetic patient presentations conditioned on a specific diagnosis_code or injury_type.
• Fine-tune a language model to predict a disease label from a textual description of patient symptoms.
• Train a classifier to distinguish between mild and severe presentations of the same condition.
• Benchmark few-shot learning models on the 130 distinct medical categories (diseases and injuries).

Strengths

• Over 1000 high-quality text entries
• Covers 130 distinct medical conditions (100 diseases, 30 injuries)
• Multiple data points per condition ranging from mild to severe presentations

Limitations

• Specific row count per condition is unknown, risking class imbalance
• Geographic and demographic origin of the textbook cases is unspecified
• No column details provided, limiting understanding of data structure

Provenance

Source

Hugging Face dataset authored by BI55

Collection Method

Curated from textbook-quality patient presentations; shuffled version of an original dataset

Freshness

Last updated July 2023