{"data":{"kind":"file","path":"README.md","version_id":"xzww33ouyok1fidyfl9olopx","entry":{"name":"README.md","path":"README.md","is_directory":false,"size":4338,"modified_at":"2026-02-05T21:19:37.681000","content_hash":"80f0e95995011877bb591e6acf92aa2354bbe80443691053584d971869a2d162"},"entries":[],"content":"# OpenMed AfriMedQA Environment\n\nPan-African medical multiple-choice QA environment using [AfriMed-QA](https://huggingface.co/datasets/intronhealth/afrimedqa_v2) - the first large-scale African medical QA dataset with 4K+ expert MCQs from 60+ medical schools across 16 countries.\n\n**ACL 2025 Best Social Impact Paper Award Winner**\n\n## Task Description\n\nGiven a medical question with multiple choice options (A-E), select the correct answer. Questions cover African healthcare context and 32 medical specialties.\n\n## Dataset\n\n- **Source**: [intronhealth/afrimedqa_v2](https://huggingface.co/datasets/intronhealth/afrimedqa_v2)\n- **MCQ Questions**: 4,039 expert-curated\n- **Countries**: 16 African nations (Nigeria, Tanzania, Kenya, Ghana, Uganda, etc.)\n- **Medical Schools**: 60+\n- **Specialties**: 32 (Cardiology, Pediatrics, Infectious Disease, etc.)\n- **Contributors**: 621 clinicians\n\n## Medical Specialties Covered\n\n| Category | Specialties |\n|----------|-------------|\n| Internal Medicine | Cardiology, Endocrinology, Gastroenterology, Hematology, Infectious Disease, Nephrology, Pulmonary, Rheumatology |\n| Surgery | General Surgery, Neurosurgery, Orthopedics, Plastic Surgery, Urology |\n| Primary Care | Family Medicine, Pediatrics, Geriatrics, Preventive Medicine |\n| Specialty | Dermatology, Neurology, Obstetrics/Gynecology, Oncology, Ophthalmology, Psychiatry, Radiology |\n| Other | Anesthesiology, Emergency Medicine, Medical Genetics, Otolaryngology, Pathology, Physical Medicine |\n\n## Reward Structure\n\n| Component | Weight | Description |\n|-----------|--------|-------------|\n| Accuracy | 60% | Exact match on correct answer letter (A-E) |\n| Format | 15% | Proper `\\boxed{}` and `<think>` tag usage |\n| Thinking | 15% | Quality of clinical reasoning |\n| Option Analysis | 10% | Mentions multiple options in reasoning |\n\n### Why These Rewards Work\n\n- **Accuracy (60%)**: Verifiable metric - exact letter match\n- **Format (15%)**: Encourages structured output\n- **Thinking (15%)**: Rewards clinical reasoning before answering\n- **Option Analysis (10%)**: Encourages considering alternatives (differential reasoning)\n\n## Example\n\n**Input:**\n```\nMedical Specialty: Infectious Disease\n\nQuestion:\nA 35-year-old patient in Lagos presents with cyclical fever every 48 hours,\nheadache, and fatigue for 5 days. Blood smear shows ring-form trophozoites.\nWhat is the first-line treatment?\n\nOptions:\nA. Artemether-lumefantrine (ACT)\nB. Chloroquine monotherapy\nC. Quinine alone\nD. Primaquine only\nE. Doxycycline prophylaxis\n\nSelect the correct answer.\n```\n\n**Expected Output:**\n```\n<think>\nThis patient presents with classic malaria symptoms in an endemic region (Lagos, Nigeria).\nThe cyclical 48-hour fever pattern and ring-form trophozoites on blood smear confirm\nPlasmodium falciparum or P. vivax infection.\n\nOption A: Artemether-lumefantrine is WHO-recommended first-line ACT for uncomplicated malaria\nOption B: Chloroquine resistance is widespread in Africa, not recommended as monotherapy\nOption C: Quinine is reserved for severe malaria, not first-line for uncomplicated cases\nOption D: Primaquine is for P. vivax/ovale radical cure, not primary treatment\nOption E: Doxycycline is prophylaxis, not treatment\n\nACT (Artemether-lumefantrine) is the correct first-line treatment per WHO guidelines.\n</think>\n\\boxed{A}\n```\n\n## Usage\n\n```python\nfrom OpenMed_AfriMedQA import load_environment\n\nenv = load_environment()\n```\n\n## Why AfriMed-QA for Medical RL?\n\n1. **Underrepresented context**: African healthcare scenarios rarely in training data\n2. **Expert quality**: Curated by 621 clinicians from 60+ medical schools\n3. **Verifiable answers**: MCQ format allows exact accuracy measurement\n4. **Clinical diversity**: 32 specialties, real-world African health challenges\n5. **Social impact**: Improving AI for African healthcare access\n\n## Citation\n\n```bibtex\n@inproceedings{afrimedqa2025,\n  title={AfriMed-QA: A Pan-African Multi-Specialty Medical Question-Answering Benchmark Dataset},\n  author={IntronHealth and Contributors},\n  booktitle={Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (ACL 2025)},\n  year={2025},\n  note={Best Social Impact Paper Award}\n}\n```\n\n## License\n\nDataset license follows [intronhealth/afrimedqa_v2](https://huggingface.co/datasets/intronhealth/afrimedqa_v2) terms.\n","encoding":"utf-8","truncated":false,"total_bytes":4338},"status":null}