wooshuchen/geriatric-depression-llm
The wooshuchen/geriatric-depression-llm is a 1.5 billion parameter Qwen2.5-based language model fine-tuned for generating Chinese clinical narratives. It specializes in assessing depression risk in older adults based on 15 specific patient features, offering a context length of 32768 tokens. This model is optimized for clinical decision support, providing narrative explanations of depression risk rather than just a binary classification.
Loading preview...
Geriatric Depression Clinical Narrative Model
This model, developed by wooshuchen, is a fine-tuned Qwen2.5-1.5B-Instruct variant specifically designed for generating Chinese clinical narratives. Its primary function is to assess and explain depression risk in older adults based on comprehensive patient health profiles.
Key Capabilities
- Specialized Narrative Generation: Produces detailed clinical narratives explaining depression risk (high or low) for geriatric patients.
- Feature-Rich Analysis: Utilizes 15 distinct patient features, including age, gender, education, marital status, chronic diseases (diabetes, hypertension, heart disease, stroke, lung disease), ADL/IADL limitations, cognitive scores, grip strength, fall history, and current drinking habits.
- Clinical Decision Support: Aims to assist clinicians by providing explanatory text rather than simple risk scores.
- Performance: Achieves 63.5% accuracy, 98.0% specificity, and 93.5% PPV on a test set of 200 CHARLS patients.
Training Details
The model was fine-tuned using the LoRA method (r=8, alpha=16) on 1,200 balanced samples from the China Health and Retirement Longitudinal Study (CHARLS) dataset.
Good For
- Generating automated clinical notes for depression risk assessment in elderly Chinese populations.
- Research into AI-assisted diagnostic narrative generation in geriatrics.
- Applications requiring detailed, feature-based explanations of health risks.