wooshuchen/geriatric-depression-llm

Geriatric Depression Clinical Narrative Model

Fine-tuned qwen2.5:1.5b for generating Chinese clinical narratives assessing depression risk in older adults.

Model Details

• Base Model: Qwen/Qwen2.5-1.5B-Instruct
• Fine-tuning Method: LoRA (r=8, alpha=16)
• Training Data: 1,200 balanced samples from CHARLS (China Health and Retirement Longitudinal Study)
• Task: Generate clinical narratives explaining depression risk based on 15 patient features

Input Features

Feature	Description
age	Age
female	Gender (0=Male, 1=Female)
education_years	Years of education
married	Marital status (0=Unmarried, 1=Married)
has_diabetes	Diabetes
has_hypertension	Hypertension
has_heart_disease	Heart disease
has_stroke	Stroke history
has_lung_disease	Lung disease
adl_total	ADL limitation count (0-6)
iadl_total	IADL limitation count (0-6)
cognition_total	Cognitive score (0-21)
grip_strength	Grip strength (kg)
fall_history	Fall history
drink_now	Current drinking

Usage

from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("your-username/geriatric-depression-llm")
tokenizer = AutoTokenizer.from_pretrained("your-username/geriatric-depression-llm")

system_prompt = "你是一位临床医生，请根据患者健康档案判断抑郁风险。回答必须以'该患者抑郁风险较高'或'该患者抑郁风险较低'开头。"
profile = "年龄72岁，女性，独居，患有糖尿病和高血压，ADL受限2项，认知得分8/21，握力18kg，生活满意度2。"

messages = [
    {"role": "system", "content": system_prompt},
    {"role": "user", "content": f"请评估以下患者的抑郁风险：\n\n{profile}"},
]
text = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
inputs = tokenizer(text, return_tensors="pt")
outputs = model.generate(**inputs, max_new_tokens=400, do_sample=True, temperature=0.3)
print(tokenizer.decode(outputs[0][inputs["input_ids"].shape[1]:], skip_special_tokens=True))

Performance

Evaluated on 200 real CHARLS patients:

• Accuracy: 63.5%
• Specificity: 98.0%
• PPV: 93.5%

Citation

If you use this model, please cite:

[Paper title TBD]
[Authors TBD]

License

Apache 2.0