Expert-guided Clinical Text Augmentation via Query-Based Model Collaboration

Data augmentation is a widely used strategy to improve model robustness and generalization by enriching training datasets with synthetic examples. While large language models (LLMs) have demonstrated strong generative capabilities for this purpose, their applications in high-stakes domains like healthcare present unique challenges due to the risk of generating clinically incorrect or misleading information. In this work, we propose a novel query-based model collaboration framework that integrates expert-level domain knowledge to guide the augmentation process to preserve critical medical information. Compared to existing LLM-based and traditional augmentation methods, our generated data significantly improves preservation of critical medical information and reduces hallucinations at both the token and concept levels. Experiments on downstream clinical prediction tasks demonstrate consistent performance gains over existing augmentation methods. This lightweight collaborative framework addresses the gap between LLM augmentation potential and the safety requirements of specialized domains.

AI systems can be made more reliable by training them with extra synthetic examples, but this is risky in healthcare: a rewritten clinical note that changes a symptom, medication, or diagnosis can teach a model the wrong information. We propose a safer way to create synthetic clinical notes by pairing a strong text-generating AI model with a smaller medical expert model. The expert identifies important medical details that should not be changed, and the text generator rewrites the note while preserving them. Across several clinical prediction tasks, our method produced notes with fewer invented medical details, preserved more critical information, and improved model performance compared with common text augmentation methods. This approach helps make AI-generated training data more useful and safer for high-stakes domains like healthcare.