This research introduces a novel approach to enhance LLM performance on structured medical data by using small models as knowledge triggers.

• Addresses LLMs' numerical insensitivity and modality discrepancy with tabular medical data
• Employs specialized small models to extract and translate tabular patterns for LLMs
• Bridges the gap between statistical tabular learning and LLM reasoning capabilities
• Demonstrates significant improvements in medical tabular prediction tasks

For healthcare organizations, this approach offers a practical way to leverage both structured medical datasets and powerful LLM capabilities without extensive retraining or model modifications.

Small Models are LLM Knowledge Triggers on Medical Tabular Prediction