This research introduces an aspect-oriented summarization approach to improve LLM performance on complex medical prediction tasks without requiring extensive labeled data.

• Addresses the challenge of processing lengthy medical documents with LLMs
• Proposes a two-step process: targeted summarization followed by supervised fine-tuning
• Demonstrates improved prediction accuracy for psychiatric readmissions
• Offers a practical solution when full document processing exceeds LLM capabilities

Why it matters: This approach creates a more efficient pathway to leverage LLMs in healthcare settings where document length and complexity present barriers, potentially improving patient outcomes through better readmission prediction.

Aspect-Oriented Summarization for Psychiatric Short-Term Readmission Prediction