This research demonstrates how Large Language Models can effectively encode complex Electronic Health Records (EHRs) for improved clinical predictions without needing domain-specific training.

• General-purpose LLMs outperform specialized healthcare models on multiple clinical prediction tasks
• Zero-shot prompting strategies enable effective extraction of structured information from messy health data
• Models show strong cross-dataset generalization, maintaining performance across different healthcare systems
• Findings indicate potential for significant cost and time savings in healthcare analytics

This breakthrough matters because it democratizes advanced healthcare analytics, allowing smaller institutions to leverage powerful AI without massive training datasets or specialized models, potentially accelerating clinical decision support and predictive care.

Large Language Models are Powerful EHR Encoders