(Paper in progress)

Background: Generative AI, particularly large language models (LLMs), is transforming clinical workflows. Through advanced prompt engineering, these models can generate high-quality, relevant medical content, optimizing time-intensive tasks such as discharge report creation.

Objective: The Informe de Alta mediante IA (IAIA) system was developed to automatically generate hospital discharge reports in the Pneumology department. Using clinical course data, IAIA produces structured summaries, detailing diagnoses, treatments, and follow-up recommendations. This operational system ensures continuity of care, reduces documentation time, and enhances efficiency.

Methods: IAIA’s development followed a rigorous, six-phase iterative process. Various LLMs were evaluated using advanced prompt engineering strategies, such as chain-of-thought reasoning, domain-specific knowledge integration, and few-shot prompting. The system was tested on 120 clinical courses, generating and evaluating discharge reports through ROUGE metrics, technical analysis, and clinician surveys.

Results: IAIA is fully operational in Pneumology, producing discharge reports with quality and consistency comparable to human-authored ones. The majority of the generated reports were rated as clinically acceptable with minimal revisions. The system significantly reduces administrative burden, freeing clinicians to focus on patient care.

Conclusion: IAIA demonstrates the capability of generative AI to revolutionize clinical documentation in real-world settings. Its success in Pneumology highlights the potential for widespread adoption, offering benefits such as improved documentation quality, efficiency, and consistency.

Next Steps: Following its success in Pneumology, IAIA is being adapted for implementation in other medical specialties. This expansion will explore its scalability and further refine its integration into diverse clinical workflows.

Keywords: IAIA, Generative AI, Pneumology, Discharge Report, Clinical Documentation.

For more information contact [email protected]