Rapid Response to the Surge of Non-Standard Medical Records in Catastrophic Events Using Multimodal Information Processing and AI-Driven Natural Language Processing

Introduction: Major public health emergencies generate a surge in non-standard medical records, which challenge efficient patient care and data management. Handwritten or incomplete records, along with simplified Hospital Information Systems (HIS), struggle to manage the surge in patient data, leading to incomplete, unstructured, and difficult-to trace records. This study explores the use of AI and NLP to standardize and manage these records during disasters. Methods: A deep learning-based system using pre-trained NLP models such as BERT and T5 is being developed. The system extracts and standardizes patient information, symptoms, and treatment details. Training will occur within the world’s only non-military, WHO-certified Type 3 International Emergency Medical Team (EMT), utilizing a large dataset from high-risk natural disaster areas in Southwest China (sourced from publicly available or open-source databases). The evaluation will focus on handling diverse and unstructured data and employing metrics such as precision, recall, and F1 score. Results: Initial testing shows promising outcomes, with the system demonstrating its potential to extract and standardize key medical information. Further validation is needed to ensure real-world reliability. Conclusion: AI and NLP present a promising solution for managing non-standard medical records in disaster medical rescue. Automating record standardization can enhance emergency response efficiency and improve patient outcomes. Future work will aim to enhance real-time data collection, integrate the system into daily prehospital and emergency care, and collaborate with experts to refine the system.