Development and Validation of a Generative Artificial Intelligence-Based Pipeline for Automated Clinical Data Extraction From Electronic Health Records: Technical Implementation Study

Comprehensive Summary

This study conducted by Carlisle et al. focuses on how LLM based systems can help extract data automatically from unstructured electronic health records (EHR). The study uses UODBLLM, a python-based modular LLM used to extract data from different clinical reports. The study consisted of 1800 prostate MRI radiology reports. Results showed the average processing speed to be 8.9 seconds per report across 1800 reports. While the study did not re-evaluate extraction accuracy, it was found that a 98.1% median field level accuracy was achieved for important clinical variables. At the end, the study was able to successfully have a completion rate of 100%, demonstrating that LLM agnostic systems can be used for urological documentation.

Outcomes and Implications

This research is important because the push for combining artificial intelligence and data management has seen significant progress, and several different institutes have started to adopt this. The study shows that evaluating existing structured data becomes easier to check for temporal relevance and completeness before proceeding with the process of treatment.