Scout: A Novel LLM-Driven EHR Search and Synthesis Tool

The burden of clinical documentation and data extraction from Electronic Health Records (EHRs) significantly contributes to clinician burnout. In response, we created Scout, an innovative platform that leverages large language models (LLMs) to allow healthcare professionals to search EHR data using natural language queries. Each output from Scout is accompanied by citations that link back to the original data, ensuring that clinicians can easily verify the information provided. We executed a randomized, evaluator-blinded crossover study involving 20 participants across seven clinical specialties, focusing on 200 structured cases. Participants engaged in realistic clinical tasks utilizing either Scout or the traditional EHR system. The results indicated that Scout reduced the time taken to complete tasks by 37.6% and notably lowered perceived workload, particularly in areas such as mental demand and effort. Furthermore, non-inferiority tests confirmed that the accuracy, completeness, and relevance of tasks performed with Scout were on par with those done via the EHR alone. A pilot deployment involving over 200 users across more than 20 specialties yielded over 6,600 interactions within three months, highlighting various clinical and administrative applications. Automated evaluations identified errors infrequently, and a manual review showed that many flagged claims were indeed supported by patient records, underscoring the necessity of human oversight. These preliminary findings suggest that LLM-enhanced EHR tools can significantly ease clinician workloads while preserving the quality of outputs.