ClinicalGraph: An Applied Approach in Clinical EHR Knowledge Graph Generation for Optimized Clinical Decision Support System

Electronic Health Records (EHR) are well-known for their extensive capabilities in storing pertinent patient information across various specializations. However, EHRs are part of a broader ecosystem of applications that support the medical treatment process. This ecosystem, although somewhat centralized, does not have the interoperability that many assume. The main issue is that current triage models are not personalized to each patient. Context matters in a medical emergency situation especially when resources are low. In this research, we propose a possible solution by applying state of the art techniques in order to develop a clinical knowledge graph that contains relevant data used for triage optimization. We utilize a fine-tuned named entity recognition model (NER) to extract 41 label entity categories from previous medical records. Additionally, we employed prompt engineering utilizing a large language text generation model with medical knowledge to generate relationships. This resulted in a sum of 1,429 relationship type categories and approximately 999 entity nodes were created with 2,387 relationships.