Fig. 2: Schematic diagram of the COMPOSER-LLM real-time deployment pipeline. | npj Digital Medicine

Fig. 2: Schematic diagram of the COMPOSER-LLM real-time deployment pipeline.

From: Development and prospective implementation of a large language model based system for early sepsis prediction

Fig. 2

The real-time platform extracts data at an hourly resolution of all active patients using FHIR APIs, and passes the input feature set (consisting of laboratory measurements, vitals measurements, comorbidities, medications, clinical notes, and demographics) to the COMPOSER-LLM inference engine. The sepsis risk scores generated by the COMPOSER-LLM pipeline are then written back to the EHR (as a flowsheet item) through an HL7 device data interface. The flowsheet then triggers a nurse-facing Best Practice Advisory that alerts the caregiver that the patient is at risk of developing severe sepsis. HL7 health level 7, AWS Amazon Web Services, FHIR Fast Healthcare Interoperability Resources, RDS relational database service, EC2 Elastic Compute Cloud.

Back to article page