From Development to Clinical Practice: Deploying an Interoperable and Secure ML-based CDSS for Early Sepsis Detection

Details

This paper reports on the entire process, from developing an ML-based CDSS to its deployment in a clinical setting. A system called BIAlert was developed for early sepsis detection using EHR data. Key technical features include four components: a connector that uses Apache Kafka and exchanges messages in HL7 FHIR format; a writer that writes data in FHIR format; a predictor utilizing an ML model; and a Model Evaluator for assessment. The performance evaluation demonstrated high efficacy at both HSLL and H12O hospitals. The system showed clinical utility, particularly evidenced by strong metrics like AUC and PPV. Crucially, the utilization of standards such as SNOMED CT, LOINC, ATC, and HL7 FHIR contributes significantly to ensuring interoperability. The findings provide a concrete architecture and performance validation process for applying AI technology in healthcare settings, offering valuable insights for data linkage and CDSS implementation within Japanese healthcare IT.