Integrated NLP for HealthLake: Extracting Information and Mapping to FHIR from Unstructured Data

Details

AWS HealthLake's integrated NLP function processes unstructured data within FHIR DocumentReference resources. It achieves this by calling Amazon Comprehend Medical DetectEntities-V2, InferICD10-CM, and InferRxNorm API actions. The results of these actions are appended as extensions to the DocumentReference resource. Crucially, when the Amazon Comprehend Medical API detects entities with traits such as SIGN, SYMPTOM, or DIAGNOSIS, a FHIR Linkage resource is generated. Furthermore, new Condition and Observation resources are created from these specific entities and linked back to the source document using the Linkage resource. Note that while GET requests are supported for the FHIR resources generated by HealthLake integrated NLP, standard FHIR API search functionality does not support them. Since this integrated NLP feature is off by default, users must submit a support case via the AWS Support Center Console to activate it.