Using automated resource generation based on natural language processing (NLP) of the FHIR DocumentReference resource type in AWS HealthLake
Note
After February 20, 2023, HealthLake data stores do not use integrated natural language processing (NLP) by default. If you are interested in turning on this feature on your data store, see How do I turn on HealthLake's integrated natural language processing feature? in the Troubleshooting chapter.
If you have turned on Amazon Comprehend Medical's integrated NLP, then when you create or update DocumentReference
resources, you will incur charges in your AWS account. For more details, see AWS HealthLake pricing
Amazon Comprehend Medical isn't available in Asia Pacific (Mumbai). HealthLake data stores created in the Asia Pacific (Mumbai) region do not support integrated natural language processing (NLP).
HealthLake automatically provides you with integrated natural language processing (NLP) using Amazon Comprehend Medical for unstructured data processing for data stored in the DocumentReference
resource type. To do this, HealthLake calls the Amazon Comprehend Medical DetectEntities-V2
, InferICD10-CM
, and InferRxNorm
API operations. The results are automatically appended to the DocumentReference
resource as an extension. When the Amazon Comprehend Medical API operations detect traits that are SIGN
, SYMPTOM
, and DIAGNOSIS
, a Linkage
resource type is generated automatically. New condition and observation resources are created from entities identified with the traits of SIGN
, SYMPTOM
, or DIAGNOSIS
, and they are linked to the source document with this linkage resource.
For resources generated by the integrated NLP, you can make GET
requests, but searching these new resources is not supported.
To learn more about searching these extensions using HealthLake's integration with Athena, see Query your HealthLake data store using SQL.