How Amazon Comprehend Medical is integrated with HealthLake
HealthLake infers data found in the DocumentReference
resource type using Amazon Comprehend Medical. The Amazon Comprehend Medical API operations DetectEntities-V2
, InferICD10-CM
, and InferRxNorm
detect medical conditions as traits. Each operation provides different insights.
Language support
Amazon Comprehend Medical API operations only detect medical entities in English language texts.
-
DetectEntities-V2: Inspects the clinical text for a variety of medical entities and returns specific information about them, such as entity category, location, and confidence score.
-
InferICD10-CM: Detects medical conditions in a patient record as entities, and it links those entities to normalized concept identifiers in the ICD-10-CM knowledge base from the CDC's National Center for Health Statistics under authorization by the World Health Organization (WHO).
-
InferRxNorm: Detects medications as entities listed in a patient record, and it links them to the normalized concept identifiers in the RxNorm database from the National Library of Medicine.
The supported traits for each API operation are SIGN
, SYMPTOM
, and DIAGNOSIS
. If traits are detected, they are added as FHIR-compliant extensions to different locations in your HealthLake data store.
Locations where extensions are added.
-
DocumentReference
: The results from the Amazon Comprehend Medical API operations are added as anextension
to each document found within the DocumentReference resource type. Results in the extension are divided into two groups. You can find them in the results based on theirURL
.-
http://healthlake.amazonaws.com/system-generated-resources/
-
These are resource types that have been created or added to by HealthLake.
-
-
http://healthlake.amazonaws.com/aws-cm/
-
Where the raw output of the Amazon Comprehend Medical API operations is added to your HealthLake data store.
-
-
-
Linkage
: This resource type is either added or created as a result of the integrated NLP. AGET
request on a specificLinkage
returns a list of linked resources. To identify if aLinkage
was added by HealthLake, look for the added"tag": [{"display": "SYSTEM_GENERATED"}]
key-value pair. To learn more about the FHIR specifications for Linkage, see Resource type: Linkagein the FHIR Documentation Index. -
FHIR resource types generated as a result of the Amazon Comprehend Medical API operations.
-
Observation
: Has results from the Amazon Comprehend Medical API operations DetectEntities-V2 and InferICD10-CM added to it when the traits areSIGN
orSYMPTOM
. -
Condition
: Has results from the Amazon Comprehend Medical API operations DetectEntities-V2 and InferICD10-CM added to it when the traits areDIAGNOSIS
. -
MedicationStatement
: Has results from the Amazon Comprehend Medical API operation InferRxNorm added to it.
-
Integration with the FHIR REST API operations
By default, traits detected by the Amazon Comprehend Medical API operations are not returned when making a GET
request.
To see the results of the integrated NLP operations for these resource types, you must specify a known ID
.
-
Linkage
-
Observation
-
Condition
-
MedicationStatement
The results of the integrated NLP operations outside the DocumentReference resource type are only available using a GET
request where the specified ID
is know to contain results from the Amazon Comprehend Medical API operations.
Examples of how Amazon Comprehend Medical API operations are integrated into HealthLake
Example 1: Patient record ingested into a HealthLake data store
Here is an example clinical note based off of a patient's encounter with a medical professional.
Synthetic data
The text in this example is synthetic content and doesn't contain personal health information (PHI).
1991-08-31 # Chief Complaint - Headache - Sinus Pain - Nasal Congestion - Sore Throat - Pain with Bright Lights - Nasal Discharge - Cough # History of Present Illness Jerónimo599 is a 4 month-old non-hispanic white male. # Social History Patient has never smoked. Patient comes from a middle socioeconomic background. Patient currently has Aetna. # Allergies No Known Allergies. # Medications No Active Medications. # Assessment and Plan Patient is presenting with bee venom (substance), mold (organism), house dust mite (organism), animal dander (substance), grass pollen (substance), tree pollen (substance), lisinopril, sulfamethoxazole / trimethoprim, fish (substance). ## Plan The patient was prescribed the following medications: - astemizole 10 mg oral tablet - nda020800 0.3 ml epinephrine 1 mg/ml auto-injector The patient was placed on a careplan: - self-care interventions (procedure)
As a reminder, this information is encoded in base64 format in the DocumentReference resource. When this document is ingested into HealthLake and the Amazon Comprehend Medical API operations are complete, to see the results, you can start with the GET
request on the DocumentReference
resource type.
GET https://https://healthlake.
your-region
.amazonaws.com/datastore/your-datastore-id
/r4/eeb8005725ae22b35b4edbdc68cf2dfd
/r4/DocumentReference
When the Amazon Comprehend Medical API operations are successful, look for these key-value pairs inside the extension
linked to the following "url": "http://healthlake.amazonaws.com/aws-cm/"
{ "url": "http://healthlake.amazonaws.com/aws-cm/status/", "valueString": "SUCCESS" }, { "url": "http://healthlake.amazonaws.com/aws-cm/message/", "valueString": "The Amazon HealthLake integrated medical NLP operation was successful." }
The following tabs show you how the ingested medical record is reported in your HealthLake data store based on the resource type.
Example 2: A DocumentReference
that contains MedicationStatement resource type
Here is an example of a clinical note based off of a patient's encounter with a medical professional.
Synthetic data
The text in this example is synthetic content and doesn't contain personal health information (PHI).
Tom is not prescribed Advil
The following tabs show how the ingested medical record is reported in your HealthLake data store based on the resource type.