Healthcare Natural Language API

The Healthcare Natural Language API is a part of the Cloud Healthcare API that uses natural language models to extract healthcare information from medical text.

This conceptual guide explains the basics of using the Healthcare Natural Language API, including:

The types of requests you can make to the Healthcare Natural Language API
How to construct requests to the Healthcare Natural Language API
How to handle responses from the Healthcare Natural Language API

Overview

The Healthcare Natural Language API extracts healthcare information from medical text. This healthcare information can include:

Medical concepts, such as medications, procedures, and medical conditions
Functional features, such as temporal relationships, subjects, and certainty assessments
Relations, such as side effects and medication dosage

Choosing between the Healthcare Natural Language API and AutoML Entity Extraction for Healthcare

The Healthcare Natural Language API offers pre-trained natural language models to extract medical concepts and relationships from medical text. The Healthcare Natural Language API maps text into a predefined set of medical knowledge categories.

AutoML Entity Extraction for Healthcare allows you to create a custom entity extraction model trained using your own annotated medical text and using your own categories. For more information, see the AutoML Entity Extraction for Healthcare documentation.

Available locations

The Healthcare Natural Language API is available in the following locations:

Location name	Location description
`asia-south1`	Mumbai, India
`australia-southeast1`	Sydney, Australia
`europe-west2`	London, UK
`europe-west4`	Netherlands
`northamerica-northeast1`	Montréal, Canada
`us-central1`	Iowa, USA

Healthcare Natural Language API features

The Healthcare Natural Language API inspects medical text for medical concepts and relations. You perform entity analysis using the analyzeEntities method.

Entity analysis request fields

The Healthcare Natural Language API is a REST API and consists of JSON requests and responses. The following sample shows a simple Healthcare Natural Language API request using curl:

curl -X POST \
  -H "Authorization: Bearer $(gcloud auth application-default print-access-token)" \
  -H "Content-Type: application/json; charset=utf-8" \
  --data "{
    'nlpService': 'projects/PROJECT_ID/locations/LOCATION/services/nlp',
    'documentContent': 'Insulin regimen human 5 units IV administered.'
}" "https://healthcare.googleapis.com/v1/projects/PROJECT_ID/locations/LOCATION/services/nlp:analyzeEntities"

The sample request demonstrates the following fields:

nlpService contains the resource name of the NLP service.
documentContent contains the data for the request, which consists of medical text. The maximum size of the medical text is 20,000 unicode characters.

Entity analysis response

Entity analysis returns a set of detected medical knowledge mentions, medical concepts, and relations between medical knowledge mentions.

The response to the sample request in the Entity analysis request fields section is as follows:

{
  "entityMentions": [
    {
      "mentionId": "1",
      "type": "MEDICINE",
      "text": {
        "content": "Insulin regimen human"
      },
      "linkedEntities": [
        {
          "entityId": "UMLS/C3537244"
        },
        {
          "entityId": "UMLS/C3714501"
        },
        {
          "entityId": "UMLS/C0021641"
        },
        {
          "entityId": "UMLS/C0795635"
        },
        {
          "entityId": "UMLS/C1533581"
        },
        {
          "entityId": "UMLS/C4721402"
        }
      ],
      "temporalAssessment": {
        "value": "CURRENT",
        "confidence": 0.87631082534790039
      },
      "certaintyAssessment": {
        "value": "LIKELY",
        "confidence": 0.9999774694442749
      },
      "subject": {
        "value": "PATIENT",
        "confidence": 0.99999970197677612
      },
      "confidence": 0.41636556386947632
    },
    {
      "mentionId": "2",
      "type": "MED_DOSE",
      "text": {
        "content": "5 units",
        "beginOffset": 22
      },
      "confidence": 0.56910794973373413
    },
    {
      "mentionId": "3",
      "type": "MED_ROUTE",
      "text": {
        "content": "IV",
        "beginOffset": 30
      },
      "linkedEntities": [
        {
          "entityId": "UMLS/C0348016"
        }
      ],
      "confidence": 0.9180646538734436
    }
  ],
  "entities": [
    {
      "entityId": "UMLS/C1533581",
      "preferredTerm": "Therapeutic Insulin",
      "vocabularyCodes": [
        "MTH/NOCODE",
        "NCI/C581"
      ]
    },
    {
      "entityId": "UMLS/C0021641",
      "preferredTerm": "Insulin",
      "vocabularyCodes": [
        "FMA/83365",
        "LNC/LA15805-7",
        "LNC/LP14676-8",
        "LNC/LP16325-0",
        "LNC/LP32542-0",
        "LNC/LP70329-5",
        "LNC/MTHU002108",
        "LNC/MTHU019392",
        "MSH/D007328",
        "MTH/NOCODE"
      ]
    },
    {
      "entityId": "UMLS/C0348016",
      "preferredTerm": "Intravenous",
      "vocabularyCodes": [
        "LNC/LA9437-0",
        "LNC/LP32453-0",
        "MTH/NOCODE",
        "NCI/C13346"
      ]
    },
    {
      "entityId": "UMLS/C3537244",
      "preferredTerm": "Insulins",
      "vocabularyCodes": [
        "MSH/D061385",
        "MTH/NOCODE"
      ]
    },
    {
      "entityId": "UMLS/C3714501",
      "preferredTerm": "Insulin Drug Class",
      "vocabularyCodes": [
        "MTH/NOCODE",
        "VANDF/4021631"
      ]
    },
    {
      "entityId": "UMLS/C4721402",
      "preferredTerm": "INS protein, human",
      "vocabularyCodes": [
        "MTH/NOCODE",
        "NCI/C2271"
      ]
    },
    {
      "entityId": "UMLS/C0795635",
      "preferredTerm": "insulin, regular, human",
      "vocabularyCodes": [
        "LNC/LP17001-6",
        "MSH/D061386",
        "MTH/NOCODE",
        "NCI/C29125",
        "RXNORM/253182",
        "VANDF/4017559",
        "VANDF/4017569",
        "VANDF/4019786"
      ]
    }
  ],
  "relationships": [
    {
      "subjectId": "1",
      "objectId": "2",
      "confidence": 0.53775161504745483
    },
    {
      "subjectId": "1",
      "objectId": "3",
      "confidence": 0.95007365942001343
    }
  ]
}

Entity analysis response fields

The sample demonstrates the following response fields:

entityMentions are occurrences of medical knowledge entities in the source medical text. Each entity mention has the following fields:
- mentionId: a unique identifier for an entity mention in the response.
- type: the medical knowledge category of the entity mention.
- text: consists of the textContent field, and describes the excerpt of the medical text containing the entity mention, and offset, the location of the entity mention in the source medical text.
- temporalAssessment: specifies how the linked entity relates to the entity mention, one of CURRENT, CLINICAL_HISTORY, FAMILY_HISTORY, UPCOMING, or OTHER.
- certaintyAssessment: the negation or qualification of the medical concept, one of LIKELY, SOMEWHAT_LIKELY, UNCERTAIN, SOMEWHAT_UNLIKELY, UNLIKELY, or CONDITIONAL.
- subject: specifies the subject that the medical concept relates to, one of PATIENT, FAMILY_MEMBER, or OTHER.
- linkedEntities: a list of medical concepts that might be related to this entity mention. Linked entities specify the entityId, which links a medical concept to an an entity in entities.
entities describes the medical concepts from the linked entities fields. Each entity is described using the following fields:
- entityId: a unique identifier from the linkedEntities field.
- preferredTerm: a preferred term for the medical concept.
- vocabularyCodes: the representation of the medical concept in supported medical vocabularies.
relationships define directed relationships between entity mentions. In the sample, the subject of the relationship is "Insulin regimen human" and the object of the relationship is "5 units".
confidence: an indication of the model's confidence in the relationship as a number between 0 and 1.

Apart from the listed fields, the response might also contain the additionalInfo field, which states any additional description about the entity mention type. See Additional information.

Supported languages

The Healthcare Natural Language API only supports extracting healthcare information from English text.

Supported medical vocabularies

The Healthcare Natural Language API supports the following medical vocabularies:

Foundational Model of Anatomy
Gene Ontology
HUGO Gene Nomenclature Committee
Human Phenotype Ontology
ICD-10 Procedure Coding System
ICD-10-CM (available for US users only)
ICD-9-CM
LOINC
MeSH
MedlinePlus Health Topics
Metathesaurus Names
NCBI Taxonomy
NCI Thesaurus
National Drug File
Online Mendelian Inheritance in Man
RXNORM
SNOMED CT (available for US users only)

Supported medical knowledge categories

The Healthcare Natural Language API assigns a medical knowledge category to the entityMentions.type field. A list of supported medical knowledge categories is as follows:

Medical knowledge category	Description
`ANATOMICAL_STRUCTURE`	A complex part of the human body, such as cells, organs, and systems.
`BODY_FUNCTION`	A function carried out by the human body.
`BF_RESULT`	The result of a body function.
`BODY_MEASUREMENT`	A normal measurement of the human body, such as vital signs, obtained without any complex tests or procedures using basic instruments, such as a thermometer or a stethoscope.
`BM_RESULT`	The result of a body measurement.
`BM_UNIT`	The unit for a body measurement.
`BM_VALUE`	The value of a body measurement.
`DATE`	A date annotation, such as the date of diagnosis, date of procedure, or date of a radiation treatment.
`LABORATORY_DATA`	The results of testing a bodily sample.
`LAB_SPECIMEN`	Biological material collected from the body for testing or sampling.
`LAB_RESULT`	A qualitative description of laboratory data, such as "increased", "decreased", "positive", or "negative".
`LAB_VALUE`	The value of an instance of the laboratory data.
`LAB_UNIT`	The unit of measurement for the laboratory value.
`MEDICAL_DEVICE`	A physical or virtual instrument.
`MEDICINE`	A drug or other preparation for the treatment or prevention of a disease.
`MED_DOSE`	A dose of a medication.
`MED_DURATION`	The period of administration of a medication.
`MED_FORM`	The physical characteristics of a specific medication.
`MED_FREQUENCY`	The frequency at which a medication is taken.
`MED_ROUTE`	A location in the body where a medication is administered.
`MED_STATUS`	The status of an existing medication, such as "continue", "start", "restart", "stop", "switch", "increase", and "decrease".
`MED_STRENGTH`	The amount of active ingredient in a dose of a medication.
`MED_UNIT`	The unit of measurement for the active ingredient in a medication.
`MED_TOTALDOSE`	The quantity of medication to take at one time.
`PROBLEM`	A medical condition, including findings and diseases.
`PROCEDURE`	A diagnostic or treatment procedure.
`PROCEDURE_RESULT`	The results of a procedure.
`PROC_METHOD`	The method used to conduct a procedure.
`SEVERITY`	The severity of the medical condition.
`SUBSTANCE_ABUSE`	A description of abuse of a psychoactive substance.
`CLINICAL_STATUS`	The status of a cancer case such as "active", "recurring", "relapsing", and "resolved".
`DIMENSIONS`	The measurements of a tumor, a mass, or an abnormal growth.
`GENE_STUDIED`	The genes studied which directly or indirectly lead to tumor formation, such as BRCA1, p53, and ALK.
`HISTOLOGICAL_GRADE`	A classification system to grade the appearance of a cancerous cell.
`RADIATION_DOSAGE`	The amount of radiation given to a patient.
`ONSET`	A date annotation to represent the date when a patient first observed cancer-related problems.
`VARIATION_CODE`	A code given to the specific genomic variant that's detected under a major coding system such as Clinvar and HGVS.

Supported functional feature categories

The Healthcare Natural Language API can infer functional features, or attributes, of an entity mention from context. For example, in the statement "Kusuma's mother has diabetes", the condition "diabetes" has the functional feature of subject FAMILY_MEMBER.

Temporal relationships

Temporal relationships, returned in the temporalAssessment field, describe how this entity mention relates to the subject temporally.

The Healthcare Natural Language API supports the following temporal relationships:

CURRENT
CLINICAL_HISTORY
FAMILY_HISTORY
UPCOMING
OTHER

Subjects

Subjects, returned in the subject field, describe the individual the entity mention relates to.

The Healthcare Natural Language API supports the following subjects:

PATIENT
FAMILY_MEMBER
OTHER

Certainty assessments

Certainty assessments, returned in the certaintyAssessment field, describe the original note taker's confidence. For example, if the original note contains "The patient has a sore throat", the certainty assessment returns a LIKELY value to indicate the note taker's confidence that it was likely that the patient had a sore throat. If the original note contains "The patient does not have a sore throat", the certainty assessment returns an UNLIKELY value to indicate the note taker's confidence that it was unlikely that the patient had a sore throat.

Certainty assessments can be one of the following values:

LIKELY
SOMEWHAT_LIKELY
UNCERTAIN
SOMEWHAT_UNLIKELY
UNLIKELY
CONDITIONAL

Additional information

The additionalInfo field provides additional details about an entity mention. For example, the additionalInfo field for a DATE entity mention might consist of details about the type of the date, categorized as one of the following:

ADMISSION_DATE
CONSULTATION_DATE
DISCHARGE_DATE
SERVICE_DATE
VISIT_DATE
DIAGNOSIS_DATE
MED_STARTED_DATE
MED_ENDED_DATE
NOTE_DATE
PROCEDURE_DATE
RADIATION_STARTED_DATE
RADIATION_ENDED_DATE
STAGE_DATE

Supported relationships between entity mentions

The Healthcare Natural Language API can infer relationships between entity mentions based on the surrounding medical text. In the response, the subject of the relationship is identified by subjectId and the object of the relationship is identified by objectId.

The Healthcare Natural Language API supports the following relationships between entity mentions:

Subject	Object
`ANATOMICAL_STRUCTURE`	`MEDICAL_DEVICE`
`BODY_FUNCTION`	`BF_RESULT`
`BODY_MEASUREMENT`	`BM_RESULT`
`BODY_MEASUREMENT`	`BM_UNIT`
`BODY_MEASUREMENT`	`BM_VALUE`
`LABORATORY_DATA`	`LAB_RESULT`
`LABORATORY_DATA`	`LAB_UNIT`
`LABORATORY_DATA`	`LAB_VALUE`
`MEDICINE`	`MED_DOSE`
`MEDICINE`	`MED_DURATION`
`MEDICINE`	`MED_FORM`
`MEDICINE`	`MED_FREQUENCY`
`MEDICINE`	`MED_ROUTE`
`MEDICINE`	`MED_STATUS`
`MEDICINE`	`MED_STRENGTH`
`MEDICINE`	`MED_TOTALDOSE`
`MEDICINE`	`MED_UNIT`
`PROBLEM`	`ANATOMICAL_STRUCTURE`
`PROBLEM`	`MEDICINE`
`PROBLEM`	`PROCEDURE`
`PROBLEM`	`SEVERITY`
`PROCEDURE`	`ANATOMICAL_STRUCTURE`
`PROCEDURE`	`PROC_METHOD`
`PROCEDURE`	`PROCEDURE_RESULT`
`SUBSTANCE_ABUSE`	`SEVERITY`

Healthcare Natural Language API output as a FHIR bundle

When you request the analyzeEntities method with the alternativeOutputFormat field set to FHIR_BUNDLE, the response includes the following JSON objects:

The entity mentions, the entities, and the relationships
A FHIR R4 bundle represented as a string, that includes all the entities, the entity mentions, and the relationships in JSON format

To create the FHIR R4 bundle, the Healthcare Natural Language API maps the entity mentions, entities, and relationships to FHIR resources and their elements. The following table lists some of these mappings.

Healthcare Natural Language API entity mentions	Medical Knowledge Category	FHIR R4 resources and elements
`PROBLEM`		`Condition`
`PROBLEM`		`Condition.category`
`PROBLEM`		`Condition.status`
`PROBLEM`	`ANATOMICAL_STRUCTURE`	`Condition.bodySite`
`PROBLEM`	`ANATOMICAL_STRUCTURE`	`Condition.evidence`
`PROBLEM`	`SEVERITY`	`Condition.severity`
`PROCEDURE`		`Procedure`
`PROCEDURE`		`Procedure.status`
`PROCEDURE`		`Procedure.code`
`PROCEDURE`	`ANATOMICAL_STRUCTURE`	`Procedure.bodySite`
`PROCEDURE`	`MEDICAL_DEVICE`	`Procedure.usedCode`
`PROCEDURE`	`PROBLEM`	`Procedure.reasonReference`
`MEDICINE`		`MedicationStatement`
`MEDICINE`		`MedicationStatement.status`
`MEDICINE`		`MedicationStatement.medication`
`MEDICINE`	`PROBLEM`	`MedicationStatement.reasonReference`
`MEDICINE`	`MED_DOSE`	`MedicationStatement.dosage.doseAndRate.doseQuantity`
`MEDICINE`	`MED_FREQUENCY`	`MedicationStatement.dosage.text`
`MEDICINE`	`MED_ROUTE`	`MedicationStatement.dosage.route`
`MEDICINE`	`MED_STRENGTH`	`MedicationStatement.dosage.doseAndRate.doseQuantity`
`MEDICINE`	`MED_UNIT`	`MedicationStatement.dosage.doseAndRate.doseQuantity`

To extract entities from text as a FHIR R4 bundle, see Extract output as a FHIR R4 bundle.