AWS HealthScribe
AWS HealthScribe is a new HIPAA-eligible machine learning (ML) capability that combines speech recognition and generative AI to transcribe patient-clinician conversations and generate easy-to-review clinical notes. AWS HealthScribe helps healthcare software vendors build clinical applications that reduce the documentation burden and improve consultation experience. The service automatically provides rich conversation transcripts, identifies speaker roles, classifies dialogues, extracts medical terms, and generates preliminary clinical notes. AWS HealthScribe combines these capabilities to remove the need to integrate and optimize separate AI services, enabling you to expedite implementation.
Common use cases:
-
Reduce documentation time — Enable clinicians to quickly complete clinical documentation with AI-generated clinical notes that are easy to review, adjust, and finalize in your application.
-
Boost medical scribe efficiency — Equip medical scribes with AI-generated transcript and clinical notes, along with the consultation audio, to expedite documentation turn-around time.
-
Efficient patient visit recap — Create an experience that enables users to quickly recollect key highlights of their conversation in your application.
Important
The results produced by AWS HealthScribe are probabilistic and may not always be accurate due to
various factors, including audio quality, background noise, speaker clarity, the complexity of
medical terminology, context-specific language nuances, and
the nature of machine learning and generative AI
AWS HealthScribe operates under a shared responsibility model, whereby AWS is responsible
for protecting the infrastructure that runs AWS HealthScribe and you are responsible for managing
your data. For more information, see Shared Responsibility Model
AWS HealthScribe is available in US East (N. Virginia) region.
The service is available in US English (en-US). For best results, use a lossless audio format, such as FLAC or WAV, with PCM 16-bit encoding. AWS HealthScribe supports sample rates of 16,000 Hz or higher.
AWS HealthScribe currently supports General Medicine and Orthopedics specialties.
An AWS HealthScribe job analyzes medical consultation to produces two JSON output files: a transcript file and a clinical documentation file.
In the transcript file, in addition to standard turn-by-turn transcription output with word level timestamps, AWS HealthScribe provides you with:
-
Participant role detection so you can distinguish the patients from the clinicians in the conversation transcript.
-
Transcript sectioning, which categorizes transcript dialogues based on their clinical relevance like small talk, subjective, objective, etc. This can be used to show specific portions of the transcript.
-
Clinical entities, which includes structured information like medications, medical conditions, and treatments mentioned in the conversation.
In the clinical documentation file, AWS HealthScribe provides you with:
-
Summaries containing summarized notes for key sections of clinical documentation such as Chief Complaint, History of Present Illness, Review of Systems, Past Medical History, Assessment, and Plan.
-
Evidence links which links every sentence used in the AI-generated summaries to the original consultation transcript, making it easier for users to validate accuracy of the summary in your application.
API operations specific to AWS HealthScribe:
StartMedicalScribeJob
ListMedicalScribeJobs
GetMedicalScribeJob
DeleteMedicalScribeJob
To view example AWS HealthScribe requests, see Starting an AWS HealthScribe job.
Transcript file
The transcript file provides the content of the conversation in a turn-by-turn format.
In addition, the following insights are provided for each conversation turn:
-
Participant role — Each participant is labeled as either a clinician or a patient. If a conversation has more than one participant in each category, each participant is assigned a number. For example,
CLINICIAN_1,CLINICIAN_2andPATIENT_1,PATIENT_2. -
Section — Each dialogue turn is assigned to one of four possible sections based on the content identified.
-
Subjective — Information provided by the patient about their health concerns.
-
Objective — Information observed by the clinician through physical exam, lab, imaging, or diagnostic tests.
-
Assessment and Plan — Information that relates to the doctor's assessment and treatment plan.
-
Visit Flow Management — Information that related to small talk or transitions.
-
-
Insights — Extract clinically relevant entities (
ClinicalEntity) present in the conversation. AWS HealthScribe detects all clinical entities supported by Amazon Comprehend Medical.
For more detailed output information, see Example transcript output.
Clinical Documentation file
The documentation insights file contains summaries for the following key sections of the clinical documentation.
| Section | Description |
|---|---|
|
CHIEF COMPLAINT |
Brief description for the patient's reason for visiting clinician. |
|
HISTORY OF PRESENT ILLNESS |
Notes that provide information on patient's illness, including reference to severity, onset, timing of symptoms, current treatments, and the affected areas. |
|
REVIEW OF SYSTEMS |
Patient-reported evaluation of symptoms across different body systems. |
|
PAST MEDICAL HISTORY |
Details a patient's previous medical conditions, surgeries, and treatments. |
|
ASSESSMENT |
Notes that provide information on clinician's assessment of patient's health. |
|
PLAN |
Notes that reference any medical treatments, lifestyle adjustments, and further appointments. |
Every sentence present in the Summary includes references to the
original consultation transcript, making it easier for users to validate accuracy of the
summary in your application. Providing traceability and transparency for AI-generated insights
is consistent with Responsible AI principles, like explainability. Providing these references
along with the summary notes to clinicians or medical scribe helps foster trust and encourage
safe use of AI in the clinical settings.
Every sentence in the Summary comes with EvidenceLinks that
provide SegmentId for the relevant dialogues in the transcript that were summarized.
For more detailed output information, see Example clinical documentation output.
