Ideas2IT rewards key players with 1/3rd of the Company in New Initiative.  Read More >
Back to Blogs

Amazon HealthLake: An AWS service that moved Healthcare


Healthcare systems in the world are fairly complex and different from each other in several ways. Healthcare systems in today’s world are becoming more and more data-driven. The challenge in healthcare analytics is that most information in healthcare is unstructured, dirty, and difficult to use for analysis. Most valuable information in healthcare is lost due to this nature of the data that is present throughout the healthcare system in the form of lab test reports, claims made, medical images, clinical notes, and other vital time-series statistics like heart rate, ECG races, etc. Almost all research in healthcare analytics is now focused on gathering as much information from these unstructured data, which includes discharge summaries, clinical notes, medical images.

Every health care provider, payer, and life sciences company needs to solve the problem of structuring the data to make better patient support decisions through AI. This is also important also in B2B healthcare services like claims and insurances and other services. These claims and insurance covers become very important even during the analysis of patient data to improve patient care and growth in revenue from the care-giver side as well as the claims company’s side.

Is Amazon the new go-to for solving this problem? In December 2020, Amazon launched Amazon HealthLake as an AWS service, which will automatically understand and extract medical information including rules, procedures, and real-time diagnoses. Amazon HealthLake is a HIPAA-eligible service that enables healthcare providers, payers, and pharmaceutical companies to store, transform, query, and analyze health data at a petabyte-scale. It performs much of the work of organizing, indexing, and structuring patient information, to provide a complete view of each patient’s medical history in a more secure, compliant, and auditable manner. Healthlake normalizes the data for ease in the use of further analysis. Upon ingestion, Amazon HealthLake uses machine learning models that are trained to understand medical terminology to identify and tag each piece of clinical information, index events into a timeline view, and enrich the data with standardized labels (e.g., medications, conditions, diagnoses, procedures, etc.), which facilitates quick and easy information search retrieval.

For example, if a healthcare provider wants to know “What is the immunization status for their patients?”. This could be easily queried from the timelined data by searching for “Immunization” with search parameters as “patients” for a detailed view of immunization status for different diseases for different patients.

[caption id="attachment_10866" align="alignnone" width="673"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

Healthlake recognizes interoperability standards, by including Fast Healthcare Interoperability Resources(FHIR R4), which is a standardized data sharing format in the healthcare systems. This enables providers to collaborate more effectively and allows unfettered access to their medical information. Support for other formats, like X12, CCD, etc. may be on the way. Health analytics will be much easier with all the heavy lifting being taken care of by this end-to-end service. In the next section, we will discuss some of the major capabilities of HealthLake.

Key Capabilities

Let’s look at some of the major capabilities of HealthLake in better detail to understand this service better.

  • Import | Quickly and easily ingest health data: Bulk import allows customers to easily migrate their on-premise FHIR files including clinical notes, lab reports, insurance claims, and more to an S3 bucket in their account, where their data can be used in further downstream applications.
  • Store | Store your data in the AWS Cloud in a more secure, compliant, and auditable manner: Data Store helps you index all of your information so it can be easily queried. The Data Store creates a complete view of each patient’s medical history in chronological order and facilitates the exchange of information using the V4 FHIR specification. The Data Store is always running to keep your index up to date, offering you the ability to query the information anytime using the standard FHIR Operations with durable primary storage and index scaling. Amazon HealthLake meets rigorous security and access controls to ensure patients’ sensitive health data is protected and meets regulatory compliance.
  • Transform | Transform unstructured data using specialized ML models: Integrated medical natural language processing (NLP) transforms all of the raw medical text data from the Data Store using specialized ML models that have been trained to understand and extract meaningful information from unstructured healthcare data. With integrated medical NLP, you can automatically extract entities (e.g., medical procedures, medications), entity relationships (e.g., a medication and its dosage), entity traits (e.g., positive or negative test result, time of procedure), and Protected Health Information (PHI) data from your medical text. For example, HealthLake can accurately identify patient information from an insurance claim, extract laboratory reports, and map to medical billing codes like ICD-10 in minutes, rather than hours or weeks.
  • Query | Use powerful query and search capabilities: Amazon HealthLake supports FHIR CRUD (Create/Read/Update/Delete) and FHIR Search operations. You can query records by performing a Create Operation for adding new patients and their information, like medications. You can read the most recent version of that record by performing a Read Operation. You can update a previously created record by performing an Update Operation. As per the FHIR specification, deleted data is only hidden from analysis and search results; it is not deleted from the service, only versioned. You can also search with predefined filters to find all the information on a patient.
  • Analyze: Amazon HealthLake enables customers to bulk export their FHIR data from the HealthLake Data Store to an S3 bucket. With Amazon QuickSight, developers can create dashboards on the exported and normalized data to quickly explore trends about their patients or population and predict events and personalize care at the individual level. Developers can also build, train and deploy their own ML models on their data with Amazon SageMaker.

A complete guide to creating the Data Store and importing data onto it can be found in their detailed documentation here.

At this point, when we have discussed the major services available in HealthLake, Let’s see some of the integral parts of those services, to see how the whole service comes together. In this light, it’s important to understand how the healthcare resources(here FHIR) are managed, or how NLP integration in the service makes it stand out. In the next two sections, we shall discuss this, and succeeding that would be a snippet of demonstration of the queries that you can do for information retrieval.

Managing FHIR Resources

In an active Data Store at HealthLake, users can create, delete, update resources by using the FHIR Rest APIs. The following table lists the operation that can be performed for FHIR resource management.

[caption id="attachment_10867" align="alignnone" width="679"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

There are as many as 71 resource types that are supported by Healthlake, a comprehensive list of which can be found here.

Some common search parameters for querying FHIR data can be listed as below:

[caption id="attachment_10868" align="alignnone" width="673"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

Integrated NLP

Amazon HealthLake automatically integrates with natural language processing (NLP) for the DocumentReference resource type. The integrated medical NLP output is provided as an extension to the existing DocumentReference resource. The integration involves reading the text data within the resource, and then calling the following integrated medical NLP operations: DetectEntities-V2, InferICD10-CM, and InferRxNorm. The response of each of the integrated medical NLP APIs is appended to the DocumentReference resource as an extension that is searchable. This enables users to identify patients through elements of their records that were previously buried within the unstructured text. When you create a resource in HealthLake, the resource is updated with the response from the integrated medical NLP operations. These extensions follow the FHIR format for extensions with an identifying URL and the respective value for the URL.

[caption id="attachment_10870" align="alignnone" width="675"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

Brief Query Demo

Users can query information regarding the patients, medications, disease codes from the data very easily in HealthLake. As discussed before, the NLP integration makes it possible to do very specific queries. Let’s look at the following two queries.

1. Immunization Status: Say, we would like to know what the immunization status of the patients is. We need to make a query similar to the following.

[caption id="attachment_10871" align="alignnone" width="673"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]The value is optional. You can put the patient ID for referring to a specific patient. This will return a list of all patients with immunization history and their status. One small snippet is attached for better understanding.[caption id="attachment_10878" align="alignnone" width="675"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

The above result shows that the influenza immunization has been “completed” for the patient. The patients reference ID which is seen above can be used for specific queries regarding

2. Rx norm search

Let’s say, from the clinical notes and documents, you would like to know which patients intake a certain medicine, say “Clonidine” (a drug used in the treatment of hypertension). Let’s make the following query.

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS

A snippet from the results are shown as follows.

[caption id="attachment_10880" align="alignnone" width="672"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

The specific medicine is identified, along with the matching score, category, and type. This type of information becomes necessary features for further study, say related to Hypertension and the likes of it.

At the final step, when the imported data is normalized and all transformations are ready, the end job is to export this standardized data for further analysis. The next section discusses that in brief.

Data Export

The normalized data is exported to S3 for further analysis. You can simply do this with a few clicks in the AWS console. Below is a snippet of the export job, if you wish to go into further details please visit the export job documentation here.

[caption id="attachment_10875" align="alignnone" width="669"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

You need to provide the S3 location and select a suitable IAM role and you’re good to go!

The utility of AWS HealthLake ends at the export of the FHIR data. Analysis of the data and building machine learning models need to be done separately through Quicksight and Sagemaker.


HealthLake is in Preview as of the date of writing this blog and is free to use during this period. The following pricing will be effective at General Availability.

[caption id="attachment_10876" align="alignnone" width="673"]

Amazon HealthLake: An AWS service that moved Healthcare

Image Courtesy: AWS[/caption]

A Few Shortcomings

Healthlake is a fairly end-to-end AWS service. However, there are very few shortcomings that I think the user might face while using HealthLake. The following holds true at the time of writing this blog, or until subsequent releases address them.

  • Limited data format support: Only FHIR(R4) format data is supported here. Healthcare systems that still haven’t migrated to FHIR, may face difficulty using this service. Resources need to be converted to FHIR for use in this service.
  • No Query-based export: The export job, transforms and exports the whole data pool in S3. There is no provision to export the queried data to S3, which can be useful for specific studies. For example, if a study is based on only blood sugar patients, they need to be filtered out after export.
  • Minor inconsistency in query results: This is based on the experience while using the preloaded data for queries. Sometimes, results are inconsistent with what is queried. Several redundant information is sometimes likely to appear in such queries, making the readability of the results complex.
  • Results readability: Resource search results appear in FHIR format and no other easy-to-read format like tables and data frames are not supported. Until someone is experienced with FHIR and interoperability, it might cause a user difficulty in full comprehension of the query results.

If you are looking for a job that gives you exposure to innovation-led healthcare projects and emerging cutting-edge technologies, check out open roles at

Ideas2IT Team

Connect with Us

We'd love to brainstorm your priority tech initiatives and contribute to the best outcomes.