Azure Security and Compliance Blueprint - HIPAA/HITRUST Health Data and AI

Overview

The Azure Security and Compliance Blueprint - HIPAA/HITRUST Health Data and AI offers a turn-key deployment of an Azure PaaS solution to demonstrate how to securely ingest, store, analyze, and interact with health data while being able to meet industry compliance requirements. The blueprint helps accelerate cloud adoption and utilization for customers with data that is regulated.

The Azure Security and Compliance Blueprint - HIPAA/HITRUST Health Data and AI Blueprint provides tools and guidance to help deploy a secure, Health Insurance Portability and Accountability Act (HIPAA), and Health Information Trust Alliance (HITRUST) ready platform-as-a-service (PaaS) environment for ingesting, storing, analyzing, and interacting with personal and non-personal medical records in a secure, multi-tier cloud environment, deployed as an end-to-end solution. It showcases a common reference architecture and is designed to simplify adoption of Microsoft Azure. This provided architecture illustrates a solution to meet the needs of organizations seeking a cloud-based approach to reducing the burden and cost of deployment.

The solution is designed to consume a sample data set formatted using Fast Healthcare Interoperability Resources (FHIR), a worldwide standard for exchanging healthcare information electronically, and store it in a secure manner. Customers can then use Azure Machine Learning to take advantage of powerful business intelligence tools and analytics to review predictions made on the sample data. As an example of the kind of experiment Azure Machine Learning can facilitate, the blueprint includes a sample dataset, scripts, and tools for predicting the length of a patient's stay in a hospital facility.

This blueprint is intended to serve as a modular foundation for customers to adjust to their specific requirements, developing new Azure Machine learning experiments to solve both clinical and operational use case scenarios. It is designed to be secure and compliant when deployed; however, customers are responsible for configuring roles correctly and implementing any modifications. Note the following:

  • This blueprint provides a baseline to help customers use Microsoft Azure in a HITRUST, and HIPAA environment.

  • Although the blueprint was designed to be aligned with HIPAA and HITRUST (through the Common Security Framework -- CSF), it should not be considered compliant until certified by an external auditor per HIPAA and HITRUST certification requirements.

  • Customers are responsible for conducting appropriate security and compliance reviews of any solution built using this foundational architecture.

Deploying the automation

  • To deploy the solution, follow the instructions provided in the deployment guidance.

For a quick overview of how this solution works, watch this video explaining and demonstrating its deployment.

  • Frequently asked question can be found in the FAQ guidance.

  • Architectural diagram. The diagram shows the reference architecture used for the blueprint and the example use case scenario.

  • Deployment templates. In this deployment, Azure Resource Manager templates are used to automatically deploy the components of the architecture into Microsoft Azure by specifying configuration parameters during setup.

  • Automated deployment scripts. These scripts help deploy the solution. The scripts consist of:

  • A module installation and global administrator setup script is used to install and verify that required PowerShell modules and global administrator roles are configured correctly.

  • An installation PowerShell script is used to deploy the solution, provided via a .zip file that contains a pre-built demo functions.

Solution components

The foundational architecture is composed of the following components:

  • Threat model A comprehensive threat model is provided in tm7 format for use with the Microsoft Threat Modeling Tool, showing the components of the solution, the data flows between them, and the trust boundaries. The model can help customers understand the points of potential risk in the system infrastructure when developing machine learning components or other modifications.

  • Customer implementation matrix A Microsoft Excel workbook lists the relevant HITRUST requirements and explains how Microsoft and the customer are responsible for meeting each one.

  • Health review. The solution was reviewed by Coalfire systems, Inc. The Health Compliance (HIPAA, and HITRUST) Review and guidance for implementation provides an auditor's review of the solution, and considerations for transforming the blueprint to a production-ready deployment.

Architectural diagram

Roles

The blueprint defines two roles for administrative users (operators), and three roles for users in hospital management and patient care. A sixth role is defined for an auditor to evaluate compliance with HIPAA and other regulations. Azure Role-based Access Control (RBAC) enables precisely focused access management for each user of the solution through built-in and custom roles. See Get started with Role-Based Access Control in the Azure portal and Built-in roles for Azure role-based access control for detailed information about RBAC, roles, and permissions.

Site Administrator

The site administrator is responsible for the customer's Azure subscription. They control the overall deployment, but have no access to patient records.

  • Default role assignments: Owner

  • Custom role assignments: N/A

  • Scope: Subscription

Database Analyst

The database analyst administers the SQL Server instance and database. They have no access to patient records.

The data scientist operates the Azure Machine Learning service. They can import, export, and manage data, and run reports. The data scientist has access to patient data, but does not have administrative privileges.

Chief Medical Information Officer (CMIO)

The CMIO straddles the divide between informatics/technology and healthcare professionals in a healthcare organization. Their duties typically include using analytics to determine if resources are being allocated appropriately within the organization.

  • Built-in role assignments: None

Care Line Manager

The care line manager is directly involved with the care of patients. This role requires monitoring the status of individual patients as well as ensuring that staff is available to meet the specific care requirements of their patients. The care line manager is responsible for adding and updating patient records.

  • Built-in role assignments: None

  • Custom role assignments: Has privilege to run HealthcareDemo.ps1 to do both Patient Admission, and Discharge.

  • Scope: ResourceGroup

Auditor

The auditor evaluates the solution for compliance. They have no direct access to the network.

  • Built-in role assignments: Reader

  • Custom role assignments: N/A

  • Scope: Subscription

Example Use case

The example use case included with this blueprint illustrates how the Blueprint can be used to enable machine learning and analytics on health data in the cloud. Contosoclinic is a small hospital located in the United States. The hospital network administrators want to use Azure Machine Learning to better predict the length of a patient's stay at the time of admittance, in order to increase operational workload efficiency, and enhance the quality of care it can provide.

Predicting length of stay

The example use case scenario uses Azure Machine Learning to predict a newly admitted patient's length of stay by comparing the medical details taken at patient intake to aggregated historical data from previous patients. The blueprint includes a large set of anonymized medical records to demonstrate the training and predictive capabilities of the solution. In a production deployment, customers would use their own records to train the solution for more accurate predictions reflecting the unique details of their environment, facilities, and patients.

Users and roles

Site Administrator -- Alex

Email: Alex_SiteAdmin

Alex's job is to evaluate technologies that can reduce the burden of managing an on-premises network and reduce costs for management. Alex has been evaluating Azure for some time but has struggled to configure the services that he needs to meet the HiTrust compliance requirements to store Patient Data in the cloud. Alex has selected the Azure Health AI to deploy a compliance-ready health solution, which has addressed the requirements to meet the customer requirements for HiTrust.

Data Scientist -- Debra

Email: Debra_DataScientist

Debra is in charge of using and creating models that analyze medical records to provide insights into patient care. Debra uses SQL and the R statistical programming language to create her models.

Database Analyst -- Danny

Email: Danny_DBAnalyst

Danny is the main contact for anything regarding the Microsoft SQL Server that stores all the patient data for Contosoclinic. Danny is an experienced SQL Server administrator who has recently become familiar with Azure SQL Database.

Chief Medical Information Officer -- Caroline

Caroline is working with Chris the Care Line Manager, and Debra the Data Scientist to determine what factors impact patient length of stay. Caroline uses the predictions from the length-of-stay (LOS) solution to determine if resources are being allocated appropriately in the hospital network. For example, using the dashboard provided in this solution.

Care Line Manager -- Chris

Email: Chris_CareLineManager

As the individual directly responsible for managing patient admission, and discharges at Contosoclinic, Chris uses the predictions generated by the LOS solution to ensure that adequate staff are available to provide care to patients while they are staying in the facility.

Auditor -- Han

Email: Han_Auditor

Han is a certified auditor who has experience auditing for ISO, SOC, and HiTrust. Han was hired to review Contosoclinc's network. Han can review the Customer Responsibility Matrix provided with the solution to ensure that the blueprint and LOS solution can be used to store, process, and display sensitive personal data.

Design configuration

This section details the default configurations and security measures built into the Blueprint outlined to:

  • INGEST data raw sources including FHIR data source
  • STORE sensitive information
  • ANALYZE and predict outcomes
  • INTERACT with the results and predictions
  • IDENTITY management of solution
  • SECURITY enabled features

IDENTITY

Azure Active Directory and role-based access control (RBAC)

Authentication:

  • Azure Active Directory (Azure AD) is the Microsoft's multi-tenant cloud-based directory and identity management service. All users for the solution were created in Azure Active Directory, including users accessing the SQL Database.

  • Authentication to the application is performed using Azure AD. For more information, see Integrating applications with Azure Active Directory.

  • Azure Active Directory Identity Protection detects potential vulnerabilities affecting your organization's identities, configures automated responses to detected suspicious actions related to your organization's identities, and investigates suspicious incidents and takes appropriate action to resolve them.

  • Azure Role-based Access Control (RBAC) enables precisely focused access management for Azure. Subscription access is limited to the subscription administrator, and Azure Key Vault access is limited to the site administrator. Strong passwords (12 characters minimum with at least one Upper/Lower letter, number, and special character) are required.

  • Multi-factor authentication is supported when the -enableMFA switch is enabled during deployment.

  • Passwords expire after 60 days when the -enableADDomainPasswordPolicy switch is enabled during deployment.

Roles:

  • The solution makes use of built-in roles to manage access to resources.

  • All users are assigned specific built-in roles by default.

Azure Key Vault

  • Data stored in Key Vault includes:

    • Application insight key
    • Patient Data Storage Access key
    • Patient connection string
    • Patient data table name
    • Azure ML Web Service Endpoint
    • Azure ML Service API Key
  • Advanced access policies are configured on a need basis

  • Key Vault access policies are defined with minimum required permissions to keys and secrets
  • All keys and secrets in Key Vault have expiration dates
  • All keys in Key Vault are protected by HSM [Key Type = HSM Protected 2048-bit RSA Key]
  • All users/identities are granted minimum required permissions using Role Based Access Control (RBAC)
  • Applications do not share a Key Vault unless they trust each other and they need access to the same secrets at runtime
  • Diagnostics logs for Key Vault are enabled with a retention period of at least 365 days.
  • Permitted cryptographic operations for keys are restricted to the ones required

INGEST

Azure Functions

The solution was designed to use Azure Functions to process the sample length of stay data used in the analytics demo. Three capabilities in the functions have been created.

1. Bulk import of customer data phi data

When using the demo script. .\HealthcareDemo.ps1 with the BulkPatientAdmission switch as outlined in Deploying and running the demo it executes the following processing pipeline:

  1. Azure Blob Storage - Patient data .csv file sample uploaded to storage
  2. Event Grid - Event Publishes data to Azure Function (Bulk import - blob event)
  3. Azure Function - Performs the processing and stores the data into SQL Storage using the secure function - event(type; blob url)
  4. SQL DB - The database store for Patient Data using tags for classification, and the ML process is kicked off to do the training experiment.

Additionally the azure function was designed to read and protect designated sensitive data in the sample data set using the following tags:

  • dataProfile => “ePHI”
  • owner => <Site Admin UPN>
  • environment => “Pilot”
  • department => “Global Ecosystem" The tagging was applied to the sample data set where patient 'names' was identified as clear text.

2. Admission of new patients

When using the demo script. .\HealthcareDemo.ps1 with the BulkPatientadmission switch as outlined in Deploying and running the demo it executes the following processing pipeline: 1. Azure Function triggered and the function requests for a bearer token from Azure Active directory.

2. Key Vault requested for a secret that is associated to the requested token.

**3. Azure Roles validate the request, and authorize access request to the Key Vault.

4. Key Vault returns the secret, in this case the SQL DB Connection string.

5. Azure Function uses the connection string to securely connect to SQL Database and continues further processing to store ePHI data.

To achieve the storage of the data, a common API schema was implemented following Fast Healthcare Interoperability Resources (FHIR, pronounced fire). The function was provided the following FHIR exchange elements:

  • Patient schema covers the "who" information about a patient.

  • Observation schema covers the central element in healthcare, used to support diagnosis, monitor progress, determine baselines and patterns and even capture demographic characteristics.

  • Encounter schema covers the types of encounters such as ambulatory, emergency, home health, inpatient, and virtual encounters.

  • Condition schema covers detailed information about a condition, problem, diagnosis, or other event, situation, issue, or clinical concept that has risen to a level of concern.

Event Grid

The solution supports Azure Event Grid, a single service for managing routing of all events from any source to any destination, providing:

STORE

SQL Database and Server

Storage accounts

  • Data in motion is transferred using TLS/SSL only.

  • Anonymous access is not allowed for containers.

  • Alert rules are configured for tracking anonymous activity.

  • HTTPS is required for accessing storage account resources.

  • Authentication request data is logged and monitored.

  • Data in Blob storage is encrypted at rest.

ANALYZE

Machine Learning

SECURITY

Azure Security Center

  • Azure Security Center provides a centralized view of the security state of all your Azure resources. At a glance, you can verify that the appropriate security controls are in place and configured correctly, and you can quickly identify any resources that require attention.

  • Azure Advisor is a personalized cloud consultant that helps you follow best practices to optimize your Azure deployments. It analyzes your resource configuration and usage telemetry and then recommends solutions that can help you improve the cost effectiveness, performance, high availability, and security of your Azure resources.

Application Insights

  • Application Insights is an extensible Application Performance Management (APM) service for web developers on multiple platforms. Use it to monitor your live web application. It detects performance anomalies. It includes powerful analytics tools to help you diagnose issues and to understand what users actually do with your app. It's designed to help you continuously improve performance and usability.

Azure Alerts

  • [Alerts offer a method of monitoring Azure services and allow you to configure conditions over data. Alerts also provide notifications when an alert condition matches the monitoring data.

Operations Management Suite (OMS)

Operations Management Suite (also known as OMS) is a collection of management services.