Skip to content

Latest commit

 

History

History
 
 

demo

README.md

Android FHIR SDK with the FHIR Info Gateway

Before you begin

What you'll see

The demo Android app uses the Android FHIR SDK to show a patient registration questionnaire and save them locally. It also syncs data with a FHIR server through the FHIR Info Gateway, which provides access control to FHIR resources based on the authenticated user. Finally, FHIR Data Pipes periodically transforms the data to Parquet files that you can query to perform analytics.

What you'll learn

  • How to integrate the Android FHIR SDK with the FHIR Info Gateway
  • How the FHIR Info Gateway works
  • How to implement interfaces for syncing FHIR resources
  • How to transform and export FHIR data to query with Spark SQL

The diagram below shows the different components that are involved:

Setup Overview

What you'll need

Demo app setup

  1. Clone the FHIR App Examples repo:

    git clone https://github.com/google/fhir-app-examples.git

  2. Open Android Studio, select Import Project (Gradle, Eclipse ADT, etc.) and choose the fhir-app-examples folder downloaded in the previous step. If this is your first time opening the project, the Gradle Build process should start (and take some time).

FHIR server and Data Pipes setup

  1. Clone the FHIR Data Pipes repo:

    git clone https://github.com/google/fhir-data-pipes.git

  2. Follow the instructions to set up a local test server. At step 2, follow the instructions for a "HAPI source server with Postgres". You can omit the optional step 4.

    You now have a HAPI FHIR server that you loaded with synthetic patient data; exactly 79 patients, if it only has the test data.

  3. From a terminal, run:

    PATIENT_ID1=4765
PATIENT_ID2=4767

curl -X PUT -H "Content-Type: application/json" \
  "http://localhost:8091/fhir/List/patient-list-example" \
  -d '{
      "resourceType": "List",
      "id": "patient-list-example",
      "status": "current",
      "mode": "working",
      "entry": [
         {
            "item": {
            "reference": "Patient/'"${PATIENT_ID1}"'"
            }
         },
         {
            "item": {
            "reference": "Patient/'"${PATIENT_ID2}"'"
            }
         }
      ]
   }'

    Note: If you are copy and pasting from OSX, you may need to first paste into a text editor to put all of that in a single line (and removing trailing "\"), before pasting that into shell to run.

    This creates a FHIR List on the server with the id patient-list-example, which we will use as an access control list. It contains references to Patient/4765 and Patient/4767 which the user is allowed to access.

  4. Follow the instructions to set up a single-machine analytics pipeline. This Docker image includes the FHIR Pipelines Controller plus a Spark Thrift server where data is ultimately loaded for querying.

  5. In a web browser, visit http://localhost:8090 to see the FHIR Pipelines Controller UI.

IDP and Info Gateway setup

  1. Clone the FHIR Info Gateway repo:

    git clone https://github.com/google/fhir-gateway.git

  2. Start the Keycloak Identity Provider Server. From the fhir-gateway directory, run:

    docker-compose -f docker/keycloak/config-compose.yaml \
  up --force-recreate --remove-orphans -d --quiet-pull

    The config-compose.yaml sets up a Keycloak instance that can support both a list-based access control and a single-patient based SMART-on-FHIR app (in two separate realms).

    The keycloak-config image is built using the Dockerfile here. A key component of the Dockerfile is the keycloak_setup.sh. There are two points of interest in this script: the first is this, which creates a client that authenticated users can act as, and here where we create a user that binds the patient-list-example value to the patient_list claim field that is part of the JWT access token. The default username and password used for the user are from the env file.

  3. Start the FHIR Info Gateway. From the fhir-gateway directory, run:

    docker run --rm --network host \
  -e TOKEN_ISSUER="http://localhost:9080/auth/realms/test" \
  -e PROXY_TO="http://localhost:8091/fhir" \
  -e BACKEND_TYPE="HAPI" \
  -e RUN_MODE="DEV" \
  -e ACCESS_CHECKER=list \
  -e ALLOWED_QUERIES_FILE="resources/hapi_page_url_allowed_queries.json" \
  us-docker.pkg.dev/fhir-proxy-build/stable/fhir-access-proxy:latest

    This brings up a FHIR Info Gateway, connected to the HAPI FHIR server. The TOKEN_ISSUER variable is the IP of the Keycloak IDP from the previous step, and the PROXY_TO variable is the IP of the FHIR server. As we are running the TOKEN_ISSUER and FHIR Info Gateway on the same machine (but on different ports), we need to bypass the Proxy's token issuer check by setting the environment variable RUN_MODE to DEV.

    WARNING: Never use RUN_MODE=DEV in a production environment.

    Part of setting up the FHIR Info Gateway is choosing the type of Access Checker to use. This is set using the ACCESS_CHECKER environment variable (See here for more detail). In this demo, we will use the default value of list, which will use the ListAccessChecker to manage incoming requests. This access-checker uses the patient_list ID in the JWT access token to fetch the "List" of patient IDs that the given user has access to. There are some URL requests that we want to bypass the access checker (e.g. URLs with _getpages in them) and we declare these rules in hapi_page_url_allowed_queries.json. To make the server use this file, we set the environment variable ALLOWED_QUERIES_FILE.

See the components in action

This example demonstrates several components of Open Health Stack.

Android FHIR SDK in the demo app

The Demo app uses the Structured Data Capture library to render the patient registration and survey forms, and to extract FHIR resources based on the responses. You can see a form by clicking the Add Patient button in the bottom-right of the main screen.

The Demo app also uses the FHIR Engine library to save FHIR resources in the app and sync them with a FHIR server. You can see this when resources sync from the server the first time, or when you register new patients.

  1. In Android Studio, with an Android Emulator installed, run the demo app by pressing on the Play button on the top bar. This will build the app and open the emulator.

  2. Once the app finishes building it will launch in the emulator and its logs will be available in the bottom Run tab of Android Studio.

  3. In the Emulator, press the Log In button, which will take you to the IDP login screen. Type testuser as the username and testpass as the password.

  4. The app will then start the syncing process. You can see this in the logs displayed in the Run tab.

Info Gateway

When the Demo app syncs resources with the FHIR server, it is actually communicating with a FHIR Info Gateway. It uses the List Access Checker to determine which Patient resources testuser has access to, and then fetches the resources from the actual FHIR server when allowed. The demo app is designed to only send requests that are expected to succeed, but you can follow the guide to try out the Info Gateway for more information.

FHIR Data Pipes

The FHIR Data Pipes Pipelines Controller facilitates the transformation of data from a FHIR server to Parquet files. In this guide, you use the single machine configuration which also loads the Parquet files into a Spark Thrift server for you.

  1. Visit the Pipeline Controller UI at http://localhost:8090.

  2. Click on Run Full to generate the Parquet files.

  3. Connect to jdbc:hive2://localhost:10001 using a Hive/Spark client.

  4. Count the number of patients:

    SELECT COUNT(0) FROM default.patient;

  5. From the demo app running in the Android Emulator, register a new patient by selecting the New Patient (+) button and complete the registration form.

  6. Force the app to sync with the server by tapping the menu button and selecting Sync.

  7. Update the Parquet files by visiting the Pipeline Controller UI and clicking Run Incremental.

  8. Query the number of patients again:

    SELECT COUNT(0) FROM default.patient;

If you have any errors when running the incremental pipeline or it fails to work, try using sudo chmod -R 755 on the Parquet file directory, default located at fhir-data-pipes/tree/master/docker/dwh.

Implementation details

Initial Launch

When the app is launched, the first class launched is FhirApplication, as it is a subclass of Application and specified in the "android:name" field in AndroidManifest.xml. Part of the FhirApplication class instantiates a ServerConfiguration. We pass into the ServerConfiguration the URL of the FHIR Access Proxy. As we are running the Proxy and the App from the same machine, we use 10.0.2.2 as a special alias to the host loopback interface (i.e., 127.0.0.1 on the same machine). We also pass into the ServerConfiguration an instance of Authenticator for supplying the Proxy the JWT access token; LoginRepository is the implementation of Authenticator we wrote.

Fetching Access Token

Our end-to-end setup uses OAuth 2.0 authorization code flow to retrieve an access token.

After initializing the FhirApplication class, the next class launched is the LoginActivity class, as specified by the intent filters in the AndroidManifest.xml file. The LoginActivity class initializes the LoginActivityViewModel class; the LoginActivityViewModel contains two methods that are called by LoginActivity: createIntent and handleLoginResponse. The first method returns an Intent that is bound to the Log In button. The intent is built by first fetching the Discovery Document from the Proxy. The URL to the Proxy discovery endpoint is loaded from the auth_config.json. When a request to the Proxy is made to this endpoint, it returns a response that includes the value of TOKEN_ISSUER, which is needed to create the login Intent.

When the Log In button is pressed, the Intent opens a webpage to the login screen with the value of TOKEN_ISSUER as the base URL, where the user is prompted to type in their credentials. Once the user logs in, the callback defined in the getContent variable in LoginActivity runs, which takes the response from the IDP containing an authorization code, and passes it to the handleLoginResponse method. This method abstracts the exchange of the authorization code for an access token, which is stored in the App. Any call in the app now made to LoginRepository.getAccessToken fetches the stored JWT, and if expired, refreshes the token.

Download Resources

Once the user logs in, the MainActivity class is launched, which instantiates the MainActivityViewModel class. When MainActivityViewModel is initialized, it launches an instance of SyncJob. One of the parameters we need to pass in to the SyncJob.poll method is an implementation of the FhirSyncWorker abstract class, which we provide via the FhirPeriodicSyncWorker class.

FhirPeriodicSyncWorker implements two methods, one of which is getDownloadWorkManager. The implementation of that method requires a DownloadWorkManager returned, a class that we also have to implement. We have to provide a way for the SDK to generate the FHIR download requests and handle the FHIR responses returned, and we do that via the DownloadWorkManagerImpl class. This class takes in an initial resource ID to seed the first download request; this resource ID comes from the value of the patient_list claim that is part of the JWT access token now stored on the App.

As we logged in as testuser, the value of patient_list will be patient-list-example, which we defined in Keycloak. patient-list-example is the ID of a List resource on the FHIR server that we first want to fetch. With FhirPeriodicSyncWorker and DownloadWorkManagerImpl instantiated, the SyncJob.poll method runs and downloads all resources as specified in the classes we created.