fhirstarter

An ASGI FHIR API framework built on top of FastAPI and FHIR Resources.

Supports FHIR sequences:

Installation

pip install fhirstarter

Features

Automatic, standardized API route creation
Automatic validation of inputs and outputs through the use of FHIR Resources Pydantic models
Automatically-generated capability statement that can be customized, and a capability statement API route
An exception-handling framework that produces FHIR-friendly responses (i.e. OperationOutcomes)
Automatically-generated, integrated documentation generated from the FHIR specification
Custom search parameters for search endpoints

Disclaimer

FHIRStarter was built based on the business needs of Canvas Medical. At any point in time, it may not be broadly applicable to the industry at large. Canvas Medical open-sourced the project so that it can be used by healthcare software developers whose needs it might also meet. Ongoing support and development will be based on the business needs of Canvas Medical.

Background

FHIRStarter uses a provider-decorator pattern. Developers can write functions, or handlers, that implement FHIR interactions -- such as create, read, search-type, and update -- and plug them into the framework. FHIRStarter then automatically creates FHIR-compatible API routes from these developer-provided functions. FHIR interactions that are supplied must use the resource classes defined by the FHIR Resources Python package, which is a collection of Pydantic models for FHIR resources.

In order to stand up a FHIR server, all that is required is to create a FHIRStarter and a FHIRProvider instance, register a FHIR interaction with the provider, add the provider to the FHIRStarter instance, and pass the FHIRStarter instance to an ASGI server.

Usage

Currently-supported functionality

FHIRStarter supports create, read, search-type, and update endpoints across all FHIR resource types, and will automatically generate the /metadata capabilities statement endpoint.

Handlers can be written as coroutines with async/await syntax, or as plain functions. FastAPI supports both, as does FHIRStarter.

Using uvloop can improve performance of the underlying event loop on supported platforms. FHIRStarter does not mandate the use of uvloop, but it may be enabled by importing uvloop and adding a snippet like this to your application startup script:

asyncio.set_event_loop_policy(uvloop.EventLoopPolicy())

Configuration for specific FHIR sequences

FHIRStarter will work out of the box as an R5 server. If a different sequence is desired, it must be specified with an environment variable:

FHIR_SEQUENCE=R4B

The latest version of the FHIR Resources package only supports FHIR STU3, R4B, and R5. FHIR R4 is supported by an earlier version. Because of this, if a developer desires to use FHIR R4, then the developer must pin version 6.4.0 of fhir.resources in their project. FHIRStarter will check the version of fhir.resources against the specified FHIR version in the environment variable to ensure that they are compatible.

Model imports are also affected by which version of fhir.resources is installed. For STU3 and R4B, model imports will look like this:

from fhir.resources.STU3.patient import Patient

from fhir.resources.R4B.patient import Patient

For R4 and R5, model imports will look like this:

from fhir.resources.patient import Patient

Example

A detailed example is available here: example.py.

import uvicorn
from fhir.resources.fhirtypes import Id
from fhir.resources.patient import Patient

from fhirstarter import FHIRProvider, FHIRStarter, InteractionContext
from fhirstarter.exceptions import FHIRResourceNotFoundError

# Create the app
app = FHIRStarter()

# Create a provider
provider = FHIRProvider()


# Register the patient read FHIR interaction with the provider
@provider.read(Patient)
async def patient_read(context: InteractionContext, id_: Id) -> Patient:
    # Get the patient from the database
    patient = ...

    if not patient:
        raise FHIRResourceNotFoundError

    return Patient(
        **{
            # Map patient from database to FHIR Patient structure
        }
    )


# Add the provider to the app
app.add_providers(provider)


if __name__ == "__main__":
    # Start the server
    uvicorn.run(app)

Custom search parameters

Custom search parameters can be defined in a configuration file that can be passed to the app on creation.

[search-parameters.Patient.nickname]
type = "string"
description = "Nickname"
uri = "https://hostname/nickname"
include-in-capability-statement = true

Adding a custom search parameter via configuration allows this name to be used as an argument when defining a search-type interaction handler and also adds this search parameter to the API documentation for the search endpoint.

Capability statement

It is possible to customize the capability statement by setting a capability statement modifier:

def amend_capability_statement(
    capability_statement: MutableMapping[str, Any], request: Request, response: Response
) -> MutableMapping[str, Any]:
    capability_statement["publisher"] = "Canvas Medical"
    return capability_statement

app.set_capability_statement_modifier(amend_capability_statement)

FastAPI dependency injection

FastAPI's dependency injection system is exposed at various levels:

application: the __init__ method on the FHIRStarter class
provider: the __init__ method on the FHIRProvider class
handler: the create, read, search_type, or update decorator used to add a handler to a provider

Dependencies specified at the application level will be injected into all routes in the application.

Dependencies specified at the provider level will be injected into all routes that are added to the application from that specific provider.

Dependencies specified at the handler level only apply to that specific FHIR interaction.

canvas-medical/fhirstarter