Integrating a clinical product with eClinicalWorks
The EHR connection is the product. A scribe, an intake app or a coding assistant that cannot land data where clinicians already work is a demo. At Unicodez we have shipped one EHR integration: eClinicalWorks (eCW), using developer APIs the client provided. This note is about how to plan that kind of connection. It is not a catalogue of other EHRs, and it is not a claim that every eCW practice exposes the same APIs.
Start with the workflow, not the endpoint list. Who is in the room, what they document today, and which screen they trust. For an ambient scribe, the useful question is narrower than 'integrate with the EHR'. It is: which note, for which visit type, must appear in the chart, and who has to approve it before it is filed. If that sentence is fuzzy, an API credential will not make the product clearer.
eClinicalWorks publishes a developer platform, eCW Connect, for third-party apps that use its APIs. The public documentation describes FHIR access, including how an app requests authorization and then calls a FHIR resource server. The getting-started material explains FHIR as the HL7 standard for exchanging clinical data such as patients, medications and notes, and it points at REST over HTTP as the style of those calls. Read that material: eCW API documentation and the getting started guide. The underlying resource model is the HL7 FHIR R4 specification.
Public docs and a live practice are not the same thing. A client engagement depends on what that practice has turned on and on the credentials the client can hand you. Ask for this in writing before you estimate the build. Which environment is a sandbox and which is production. Who at the practice administers the app registration. Whether the first milestone is allowed to read schedule and patient context, write a draft note, or both. Which note types and specialties are in scope. How a clinician rejects or edits a draft. What test patients you may use. If the client cannot answer those, the integration is not ready to be scoped as a fixed piece of work.
Treat the client's developer access as a supplied dependency, the same way you would treat a payment account or a map key. You should not scrape a clinic's user interface, share one staff login across the product, or store a password for the EHR. The public eCW authorization flow is an OAuth-style exchange: the app sends the user to the EHR to sign in, receives a code, and trades that code for an access token that later calls present as a bearer token. Your product should keep that token on the server side of your own app, limit it to the scopes you were granted, and fail closed when it expires. The eCW docs also note that their FHIR APIs do not send CORS headers, which is a practical reason not to call the EHR directly from a browser.
Design the first write to be boring. One specialty, one note shape, one path into the chart, with the clinician reviewing the text before anything is filed. Show that draft beside the source of the visit, not as a black box. Record which encounter it belongs to, who generated it, and whether a person accepted it. That audit is useful to the clinic and it is the only honest way to talk about accuracy: the model proposes, the clinician disposes.
Map fields deliberately. A FHIR DocumentReference or a note resource is not a blob of prose you can drop in and forget. Confirm, with the client's API materials, which resource their practice expects, which status means 'preliminary' versus 'final', and how addenda work if a clinician edits the note later. Where eCW's published mapping follows US Core profiles, use those names rather than inventing a private schema that only your database understands. The point of the standard is that the next person, including the clinic's own IT staff, can see what you sent.
Plan the failure cases in the same sprint as the happy path. The token is revoked. The patient match is ambiguous. The visit was cancelled after the recording started. The practice's FHIR module is off for that site, which eCW's own endpoint listing warns can happen. The product should say what happened in language a clinic manager can act on, and it should never file a partial note because a later call failed.
Keep the promise as narrow as the access you have. 'Works with EHRs' is not a feature. 'Writes a reviewed note into this eClinicalWorks practice through the developer APIs the client gave us' is a feature. If a second vendor comes up, it is a new project with its own credentials, its own sandbox and its own clinician review, not a checkbox on the first estimate.
A sensible sequence, once the client can provide access, looks like this. Agree the visit type and the note. Get sandbox credentials and a named clinical reviewer. Read the minimum context you need to place the note on the right chart. Generate a draft. Let the reviewer accept or discard it. Only then talk about more specialties, coding suggestions or a patient-facing flow. Each of those is a separate write, with its own risk, and each should wait until the first note is dull and reliable.
Unicodez does this work as an implementation partner. The client owns the product and the EHR relationship. We work under that client's BAA and HIPAA requirements. If you are scoping an eCW connection, the useful next step is the access list above.