2 min read

🪞Who's like me GPT?

Follow along to create a patient facing FHIR app that automatically tells you which celebrities have your medical conditions
🪞Who's like me GPT?

💡
Lets make a simple FHIR app that asks GPT which celebrities have similar medical problems to a patient given the patient's medical history.

What we are making

(that is icd-10cm Z91.89)

Fork the code

Fork my repl at https://replit.com/@patientDev/GPT3-People-Like-Me

You need the url to register an app (your url will be different than mine)

Register with Epic

Go to fhir.epic.com and login “FHIR on Epic” not “Userweb”

  1. “+ Create”
  2. Give it a name
  3. Patient is already selected
  4. Select Condition.Read Condition.Search Patient.Read Patient.Search
  5. note, some of the epic apis are specific for conditions (genomic)
  6. Add redirect as your repl.it url + “/app.html”
  7. Save

You will get a new page with a nonprod client_id

  1. Select R4
  2. Give some description
  3. Hit “Save and Ready for Sandbox”

Finish up your forked code

Go back to repl’s launch.html and change

  1. the client_id to your own
  2. the redirect to your own

Try it out

Ready to go … click the box with an arrow coming out of it on the upper right to use

[sometimes wait 30 min or so for epic to pick up the new app or you might see this]

Login as fhirderrick epicepic1

(not a secret, is at https://fhir.epic.com/Documentation?docId=testpatients )

Click “send”

The “Send to AI” is really just a lambda like this (had to hide the api-key to openai somehow … ChatGPT3 now has API as of March 1, 2023)

https://patient.dev/simple-lambda-for-openai/

You can change to the “Risk for coronary artery disease” to something like “fibromyalgia” or “diabetes” or “marfans” to get more interesting results. Then click send again.

The accuracy of GPT in this is debatable.