Has anyone joined FHIR resources?

I have FHIR-formatted data (Fast Healthcare Interoperability Resources), that I've converted to CSV using the npm package ndjson-to-csv, and on first glance, this seems useable for ML experimentation, as long as I first join the data on the relevant properties.

I could spend a while going through the approximately 12 resource types (like Device, Patient, Condition, etc) to see which resource types are joinable and if so on what property. However, I was wondering if anyone has a cheat sheet for joining FHIR resource types readily available? In other words, does anyone have an open-source approach/guide/code to de-normalizing FHIR data?

I'm just asking because FHIR is such a standardized data format that I feel like some chart / cheat sheet like this should exist. But I don't find anything relevant when googling: join fhir data by resource type.