I think there's also the gamified confusion of over how much goes to the physician versus how much goes to administrating orgs and liability coverage.
It's really easy now with Claude and GPT5; just ask them a question and tell them to answer it for you, with cites, from the NHE. (The data I argue from precedes widespread LLM research; I built a site for this a couple years ago and didn't bother to update my data, because I assume it didn't get much better.)
https://www.cms.gov/data-research/statistics-trends-and-repo...