Yes, Chez Scheme is fast but it took a lot of time (a few years!) to make the compiler translate Rhombus/Racket code to fast Chez Scheme code. And there are still a few rough corners to fix...
Also, Matthew implemented flonum unboxing to make code that uses a lot of floating points like x10 faster. I added type recovery, that gives a 10% boost in many cases. And I'm probably missing a few similar features.
In most cases the Chez Scheme team also was involved. Those changes were upstreamed, so you can enjoy them in Rhombus, Racket and Chez Scheme.
I think this depends on the context. For your epidemiologist perspective, or looking at populations, that's a big rate! (Given how much metformin is taken) Would be easy to find people who suffer from it.
From the perspective of "Should I be worried about side effects from taking this drug / should I not take this because it might give me this effect", I think this falls under "be aware of symptoms and stop or weigh the costs if you get it, but you're probably fine".
Amusingly, this was a very common if not the most common stance on this very website some time ago. Surely, don't look for assistance on the internet regarding memory issues.
Of course deskilling will happen. But marketing says the machine is more often right than the operator is, and people also want it (we can replace docs with AI today, yadda yadda). Soooo... to be expected? It's just that the machine has to work correctly, which is not on the endoscopist, right?
...and the Racket syntax-parse macro system: https://docs.racket-lang.org/syntax/stxparse.html