Some time ago, I read all of the published data on the safety of traffic circles, as I had often seen them recommended as a safer alternative to traffic lights.

I was not very impressed. The confounding factors for the traffic light to circle projects were always far too large, and it always had to be coupled with the fact that it was usually poorest intersections that get rebuilt. I no longer have a high opinion of traffic safety research as a field.

Regarding this article's claims in particular, traffic fatalities went up a certain amount after 2020, and I doubt that road design changes were a primary driver. (Policing trends are one possible candidate.)

Regardless, in the last six months or so, my Tesla's autodrive went from "expensive joke" to "safer driver than me". IMO, leveraging existing technology could cut traffic deaths to a small fraction of what they are now over 3-5 years, were we to pursue aggressive conversion of the existing vehicle fleet.