They've done those studies and context switching has historically been where the most problems occur. Whether they've repeated them with modern electronic medical records and systems, I don't know. I think most people agree there's probably a better middle ground between 8 hr shifts (3 handoffs a day) and the standards set by a dude who liked to experiment with coke and meth.

One of the big issues that I feel like doesn't get touched on as much is longer shifts allow less doctors, which reinforces the artificially low doctor graduation rates. The national board in the US pegs the graduation at X thousand new doctors every year and that number is mostly tradition / vibes. No we don't want to compromise on the ability of new doctors, but "gestures vaguely to US healthcare" good lord do we need more of them. Much the same could be said for nurses.

And all of that circles back around to not wanting to dilute traditionally higher paying job markets with more practitioners because the for-profit system will try to wring out every cent they can.

Yeh, same. Which is why I said ideally there would be 100% overlap with shifts. Always 2 doctors, offset by half a shift.
Like, that is the fix. Peer review of decisions, easy conference/council/whatever-the-word-is, context can be handed over better (outgoings doc/nurse briefs incoming doc/nurse while remaining doc/nurse listens & supplements)

But I have also been on gigs (I work in events) where there is a rig crew, a show crew and a derig crew.
When everything is meticulously planned out and everything goes according to plan with all the communications in advance, it works. It does. (As a tech, I'd rather set up the kit I'm using). If I know it has been set up according to pre-communicated spec then I can work it. If it deviates and I have been in the loop, I can work with it. But if it turn up and it doesn't make immediate sense then it is many times harder. If I am rigging kit without a clear concrete plan, then I am guessing what the tech wants.
And I also know 2 lampies can't co-light a gig unless they take turns.
Someone has to be incharge, someone has to take responsibility.

But I don't think (and from what I have read, and I'm sure I have been somewhat misinformed) that applies directly to healthcare. Meticulous plans don't exist. Every patient is different. Something minor reported and expected to go away on the last visit of the leaving doc that is then reported as slightly-more on the new docs visit... That could be significant. And a few extra hours on a shift could save a life, because of that easily dismissed/forgotten context/knowledge during a handover.

2 doctors at all times is the fix. Or, actually, a voice-to-text and an LLM.... Likely a decent usage of an LLM.
It doesn't need to know who/what the patient is. It doesn't need to know co-morbidities, existing conditions, medications, treatmens etc. Just that the doctor is interacting with patient A, and here is a summary.
Patent A is the same patient that a nurse interacts with.
Helps with hangovers and context.
Patient A is still in the hospital? Patient A still has a transcribed record that can be quickly summarised by a local (or onsite) LLM.
Using onsite LLMs is no different than using a database. And it doesn't have to be massive. 30m before a shift change, there can be a "notes after this time will not be summarised during handover so previous context can be summarised". So doctors only have to remember the last 30m during a handover, and the rest of the context (even transcripts) are provided to prompt their memory for a better handover. It's an information tool for doctors, not a crutch.
And now I sound like an AI shill.

Sorry for the wall of text. I've been drinking. I hate the "just use LLMs bro", but think they have genuine utility when applied safely and locally.
And I want doctors and nurses and janitors/cleaners/sterilisers/techs of hospitals to be treated like the fucking heros they are.