I doubt there are 400,000 papers worth reading in all of medicine. A major skill in research is understanding what to read, and how carefully.
Prepping to write a review usually involves about 150 articles, for me. About 50 of them just need a cursory look, and the abstract often suffices. Another 80 or so need some more attention; you might be looking for patterns in the methods used, cross-referencing results, figuring out what labs are doing what, tracking authors over time, etc. This is how you understand the state of a field.
The last 20 or so are the bulk of your time. A few are genuinely excellent and will be a highlight of the review. Some are simply solid work that requires a good amount of attention. A fair number are terrible and highly misleading. These are the papers that look okay, but have some flaws that need to be addressed by the field at large. It takes some tact to write about these, but it's usually not too bad.
Finding good authors, especially ones that write good reviews, can get you up to speed in a field in an afternoon. An evening chasing down citations and digging a bit on your own, and you're probably about where Bill was with this. Since Alzheimer's is a pretty big subject requiring familiarity with a variety of topics, I'd give it two weeks.
Very important studies are usually unexpected - that's part of what makes them important so we're verging on a tautology, there. So it's a matter of hindsight, in every field which contemporary articles are in fact "the classics." Your certainty and trust in current authorities broad knowledge and open-mindedness is a product of not knuckling down to read the abstracts of a few hundred thousand unrecommended journal articles - as well as a profound ignorance about the history of medicine.
That history more than suggests that if you want quality>authority (of contemporary publications) you'd better start reading, and reading nearly everything.
400,000 hours = 50,000 days (at eight hours a day) = 208 years (at 5 days a week, 48 weeks a year)
If the abstracts wern't really read and understood, just skimmed, perhaps it would be possible to do in a tenth of the time, so 21 years.
In other words, there's a LOT of dross, but there are real gems lying there unread and uncited, too.
So only one in ten or a hundred articles might take an hour or more.
Average for all but the vital or good articles (after accumulating a good background and vocabularly), when I timed it back then: 100 an hour without losing any significant information. The vital remainder of 100 abstracts viewed took me (more of a guess, now) on average not more than double that amount of time, again; perhaps less time. That varied a lot because I was looking most keenly for overlooked knowledge(studies) and pairs of studies that were significant only once linked. Therefore I sometimes was willing to follow tangents to tangents just in case and could speed up at that point because it was nearly all dross.
Back when I was in university supposedly studying the history and philosphy of science and language, I had to visit medical libraries and use the CD disks from the then-private company Pubmed, and walk back and forth to a large, printed medical encycopedia chained to a pedestal and the desk-sized CD reader Pubmed provided. I don't include that reading as I took no surviving notes and wasn't counting abstracts way back then. Being able to research online when I returned to medical reading a couple of decades later was an immense relief. Switching tabs to get to a medical dictionary isn't nearly as much exercise, but it is a lot faster. Printed medical dictionaries for professionals have a LOT of pages to turn, in addition to the walking.