It says "Among 508 (12%) patients". 508 is 12% of the whole 4226, not just the 2449 with known ages. 20% (102) of those 508 were aged 20-44.

It's not known how many of the 4226 were aged 22-44, but if it's roughly the same as the 29% of known ages, that would be 1226.

102/1226 is 8.3%.

I hate to say it, but it seems the table at the bottom is overestimating the proportion of hospitalizations by ignoring 1777 patients who have an age even if it's unknown, instead of making an estimate for them.

The reason that the unknown ages were excluded from Figure 2 is because the hospitalization status is also unknown. You assumed that 0 of the unknown cases were hospitalized. Instead, you could also assume that hospitalization rates for the unknown cases are roughly the same as for the known cases. If you do that you actually get the higher of the 2 numbers in Figure 2's range. So for the 20-44 age group, that's 20.3%. If you instead assume that people with a known age, but an unknown hospitalization status were not hospitalized, then you get he lower number in the range, 14.3%.

It's a very fair point that any of these assumptions could be inflating the rates. There are a whole bunch of other factors around asymptomatic, what the eventual outcome will be in the cases that are early-stage and who gets the limited number of tests in the USA that makes it currently impossible to know the true infection fatality rate or infection hospitalization rate.

I see, you're right. What I wrongly assumed was that they were receiving spotty data about overall cases but more thorough data about cases leading to hospitalization.

I also just noticed the age brackets for the 508 known hospitalizations only add up to about 91%, so apparently 9% of those are also unknown age.

It's too bad they didn't publish this in a live form so we could see how it changes as more data comes in. Maybe they will do a follow-up using the same methods at some point.