Case fatality rate includes all known cases, whether hospitalized or not, but only cases where we know whether they recover or whether they die. You want an idea of how deadly a disease is overall, not just in the worst cases.

Think about it, if you had a million people infected and then got over some bug and only one elderly, frail person who became hospitalized and then died, the case fatality rate would be 100% if we only consider hospitalized people. That would seem misleading, doesn't it?

Now, where you'll find wiggle room in the definition is whether you only include laboratory confirmed cases or whether you go with a symptom-based one. In a fast developing epidemic, there are just too many cases to send to a lab, so you just use symptoms. So far, China seems to only be reporting laboratory confirmed cases to the press, and it seems like all cases are being hospitalized for now, even the 4-year old case who doesn't seem sick at all.

Up to 77 confirmed cases, last I read. Half or more of the cases (depending on who you ask) don't seem to have any direct connection to live poultry. So in my mind, either the disease vector is something else besides chickens, like pigeons, or else many of these people are catching it directly from another person.

The very limited number of genetic sequences released by the Chinese so far supports the theory that person-to-person transmission is already occuring in some cases. There are samples from infected people which share genetic markers that don't occur in any of the bird samples released so far. This suggests that the virus was acquired from another human rather than from a bird.