I don't know what's correct here.
Here's the discussion from Panda's Thumb where I first read that. It doesn't look like they were able to come up with a good answer to the question either.
A snippet:
The problem with the figure Garrett [in this article] cites is that it kills ~55% of the cases we know about. This is a classic case of sample bias. Those who are most sick (and thus, most likely to die) are also most likely to go to a hospital or clinic to be examined—and therefore, are also the most likely to have a clinically-confirmed case of influenza due to the H5N1 strain. Hence, the mortality data we have for H5N1 only comes from this sickest segment of the population—artificially raising the mortality rate. Puzelli et al.’s study, then, is timely due to the fact that it shows that sub-clinical infection with avian-type influenza viruses does occur (in almost 4% of their cohort of poultry workers).
For someone versed in influenza evolution and epidemiology, this is both disturbing and unsurprising. First, a bit of background on influenza virus. The virus has an RNA genome, which is segmented into 8 parts. These parts can re-combine and make a novel virus, containing a few segments of each of the parent viruses. In some cases, this progeny virus may be better adapted for a new host than one (or both) of the parent viruses.
An example. Say a human currently is infected with a (human) influenza virus. Let’s say it’s one of these folks in the Italian cohort of poultry workers. Maybe he’s not sick yet, so he goes into work, where he’s exposed to one or more types of avian influenza virus. If he becomes co-infected, the two types of virus can mix, possibly producing a “humanized” avian influenza virus, which may be transmissible between humans while the parent avian virus was not. This is the stuff of pandemics, and the kind of thing that keeps influenza researchers awake at night.