Data diving at the FRA gives us this set of data for passenger miles and passenger injuries from 1976 through 2011 which I've thrown together into a simple chart for ease of visual use. A couple of quick caveats: The absolute number is not terribly high: 2007 was the worst year on record with 1,514 total injuries, but as an absolute number that's still fairly low. Furthermore, fatalities are on a decreasing trend, which is a plus. This data is also not necessarily comparable with European or other foreign data as a passenger injury may be from any cause so long as it happened on a train or in a station. A mugging on the platform would qualify as a passenger injury as a result. There's also some occasional screwiness with the FRA's figures, possibly due to typos that get entered into the system.
As data is in terms of millions of passenger miles per injury, higher is better.
What's the cause? No idea I'm afraid, but it does seem like something that warrants further investigation