I don't like hours played as a measure of review.
Which is why it'd be per review (as in, total number of hours across every user tracked, not just those reviewing) or along those lines, not total hours invested by the userbase or whatev'. It's not supposed to be a metric of anything resembling worth, just looking for outliers on a user/playing scale, or (in the case of the word count thing) individual investment in review. Things with abnormal stats on stuff like that. It would indeed spit out passive stuff or things folks leave on in the background for whatever reason or any number of other things that mean little, but... that's fine. You'd just be using it to check for oddities, after which you can filter for your own preferences.
Basically, it'd be neat to see if the amount of time the userbase as a whole plays before someone reviews, or the amount of (lazily measured) effort any particular reviewer is putting into their spiel, is out of whack (either way, mind you) for whatever reason. If you're already doing stuff like the mentioned article did, not much reason save the effort to not plug in a bunch of other variables and see what pops out, y'know?