To take your second example: most drunk drivers get home safely. In fact, I'm guessing that something on the order of only one in ten thoiusand drunk driving sectors ends with anything but an arrival at home.
If you want to get causality and correlations of drunk driving with car accidents you have to get a proportion of drunk drivers rides in which the drivers are killed themselves (and/or take the opportunity to kill others) from the whole number of drunk driver rides.
Then you have to compair it with the proportion you get of "all within the rules" accidents (i.e, excluding drunk driver AND OTHER ACCIDENT HIGHLY CORRELATED CAUSES) to the whole number of "all within the rules" car rides.
It works... Believe me !!