My opinion: what would be nice is a robust set of studies that you can do a sort of meta analysis of. Ideally it would include measures that _decrease_ drunk driving offenses too, although I admit that it is unlikely to find a good example of that in today's world.

I do think that one study that shows a robust correlation is a good start, though. That would be the starting point for trying to decide what the optimal tradeoff is.

The thing is though, I think it's a matter of diminishing returns. I bet if you made the penalty for drunk driving death it would reduce drunk driving further, but I don't think many would agree that is a measured and reasonable response to the offense.