I don't think this paper actually disentangles the causation direction. It's just another study showing that people with mental problems tend to like smoking. The 20 year gap between starting smoking and hospitalization they're talking about leaves a lot to be desired.

The most surprising thing to me is that they managed to find an n=131,140 of people in the system with enough genetic testing coverage to calculate the polygenetic scores for the smoking models they used. That many people had detailed gene sequencing done and in accessible public databases?