How are we sure about the direction of cause and effect here? I'd expect more healthier people to self-select the working cohort, all else being equal.

Yeah this seems like an obvious confounder.