Edit: The authors of the study don't imply causation, but it's tempting to infer it yourself. Thus, my comment. OK. So this is a observation study, not an expe

Edit: The authors of the study don't imply causation, but it's tempting to infer it yourself. Thus, my comment. OK. So this is a observation study, not an experiment. Only correlation can be inferred, not causation. To make it clear why you can't, let me come up with a new, parallel study: Track everybody who self reports being told they have a genetic disposition to cancer by a doctor. Divide them by those that decide to go a cancer treatment center vs those who don't. The study would probably show that cancer treatment centers are associated with shorter lifespans. That makes sense, because you're not going to go to a cancer center treatment if you don't have cancer. But it doesn't mean that going to a cancer treatment centers are ineffective. Self selection in a study means you can't infer causation. You can only infer correlation. Note: I'm not actually weighing in on whether I think therapy is good or bad. I'm say the results of this study could happen either way. Edit: I think you could reasonably infer that therapy isn't enough whenever it is used. But you can't tell if not going to therapy would have been better than going to therapy for those who did - so if there's an action item from this study, it's to improve therapy. But I don't think you can unequivocally say: this study shows therapy is bad.