From the Methods section of the paper (under "study design"): "We also included meta-analyses that pooled data from primary prospective/retrospective cohort or case-control studies. These studies were the focus because of their ability to assess causality for observational research".

While "causality" is a strong word in the sentence above, the data is much stronger than simple correlation. Of course, outright causality has not been established, but the evidence to determine a predictive association is strong.

We only have correlation for lots of important results. Smoking cigarettes is one example. Experiments are almost always better, and a lot of correlational research is terrible. But knowing something for sure doesn’t always require an experiment.

That's a bit different.

There's no reasonable path to reverse-causality with smoking. "People prone to developing lung cancer are more likely to take up smoking" doesn't really make sense.

On the other hand, "Fat and unhealthy people are more prone not to exercise" is a reasonable reverse-causality position, for all kind of reasons. Your knees hurt, you're self-conscious, it's extremely uncomfortable, your fat jiggles, etc.

I'm not saying direct causality doesn't exist here, I'm just saying that reverse-causality is reasonable, whereas with smoking it is not. Assuming causation from this particular correlation is harder than it is with smoking.

"People who smoke are more likely to do other things that the medical establishment condemns" is a plausible non-causal explanation for the correlation between smoking and lung cancer.

I think that we just have to look at the best explanation for the current data to determine causality, but it’s always a guess.

That is a plausible explanation for a present correlation, but I think less so for when the correlation was first noted.