This type of statistical associations on 'non-randomized samples' are equally indicative of causal role of a substance X (tobacco smoke) in disease D (lung cancer), as it is for protective/therapeutic role of X for D or D's precursors/causes. For example, people taking high blood pressure meds are statistically more likely to suffer heart attacks and strokes than those not taking these medications. This association doesn't mean that these meds cause heart attacks and strokes. In fact they protect against underlying causes of these very same events with which they associate (that's what gives rise to this association).
Since tobacco smoke happens to have potent anti-inflammatory effects and it also upregulates key detox and antioxidant enzymes (nearly doubling glutathione, catalase and SOD), one would expect it to be statistically associated with variety of toxic, including carcinogenic, exposures since it provides relief against the effects of such exposures (e.g. doubling the detox rate and reducing inflammatory reactions).
While there were no genuine randomized smoking experiments on humans, there were thousands such experiments on animals but you will never hear about these findings since they all went the "wrong way" -- smoking animals get fewer cancers and live ~20% longer than non-smoking animals, while staying thinner (by 15%) and sharper into the old age.
You can find references and discussions of the above facts in the longevity/nootropics forum (longecity) in a thread "Smoking is good for you" (I post as 'nightlight'). The TOC of the thread topics is in another post on that site.
Sonds like the study done by the World Health Organization that you won't find bandied about by any anti-smoker, even though it's the largest study on smoking ever done.
I'll give you one guess on that outcome.