Table 1.

Policy summary

Background	There is widespread evidence that smoking is a leading behavioral risk factor for numerous health outcomes; despite this evidence, smoking remains a persistent public health challenge. Although there is substantial research on the subject, few attempts have been made to evaluate the dose–response relationships between smoking and its many potential health outcomes in a systematic and comprehensive way or to assess the strength of the evidence for these relationships. Due to limited assessments of this kind, current guidance on and regulation in response to the risks of smoking are often based on expert evaluation of heterogeneous evidence. Expert committees are valuable in large part because they are able to consider the nuances of the available evidence, but they are also inherently subjective. In the present study, we used a systematic review approach and employed a meta-analytic framework29 to assess all available evidence in a way that did not force log(linearity), more fully incorporated between-study heterogeneity into estimates of uncertainty and addressed many of the other limitations of existing meta-analytic approaches.
Main findings and limitations	Even when between-study heterogeneity and other forms of uncertainty were incorporated into our assessment of risk—leading to a very conservative interpretation of the available data—smoking had a strong-to-very-strong (>50% increase in risk) association with 8 of the 36 outcomes selected for this analysis. A further 24 outcomes had weak-to-moderate evidence of an association (including 1 outcome for which smoking was protective), whereas 4 had no evidence of an association once between-study heterogeneity had been incorporated. On a 5-star scale with 1 suggesting no evidence of association (no increase in risk) and 5 very strong evidence of association (>85% increase in risk), the 8 highest pairs received 4–5 stars, the next 24 received 2–3 stars and the final 4 received just 1 star. These findings confirm that smoking is irrefutably highly harmful to human health.
	Limitations of the present study include the fact that the bias covariates we used were based on observable study characteristics and thus may not fully capture characteristics such as study quality or context; our approach to selecting which risk estimates from a given study should be included, if multiple estimates with different adjustment levels were reported, limited our ability to make full use of all available risk estimates in the literature, and we did not test for additional forms of bias such as whether studies are more consistent with each other than expected by chance.
Policy implications	The available evidence demonstrates that smoking is a highly harmful risk factor for a wide array of serious health outcomes, most notably laryngeal cancer, aortic aneurysm, bladder cancer, lung cancer and other pharynx cancer (excluding nasopharynx cancer). Policy makers should pay particular attention to the 5- and 4-star smoking–outcome pairs, for which the evidence of an association is strongest, but should not ignore lower starred pairs, particularly those with a high prevalence or severity of outcome. Our findings further validate previous conclusions about the high risk of smoking and, using our meta-analytic methods, we offer the added value of minimizing the chance that risk has been overestimated. Policy makers can therefore be confident that smoking recommendations and regulations made based on the findings from the present study are not unnecessarily restrictive.