The causal association between a risk factor and a disease outcome is often informed by a wider body of evidence than epidemiological studies of RRs for specific measures of exposure, especially when disease-specific and age-specific RRs are needed. For example, although smoking is an established cause of cardiovascular diseases, when cohorts are analysed in fine age groups, the 95% CI for the effect of smoking on stroke spans 1·0 in several age groups.38 Similarly, randomised trials of zinc supplementation were designed to detect effects on total mortality.36,153 Re-analysis of the same trials for disease-specific outcomes, which is necessary to extrapo late effects to populations with different causes of death, reduced their statistical power and gave 95% CIs that spanned 1·0. To use the broad evidence while accounting for the uncertainty of the subgroup RRs, we included in the uncertainty analysis all draws of the RR distribution, including those that show a protective effect as long as the overall relation for the risk factor across all ages is significant. In other cases, if there are different degrees of exposure for a
