a. Study-level risk of bias assessments reported in the 2011 Cochrane update were used, except for two newly extracted risk of bias items where outcome-level assessments were more appropriate (blinding and incomplete outcome data). No studies had a high risk of bias due to sequence generation (11 were unclear and 6 low), or allocation concealment (6 were unclear and 11 low). Only two studies representing a small weighting in the pooled analysis (Lerman et al. and McAlister et al. ) had a high risk of bias due to incomplete outcome data (1 was unclear and 14 low). Ten studies could be considered to have a risk of bias due to inadequate blinding of outcome assessment (4 were evidently unclear and 3 were evidently low), but for these studies the accuracy of knowledge of outcome probabilities was generally assessed using objective a priori criteria, thus inadequate blinding of outcome assessment was not considered serious. Most studies of decision aids do not blind the personnel delivering the intervention, and risk of bias was rated down for this reason.
b. Inconsistency was not rated down since there was a uniform direction of effect with all studies favoring decision aids and a large proportion of the heterogeneity is explained by the variation in the control event rate.
c. Imprecision was not rated down since the confidence intervals for the pooled RR are uniformly greater than 1.25 (with greater relative effects predicted for lower control event rates), and this estimate is based on over 2,000 patients in each arm.
d. Investigation of reporting bias using funnel plots was not feasible for this outcome. Reporting bias was considered unlikely based on the thoroughness of the search and discussion provided in an earlier Cochrane update.