Figure 5 Only omission of the Olsen study changed the estimate of effect size. Because the Oresland study was an outlier, we repeated the sensitivity analysis with a fixed effects model. The Oresland study did not produce a significant change in the estimate of effect size, as it contributed relatively little to the final weighted estimate (1.2%). Only omission of the relatively large Olsen study produced a substantial shift in the effect size estimate, with a resulting relative risk of 2.82 (95% confidence interval 1.94, 4.12).