In model-based health economic evaluation, uncertainty analysis is often done using parametric bootstrapping. This requires specifying probability distributions for the model variables that are uncertain.
The effect size of the intervention is often expressed as a relative risk, and the standard assumption for a relative risk is that it has a lognormal distribution with the natural log of the relative risk and its standard error as parameters. The problem with this assumption is that the mean of the bootstrap draws from the lognormal distribution is always higher than the relative risk.
This article looks at two ways to correct for this effect and discusses their advantages and drawbacks. Both methods return a bootstrap mean equal to the relative risk, but the first returns an uncertainty interval that is narrower than the corresponding confidence interval, although the second method retains the corresponding width.
The article concludes that the second correction method is preferred.