Estimating causal direct and indirect effects in the presence of post-treatment confounders: a simulation study