Improved estimation of controlled direct effects in the presence of unmeasured confounding of intermediate variables. Academic Article uri icon