The issue of bias due to unmeasured confounding when estimating association between a given outcome and an exposure has been (and it is still) of great interest in occupational epidemiology. The main concern is that personal risk factors are lacking on administrative data, such as smoking. Different approaches, that require different degrees of available information, have been developed to evaluate the impact of the bias due to the failure in controlling for unmeasured confounders.Methodologies considered in the analysis not only include models for complete individual data, but also the indirect control through aggregate data, adding Monte Carlo simulation for the uncertainty of the estimates as well as techniques that do not require any information.These techniques are applied to a population cohort and results are compared to evaluate how the estimates vary across the different scenarios that have been proposed to adjust for confounding effect of smoking habit.