Smoking as "independent" risk factor for suicide: illustration of an artifact from observational epidemiology?