So the referenced study prompts a basic but hopefully interesting question about statistics and epidemiology which I hope someone could answer. When they contr

So the referenced study prompts a basic but hopefully interesting question about statistics and epidemiology which I hope someone could answer. When they controlled for former problem drinking status and medical status (among other factors), there was a major change in abstainers from 200% increased mortality over moderate drinkers to 51%. So it seems like current abstainers included a significantly death-prone cohort who were former problem drinkers and/or in poor health. Okay, so that makes me wonder what the difference is between controlling for major factors and simply excluding the cohort with factors having that very significant impact on mortality. My question is, at what point should one correct for major factors and at what point is it appropriate to completely segregate the data? What if the size of the group with major factors excluded would be statistically insignificant (i.e. heavy drinkers who have never been problem drinkers)?