Confounding or intermediate effect? An appraisal of iatrogenic bias in perinatal AIDS research.

Many epidemiological reports present estimates of exposure effect adjusted for a host of variables thought to be risk factors for the disease. The term iatrogenic bias is proposed to denote bias introduced by the analyst when inappropriately controlling for variables as though they were confounders. The focus of this commentary is iatrogenic bias resulting from the control of variables intermediate in the causal pathway between exposure and disease. Perinatal epidemiological studies are particularly vulnerable, and iatrogenic bias will typically mask or diminish the true effect of the exposure. The problem is illustrated with three examples from the field of perinatal AIDS epidemiology. It is recommended that researchers carefully scrutinize each variable considered for adjustment and use path diagrams to sort out causal pathways. In some instances, it may be most appropriate to present both adjusted and unadjusted estimates of effect.