Clinical Trials

Clinical Trials: A Practical Guide ■❚❙❘What is the difference between interaction (or effectmodification) and confounding?Consider a study investigating the effect of vitamin A supplements on childhoodgrowth. Among children who are vitamin A deficient, it is likely that vitamin Asupplements will increase growth, while supplements may have no effect inchildren who are not deficient in vitamin A. This is an example of interaction (oreffect modification). This can be defined as a situation where the treatment effect(ie, vitamin A supplements) on the primary outcome (ie, height) varies accordingto the levels of a third factor (ie, the level of vitamin A before supplementation).In the evaluation of a clinical trial, confounding and interaction effects aretwo different things. Confounding is a nuisance effect that distorts the observedtreatment effect on an outcome of interest because the confounder is associatedwith the outcome and is unequally distributed between the treatment groups.We aim to control confounding in the design and analysis stages of a clinicaltrial to enable the true treatment effect to be estimated.Interaction is a real effect, independent of the study design, that causes the treatmenteffect to vary according to the level of a third factor, which we want to detect.Exploring the nature of interaction can be very helpful in understanding thebiological processes underlying an association between a treatment and anoutcome. Interaction is discussed further in the next chapter.ConclusionConfounding is the situation where the observed association between a prognosticfactor (such as treatment) and an outcome measure is made stronger or weakerby the imbalance of another factor. In a clinical trial, confounding factors canfalsely obscure or accentuate the treatment effect. Therefore, attention should bepaid to controlling for potential confounders during the design and analysis stagesof clinical trials.When designing a clinical trial, the most effective way to reduce the possibility ofconfounding is to employ a suitable randomization method. When analyzing thedata, unexpected confounders can be controlled for by compensating for theirimbalance, so that the net impact of the treatment effect can be estimated, whilecontrolling for the effect of these confounders.303