The need for randomization as a means of controlling confounders is accentuated in the study of intended effects (efficacy) as compared with unintended ones (toxicity). The basic reason is that the indication for intervention is inherently a confounder in the study of efficacy but not of toxicity, whereas contraindications represent only a minor confounder even in toxicity research. Moreover, control of the indication in non-experimental terms is commonly infeasible owing to the complexity and subtlety of the indication.