Beyond intention to treat.

It is a special pleasure to review approaches beyond intention to treat (ITT) for an audience of epidemiologists. Experts in clinical trials are suspicious about departures from ITT because the typically hard issues of observational studies then surface in randomized trials (1). This pill is hard to swallow because experimenters primarily test for effects of exposures that can be prescribed (and marketed). They must avoid false effect claims that would introduce costly, useless, and possibly toxic interventions into society. Hence, one carefully designs randomized trials, constructs comparable groups, and draws robust conclusions based on simple contrasts between groups as randomized. Life would be that simple were it not for human beings. In practice, the actual exposures to treatments vary because of variation in adherence to treatment assignment and can deviate substantially from any assigned schedule. One would wish not to underestimate the effect of a prescription rule (good or bad) because it is evaluated on partially compliant patients. The cry to acknowledge the range of actual exposures and estimate their effects thus persists, but drug-regulating authorities fuel resistance. They know that vested interests are poor guides for the scientific method and that one leaves the trodden ITT path at no small price.