A well-controlled study permits a comparison of subjects treated with the new agent with a suitable control population, so that the effect of the new agent can be determined and distinguished from other influences such as spontaneous change, "placebo" effects, concomitant therapy, or observer expectations. FDA regulations cite five different kinds of controls that can be useful in particular circumstances:
- placebo concurrent control
- dose-comparison concurrent control
- no-treatment concurrent control
- active-treatment concurrent control, and
- historical control
No general preference is expressed for any one type, but the study design chosen must be adequate to the task. Thus, in discussing historical controls, because it is relatively difficult to be sure that historical control groups are comparable to the treated subjects with respect to variables that could affect outcome, use of historical control studies has been reserved for special circumstances, notably cases where the disease treated has high and predictable mortality (a large difference from this usual course would be easy to detect) and those in which the effect is self-evident (e.g. a general anesthetic).
Placebo control, no-treatment control (suitable where objective measurements are felt to make blinding unnecessary), and dose-comparison control studies are all study designs in which a difference is intended to be shown between the test article and some control. The alternative study design generally proposed to these kinds of studies is an active-treatment concurrent control in which a finding of no difference between the test article and the recognized effective agent (active-control) would be considered evidence of effectiveness of the new agent. There are circumstances in which this is a fully valid design. Active-controls are usually used in antibiotic trials, for example, because it is easy to tell the difference between antibiotics that have the expected effect on specific infections and those that do not. In many cases, however, the active-control design may be simply incapable of allowing any conclusion as to whether or not the test article is having an effect.