Choosing cases and controls: the clinical epidemiology of "clinical investigation".

Choosing a suitable "control" has never been an easy job.One immediate difficulty is in the concept itself (1).The word "control" has been applied for many different things beyond the standard idea ofcomparison.They include the idea of regulation-such as a controlled environment, a well-con- trolled blood sugar, or quality control in a measurement process.The ideas also include an antecedent baseline-such as the control value or control period that precedes the subse- quent values in a before-and-after set of observations.Statistically, the word control can denote an analytic activity, such as the partition of control strata or the variables that are con- trolled in a multivariable analysis.In most types of medical scientific thought, however, control refers to something chosen as the comparison for the prin- cipal agent that presumably leads to the outcome of a cause- effect relationship.For the comparisons of either therapeutic or etiologic agents, a placebo may be the control for active treatment, medical therapy for surgery, unexposed nonsmokers for exposed smokers.The controlled comparison may ex- tend into a "bioassay" or "dose-response curve" if the effects of higher doses or durations of the active agent-smoking, alcohol, pharmaceutical substances-are contrasted against the effects of lower doses or none.This straightforward scientific concept of control is easily converted into a research design if the investigator can govern (i.e., control) the assignment ofthe compared agents.This type of governance has always occurred in laboratory experiments with animals or inanimate substances.The governance has also occurred in experiments with people during the past four decades when randomized clinical trials have been used to test therapeutic agents.In these governing circumstances, the in- vestigator decides which active and control agents will be compared, arranges to assign them appropriately, and then carries out the follow-up observations.In many research studies of people, however, the investi- gator does not have the opportunity to construct an experimental test of the active agent.If it is suspected of being a noxious cause of disease, randomized trials will usually be unfeasible and often unethical.Few people will volunteer to participate in a randomized trial of smoking vs. nonsmoking, or to be exposed to possibly hazardous chemical agents.If the research is concerned with pathophysiologic effects rather than