In clinical trials, random assignment of patients or volunteers to the control group or the group receiving the experimental therapy is the optimal method for distributing the known and unknown variables that affect outcome between the treatment and control groups. Randomized clinical trials may be impossible to use studying all experimental therapies; for patients who cannot—by regulation, ethics, or both—be studied with this design (e.g., children or fetuses) or for disorders with a typically fatal outcome (e.g., rabies), it may be necessary to use historical controls.
Concealing participant assignment is referred to as blinding or masking. Participants in the control group will receive an inactive replica of the drug, a placebo. In a single-blind study, participants are blinded to treatment assignment, but investigators are not. In a double-blind study, neither the participants nor the investigators know whether the active agent is being given. Blinding the investigators not only removes bias in patient management and outcome interpretation but also eliminates selectivity in the enthusiasm for therapy typically conveyed by clinicians. By eliminating participant and observer bias, the randomized, double-blind, placebo-controlled trial has the highest likelihood of revealing the truth about the effects of a drug. This design permits evaluation of subjective end points, such as pain, that are powerfully influenced by the administration of placebo. Striking examples include pain in labor, where a placebo produces ~40% of the relief provided by the opioid analgesic meperidine with a remarkably similar time course, angina pectoris, where as much as a 60% improvement in symptoms is achieved with placebo, and depression, where the response to placebo is often 60-70% as great as that of an active antidepressant drug.
The existence of a therapy known to improve disease outcome provides an ethical basis for comparing a new drug with the established treatment rather than placebo. If the aim is to show that the new drug is as effective as the comparator, then the size of the trial must be sufficiently large to have the statistical power needed to demonstrate a meaningful difference. Trials conducted against comparators as controls can be misleading if they claim equal efficacy based on the lack of a statistical difference between the drugs in a trial that was too small to demonstrate such a difference. When trials against comparator drugs examine the relative incidence of side effects, it also is important that equally effective doses of the drugs are used.
A clear hypothesis should guide the selection of a primary endpoint, which should be specified before the trial is initiated. Ideally, this primary endpoint will measure a clinical outcome, either a disease-related outcome, such as improvement of survival or reduction of myocardial infarction, or a symptomatic outcome, such as pain relief or quality of life. Examination of a single, prospectively selected endpoint will most likely yield a valid result. A few additional (secondary) endpoints also may be designated in advance; the greater the number of such endpoints examined, the greater the likelihood that apparently significant changes in one of them will occur by chance. The least rigorous examination of trial results comes from retrospective selection of endpoints after viewing the data. This introduces a selection bias and increases the probability of a chance result; retrospective selection therefore should be used only as the basis to generate hypotheses that then can be tested prospectively.
Therapeutic decisions sometimes must be based on trials evaluating surrogate endpoints— measures such as clinical signs or laboratory findings that are correlated with but do not directly measure clinical outcome. Such surrogate endpoints include measurements of blood pressure (for antihypertensive drugs), plasma glucose (drugs for diabetes), and level of viral RNA in plasma (for antiretroviral drugs). The extent to which surrogate endpoints predict clinical outcome varies, and two drugs with the same effect on a surrogate endpoint may have different effects on clinical outcome. The effect of a drug on a surrogate endpoint may lead to erroneous conclusions about the clinical consequences of drug administration. In one compelling example, the CAST study showed that—despite the ability of certain antiarrhythmics to suppress ventricular ectopy after myocardial
Was this article helpful?