Control Groups in Microbicide Trials: In Defense of Orthodoxy

To the Editor—The Journal recently published 2 commentaries on the design of trials testing the effects of microbicides against HIV and other venereal infections. Both commentaries propose that, in addition to a blinded-control group, a second control group—which would remain “unblinded” as well as untreated—should be included within such trials [1, 2]. The additional control group—named “condom-only” or “non–gel-using”—would undergo the same recruitment procedures, follow-up, and counseling that are used in the other arms but would not be subject to further intervention. We think that this approach is mistaken. The design retains one fundamental requirement of such trials—namely, random assignment—but abandons another—namely, “blinding” [3]. We foresee 2 problems and, to address them, bring forward 2 arguments The first problem relates to the standard against which protection is to be judged: a placebo, if not entirely inert, might give as much as or more protection than does the putative microbicide. Although comparisons between blinded groups are ordinarily sufficient to support conclusions about treatment effects, a result showing no difference could be ambiguous and allows 3 possibilities: both treatments could be equally effective, equally ineffective, or equally harmful—hence the justification for including an open “non–gel-using” arm The second problem stems from the considerable protection that proper condom use confers against sexually transmitted infection. Ethical research requires that investigators counsel at-risk women to insist on condom use in all sexual encounters. Should the women truly conform, reduced frequency of infection will render detection of microbicidal effects less likely. Reported condom use inevitably leaves actual use uncertain; moreover, among “gel-using” groups especially, differential misreporting may well conceal reduced condom use. To protect against such bias, both commentaries again recommend the use of an unblinded condom-only group as a “true” control, the better to represent the “real world.” We submit that an unblinded group can serve neither as a suitable substitute for a blinded control nor as a true representation of the real world With regard to the first issue—the use of comparison groups in controlled trials—all investigators well know that this standard aims to keep differences between the test group and the control group(s) inapparent, both to the research teams and to the subjects; long-established and virtually ironclad rules insist on such double blinding, to avoid bias. By definition, an open arm violates this standard. Thus, open-arm subjects will surely be exempted from counseling and questioning routines on use of the prescribed gel. An open condom-only arm not only makes the disparity vis-à-vis the blinded-control and treated arms obvious to all investigators but can have untoward consequences. Thus, some field workers have found it more difficult to maintain subjects’ participation in a condom-only than in a blinded arm. Two concerns therefore remain unresolved: the intervention is not identical across groups and less-complete follow-up in open arms than in closed arms prejudices randomization and reduces statistical power With regard to the second issue—representativeness—normal research procedure almost invariably requires selected participants. In the situation discussed by the 2 commentaries, it is clear that the populations are selected in at least 4 respects: they have a limited age range, are seronegative for HIV, are active sexually (and prefer to avoid conception), and readily conform to a demanding procedure. Therefore, the sample does not necessarily fully represent the population that yields such subjects For the selected subjects, of course, awareness of their treatment status could influence their actual as well as their reported behavior, both of which are critically pertinent to the outcome of the trial. This is precisely a situation in which it is crucial to sustain blindedness among all participants. Blinded trials of microbicides are specific and particular in intent—they are a beginning and not an end. They test afresh, on the human scene, what has already been tested in the laboratory. They aim to detect the differences in effect between preparations, not to resolve speculation about what might happen later in one or another community in the real world. Some describe this stage of microbicide testing as “proof of concept” (a stage that necessarily precedes licensing considerations, which must furnish their own criteria [4]) The choice of suitable controls is not solely an academic matter. Coplan et al. [4] are obviously correct to be concerned about the burdens that the use of an unblinded-control arm adds: it increases the number of subjects who must be recruited; it increases costs; it requires more-extensive analysis; and it requires both that the duration of the trial be prolonged and that the implementation of treatment be deferred. Multiple trials are under way, and they enhance the importance of economy in all senses. Although Gross has published a strong criticism of such multiplicity [5], advantages may derive from it; because test populations and their circumstances vary, multiple trials can be of benefit by either repeatedly reinforcing or repeatedly disputing the general validity of particular results. Some trials test rather similar preparations, and several wisely use the same placebos