The plight of the many patients with human immunodeficiency virus (HIV) infection is heart-rending to all who care for them. The desire of the patient, the patient's family, and the doctor to do something can be overwhelming. Frustration about the lack of a preventive vaccine or curative treatments has led to demands that new therapies for the acquired immunodeficiency syndrome (AIDS) be exempted from the usual process of rigorous scrutiny used in the evaluation of new drugs. Compassion combined with heavy lobbying pressure by activists has led to the granting of exemptions. What changes in drug testing have been granted and how do the new policies affect the evaluation of treatments for HIV infection?

Early release of drugs has been an important demand of activists [1]. Because of the desperate cry for new therapies, drugs have been rushed to market even before they have been fully tested for both efficacy and safety. Another modification of established policy was the move to conduct "community" studies without requiring randomization [2]. Patients who elected to participate in trials could specify their own drugs. The rationale for these modifications was that the risks of early release would be countered by testing the drugs as they were being used by conducting nonexperimental trials as patients began to use the new drugs. Thus, patient experience would be observed and recorded in a standardized way. The usual randomized controlled trials would run parallel to the nonrandomized studies, so the more rigorous design could also be undertaken even while the drug entered into common use.

Fischl and colleagues [3], who, in this issue of Annals, report a study comparing the efficacy of zalcitabine and zidovudine in patients who had already taken zidovudine, may have fallen victim to this compassionate policy. The authors aimed for 160 patients in each arm of their trial and wound up with 59 and 52 patients in each arm, about a third of their projected total. This reduced the power of the study so that the authors could have detected only a very large difference between the two treatment groups. Thus, a modest but both statistically and clinically important difference between the two treatments stood a good chance of being missed in this too small study. It would have been helpful if the authors had discussed the reasons for the failure to achieve the intended size of their study and if they had provided the reader with an explanation of the reduced power of the final study (now one third the size of that intended, rendering their stated power calculations inoperative).

Many believe that the ready availability of drugs discourages entry into randomized trials. Once a drug becomes available, it may be thought of as being of proven value, so why enter a trial? And if one must enter a trial for the chance to use a new drug, why should it be a randomized trial, where the participant relinquishes the choice of a specific treatment, when one could as well enter a trial in which patients themselves choose their treatments. Fischl and colleagues [3] speculated that the "broad availability of didanosine (ddI)" may have led to both slow accrual and early discontinuation of the two study drugs in the trial, but they present no data to support this speculation.

The Editors of Annals published this admittedly too small study because it represents some of the best of limited data currently available about the relative efficacy of zalcitabine and zidovudine in patients who have already completed a course of zidovudine. These drugs are already being obtained through drug clubs and are commonly used in this way in the absence of strong clinical research that they do more good than harm.

The trend toward abandonment of randomization in large community-based trials represents a lowering of research standards. It is hard to exaggerate the importance of randomization in comparative drug trials. When each eligible patient has an equal chance of being randomly assigned to either one of the two arms of a trial, prognostic factors such as stage of disease, duration of illness, presence of opportunistic infection, and age tend to be evenly distributed between the two groups. Any difference between the groups found at the end of the randomized trial can be attributed to differences in treatments rather than to differences in patient characteristics or stage of disease. Even those prognostic factors that are unknown or unmeasured will tend to distribute evenly between the two groups if randomization is used. When the Food and Drug Administration and the National Institutes of Health agreed to nonrandomized designs, they may have given away too much. Similarly, much is lost when patients in randomized trials who, in an effort to maximize their chances for a new and promising treatment, pool their drugs with those given to other participants in their trial and dismantle the scientific strength of the study. An additional problem arises when patients with AIDS use drugs available from underground sources [4, 5].

Many years ago, before we knew how to do randomized controlled trials, another treatment for a sexually transmitted disease was widely used and believed in without strong scientific evidence that it did more good than harm. Malariotherapy was the acknowledged treatment of choice for central nervous system syphilis for 20 years and earned its discoverer the Nobel Prize in Medicine in 1927. But, to this day, there is no way to know from the many published papers reporting results of this treatment whether it ever really worked. We are unable to interpret these early studies (published from 1917 to 1940) because none of them was randomized; thus, it is not possible to attribute the improved outcome reported for malariotherapy to the actual treatment. It may have been that healthier patients (who were likely to withstand the high fevers the therapy induced) were given the new treatment, which would account for the enthusiastic reports of success [6, 7].

Is history about to repeat itself with this new and dreaded sexually transmitted disease, AIDS? We have the methods to answer the important questions about new therapies for AIDS, but, because of misguided compassion, we may fail to use them. Early release of untested drugs and nonrandomized comparative trials may result in the medical community never learning enough about the new drugs to use them properly and understand their true value. These policies may retard the development of better or more efficacious drugs.

Some good has come from these new policies. Patients are more involved in how trials are carried out—for example, in identifying participants and assuring informed consent. Some treatments have been placed on a fast track and have moved through the approval process more quickly, without sacrifice to the usual scientific concerns for safety and efficacy. However, studies that are too small to answer the posed question and studies where the groups compared are subject to the possibility of striking biases do not advance knowledge. When the drug regulatory agency decided to exempt AIDS from the usual legal requirements for demonstrating safety and efficacy, it may have set back—rather than advanced—the search for better therapies for AIDS.