Every clinical trial must be conducted according to a protocol that is written and approved before the study starts. The most satisfactory protocols are those that are designed with the collaborative effort of a team of experts representing various disciplines. The trial protocol should include a justification for the trial, and should clearly define the question that the trial is designed to answer. The study population must also be clearly defined, indicating both inclusion and exclusion criteria and the procedures to be used for recruiting study participants and allocating them to various treatment protocols. Study patients should have confirmed HIV infection, either asymptomatic or early symptomatic; in most cases, children and pregnant or lactating women should be excluded. Patients may be recruited from voluntary testing centres and from clinics treating either AIDS or other sexually transmitted diseases. Appropriate informed consent must be obtained from each patient, and each patient should have the opportunity to receive appropriate counselling. The protocol should define appropriate clinical monitoring to detect toxicity as well as to determine efficacy and a plan to deal with drug toxicity should it occur.

An accurate record must be kept for each patient in the study, which should include documentation of informed consent, a medical history, details of treatment received and succinct reports of all physical examinations, follow-up evaluations and laboratory test results.

Efficacy should be judged on the basis of such defined end-points as specific clinical symptoms or signs, the development of particular opportunistic infections, or defined prognostic laboratory markers. Safety should be monitored on the basis of either symptoms or signs, particular attention being given to end-points that may signal forms of toxicity that might be anticipated. Laboratory indicators of liver, renal, cardiac and haematological toxicity should also be monitored.

Evaluation of the trial should be undertaken using appropriate statistics.

Ideally, the study design should be blind, randomized and placebo-controlled. A cross-over design may present problems in interpretation of study results, both because of uncertainty concerning the time course over which a drug may act and because a patient’s status may change during the two phases of the study.

Every effort must be made to address the problems concerning preparation, quality control and dose standardization for herbal preparations, and to find a satisfactory placebo.