The hypothesis to be tested, and to whom the results will be applicable, must be clearly defined. For example, is the trial to test an intervention that could be followed in a real-life situation, or is it so extreme that it could only be followed by volunteers in a well-controlled laboratory environment? Both types of study are valuable, but in different ways. In the latter case, the outcome would give us precise information about the effect of the intervention on short term outcome measures. However, it would not enable us to make any recommendations to the general public. If we want to know what effect an intervention is likely to have on public health, we must use an intervention that can be followed by free-living individuals over a prolonged period.
It is essential to draw up criteria of eligibility for the trial. The aims of these criteria are to identify a group of people in whom the influence of various extraneous factors can be avoided. Thus subjects in a restricted age group may be examined, leaving aside children or the elderly. If the intervention involves providing treatment for a disease, some method of assessing the extent of the disease is required to ensure that cases of similar severity are involved. Also, if there are known factors that influence a subject's response to treatment, other than the severity of the disease, it is important to take these into account in defining the eligibility criteria.
Subjects should be allocated to the intervention and control groups at random. This can be done using prepared sealed envelopes. When a subject has been ascertained to be eligible for the study and has agreed to take part, the next envelope is opened and the subject allocated to the group specified within it. Randomising subjects to intervention and control groups should ensure that the two groups are similar in all respects other than the intervention being tested. If subjects are not randomised to the intervention and control groups, one cannot be confident that the two groups are similar.
If a crossover design is used, subjects should be randomly allocated to the intervention and control groups for a defined time period. The two groups should then be transposed. This enables each subject to act as his/her own control, thereby eliminating differences between individuals. It is, however, important to check whether there is an order effect, i.e. if the intervention period is first, is there a residual effect on the control period that is not seen when the intervention period is second? If subjects are not randomly allocated to intervention and control groups, there is a possibility of unconscious bias or of systematic changes over time, e.g. seasonal variation or laboratory drift.
It is necessary to ensure that all the relevant baseline data are collected on every subject. This is to:
The treatment regimen should be standardised. The aim of this is to reduce the uncontrolled variation in the treatment given to individuals in the study. This is particularly important if the intervention is given by several members of staff or if the study is being carried out in several centres.
Dietary trials have the complication that it may be impossible to alter the intake of one foodstuff or nutrient without affecting the intake of others. Such trials, conducted in real life situations are used to measure the effect of the intervention "package" rather than a specific nutrient. For example, if people are advised to increase their intake of bread, they may eat more butter or other spread. It is important to monitor intake of both the intervention and control groups during the trial. This serves as a measurement of compliance with the intervention as well as enabling evaluation of other dietary changes.
Assessment of outcome measures must be unbiased and accurate. If the observer knows to which group a subject has been allocated, this may influence the results of the study, e.g. blood pressure, plasma cholesterol, or even the diagnosis of myocardial infarction. If the subject believes that they are on the most effective treatment, this may also influence the results of the study, e.g. reported symptoms. The trial should therefore ideally be double-blind, i.e. neither the subjects nor the investigators know who is in the intervention group and who is in the control group. This is straightforward if the trial is of a drug or vitamin supplement as the control group can be given a placebo. It is more difficult for dietary trials, although it has been attempted in some studies by using "sensible eating" type advice for the control group. At the very least, the study should be single-blind, i.e. those collecting data on outcome measures are unaware of each subject's allocated group.
Follow-up of subjects must be as complete as possible. Any subjects who are lost to follow-up or withdraw during the trial reduces the overall value of the intervention study.
