KISS is the operative phrase in the design of the large-scale long-term practical steps you can take to reduce losses
• Target recruiting efforts at those both eligible and likely to participate. See Chapter 9.
• Review and, if possible, loosen eligibility requirements.
• Select appropriate participants, meaning those most likely to remain in compliance
• Establish measures to increase compliance. See Chapter 5.
randomized controlled clinical trials that are our chief concern in this guide. We advise you to resist all attempts to measure redundant variables ("but surely as long as the patient is in my office you won't mind if I perform one or two tests of my own") or complicate the design. On the other hand, short-term clinical trials of more limited scope whose objective is to determine the maximum tolerable dose or to establish efficacy can often benefit from the use of more complex designs such as a crossover or a fractional factorial.
In a crossover design, each patient receives all treatments in order, treatment A followed by treatment B, or treatment B followed by treatment A (or, if there are more alternatives, A followed by B followed by C). Thus each patient serves as her own control, reducing the individual-to-individual variance to an absolute minimum.
In a fractional factorial design, best employed when there are adjunct treatments and/or multiple cofactors, only some and not all treatment combinations are tested. Sophisticated statistical methods are used during the analysis phase to compensate for the missing data.
The advantage both of these design types offer is that they markedly reduce the total number of patients required for the trial. Their disadvantage, again in both cases, is that their validity rests on certain key assumptions that are seldom realized in practice.
To use a crossover design, one has to assume that neither treatment has a residual effect, that using B after A has exactly the same effect on a patient as if A had never been used.19 In particular, one has to assume that trace quantities of A or B and their metabolic byproducts do not linger in the body after treatment with A or B is ended. If crossover trials are contemplated, a pharmacokineticist is an essential addition to the design team.
To maintain the validity of a fractional factorial design, one has to be able to assume that the effect of treatment A is the same at all levels of the cofactor and in all subgroups. Again, these assumptions are seldom realized in practice and represent major drawbacks for the methodology.
But the main objection to these designs is that full-scale long-term clinical trials have not one but two purposes: to demonstrate both efficacy and safety. A sample size that might be adequate for demonstrating the one may be far too small to establish the other. The chief
19Or that the two treatments have the same residual effects, an even more unlikely prospect.
advantage of crossover and fractional factorial designs—reduction in sample size—is lost while their disadvantages remain.
A third type of study, occasionally used to demonstrate efficacy, employs case controls. The data for these controls are obtained by referencing historical databases and attempting to find patients whose profiles (demographics, risk factors, laboratory values) align as closely as possible with those of patients who received the investiga-tional intervention. As the allocation of patients to treatment was not made at random, nor were the treatments of control and experimental subjects contemporaneous, this type of design is not appropriate for full-scale clinical trials. They can be useful in demonstrating to the regulatory agency the validity of going forward with large-scale clinical trials (see Chapter 8).
If death is a possible outcome for an untreated or inadequately treated patient as it is, for example, with AIDS, you may want to consider the use of response adaptive randomization in which the majority of new patients are assigned to the currently most successful treatment. If the "success" was temporary or merely a chance event, then the proportions will gradually even out again or perhaps go the other way. But if further trials sustain the advantages of one treatment over another, then a greater and greater proportion of patients will be assigned to the preferable treatment and the number of deaths during the trials will be kept to a minimum.
The analysis of such trials is complicated, but it is well understood and thoroughly documented; see, for example, Yao and Wei (1996). The chief drawback is the lack of commercially available software with which to perform the analysis.
Was this article helpful?