The Design of Clinical Trials

Summary Clinical trials have grown in number and size since their introduction over 30 years ago. They embody Fisherian principles of experimentation, with emphasis on randomization rather than on complex structure. Randomization has recently come under attack, regrettably in the author's view. Much work has been done on methods of achieving balanced randomization, although the advantage of such schemes over post‐trial adjustment is not clear. Trials are often too small to achieve sensible objects. Size is mainly determined by power considerations, decision‐theoretic models being difficult to relate to practical needs. Sequential methods are available, and have become more flexible. Problems of planning and execution, on which the statistician's view is important, include the homogeneity or otherwise of patients, uniformity of treatment schedules, procedure for dealing with protocol departures and the avoidance of bias in assessment of response. Cross‐over designs are useful for the study of short‐term alleviation of chronic disease symptoms, but more attention should be paid to the detection of treatment × period interaction, which is based on between‐subject comparisons. More use should be made of factorial designs, although they give rise to difficulties of interpretation when factors interact.