The impact of patient preference on the design and interpretation of clinical trials

Research on several health problems shows that patients and health care providers do not use the same criteria to evaluate the effectiveness of treatment and often disagree on the severity of symptoms. When the disease is chronic and the main aim of treatment is to improve quality of life, we argue that variables rated as important by patients should be used as outcomes in clinical trials, and that in most cases these need to be measured from subjects' self-reports. In many non-pharmacological randomized clinical trials, the subjects cannot be blinded to treatment. Furthermore, many of them will probably have a preference for a particular treatment option. It has been proposed that emotional responses following assignment of treatments, which may or may not be preferred, will strongly influence the outcome, especially when it is based on self-reports of treatment satisfaction. Because of this concern, some investigators have suggested alternative study designs that incorporate preference. Brewin & Bradley (Br Med J 1989; 299 [6694]:313-5) have proposed allocating subjects to treatment methods according to their preferences, and randomizing those individuals with no preference. To determine the influence of preference on treatment outcome, they recommend comparing results from the preference group with those of the randomized group. However, we have found that there are dear differences in level of education and in the pre-treatment state between individuals with preferences and those with no strong preferences. Therefore, we believe that the design proposed by Wennberg et al. (Ann N Y Acad Sci 1993;703:52-62) is more appropriate. Ln it, subjects are randomly allocated to a preference trial (subjects choose their treatment) or to a randomized trial (random allocation to treatment). Between-trial comparisons can then be used to determine the influence of preference on outcome. This will lead to better evaluation of treatment efficacy and allow better estimates of the true effectiveness to be made.