Patient Preferences and the Ironic Nature of Randomized Trials

Abstract

Randomized trials are an attractive form of medical research that often contains blemishes when examined carefully. One criticism relates to limited generalizability, because such studies typically include only a fraction of the available patients and often exclude the challenging individuals who have complex comorbidities. Randomized trials can also fail because of practical issues such as insufficient recruitment (inadequate power bias), a limited duration of follow-up (brevity bias), or losses to follow-up (attrition bias). Several other opportunities for misinterpretation arise if the study is not adequately blinded so that outcomes and cointerventions are managed in a slanted manner. Furthermore, the large costs of a randomized trial (both financial and operational) create a potential conflict of interest for investigators and a worrisome reality that the research will not be repeated. This issue of the journal describes a further blemish in randomized trials that is subtle, pervasive, and rarely mentioned. Specifically, such trials randomize the assignment of patients to treatments and thereby require each person to disregard his or her inherent preferences. Such a behavior style is remarkably strange for a sentient adult, does not occur in usual clinical settings, and typically must be maintained for an extended interval. Of course, patients sometimes feel conflicted when facing a choice, and such indecision may take time to resolve before a subsequent commitment. By definition, however, a randomized trial requires each patient to follow a treatment that has been randomly assigned with no basis. Most people who sustain a career, marriage, diet, or other large-stakes effort would have a strong preference or other insight that underlies their specific choice. The article by McCaffery and colleagues provides an empiric demonstration of this discrepancy between self-selection and randomized assignment. Participants underwent a 2-step randomization where eligible patients were first divided into 2 groups; namely, those who were allowed to selfselect their treatment and those who instead received randomized assignment (hence, patients in the latter group were randomized twice). The main test evaluated the observed effects among self-selected patients compared with randomly assigned patients. In this study, the patients were diagnosed with cervical atypia, the medical treatment was a new screening strategy to prevent cervical cancer, and the primary outcome was a measure of mental health. The principal finding was that mean mental health scores were higher if the patient received the new screening strategy by self-selection rather than randomized assignment (47.57 v. 46.16, respectively). These new results corroborate standard intuitions on the importance of choice. For example, consider a trial testing whether chocolate milk leads to more consumer satisfaction than strawberry milk. One way to test such a hypothesis would be to recruit participants, allow each to self-select a flavor, and measure subsequent satisfaction. An alternative approach would be to recruit a second group, serve each participant a flavor by random assignment, and measure subsequent satisfaction. Consider that chocolate is generally popular but not universally enjoyed by all. If so, the randomized trial may still show positive results in favor of chocolate; however, the magnitude of relative benefit would be attenuated by the subgroup of participants who were assigned to chocolate but who would have preferred strawberry. An accurate estimate of flavor preferences requires some studies based on self-selection and not just randomized assignment. From the Department of Medicine, Brigham and Women’s Hospital, Institute for Clinical Evaluative Sciences in Ontario, Canada (JAU) and the Institute for Clinical Evaluative Sciences, Clinical Epidemiology Program, Sunnybrook Health Sciences Centre Department of Medicine, University of Toronto, Canada (DAR).