In randomization we trust? There are overlooked problems in experimenting with people in behavioral intervention trials

OBJECTIVES: Behavioral intervention trials may be susceptible to poorly understood forms of bias stemming from research participation. This article considers how assessment and other prerandomization research activities may introduce bias that is not fully prevented by randomization. STUDY DESIGN AND SETTING: This is a hypothesis-generating discussion article. RESULTS: An additivity assumption underlying conventional thinking in trial design and analysis is problematic in behavioral intervention trials. Postrandomization sources of bias are somewhat better known within the clinical epidemiological and trials literatures. Neglect of attention to possible research participation effects means that unintended participant behavior change stemming from artifacts of the research process has unknown potential to bias estimates of behavioral intervention effects. CONCLUSION: Studies are needed to evaluate how research participation effects are introduced, and we make suggestions for how research in this area may be taken forward, including how these issues may be addressed in the design and conduct of trials. It is proposed that attention to possible research participation effects can improve the design of trials evaluating behavioral and other interventions and inform the interpretation of existing evidence.