IMPACT ON RECRUITMENT AND ATTRITION
A great strength of RCTs is that each group is generally balanced on all characteristics, with any imbalance occurring by chance. However, during any trial, participants may be lost to follow-up, which reduces statistical power by decreasing sample size. In an RCT, there is an implicit trade-off between statistical power to detect an effect and the level of control over threats to validity (Mohr et al., 2009). Loss to follow-up can greatly influence the outcome of behavioral intervention trials (Dumville, Torgerson, & Hewitt, 2006). Bias may occur from attrition when there are different rates of attrition between the treatment group and the control group or the reasons for the attrition differ between the two groups (Tansella et al., 2006). This is an important source of bias, and that bias can remain large even when advanced statistical techniques, such as multiple imputation, are used to address the attrition. Thus, it is critical to minimize dropout from the control group. Suggestions for minimizing dropout include creating a research project identity; emphasizing the importance of the contribution of the control participants to the study results before randomization; developing a strong tracking system to be able to identify, locate, and determine the status of the control group members; and maintaining contact with them through telephone reminders, postcards, and newsletters (Miller & Hollist, 2007). Retention issues are discussed further in Chapter 10.
It is important when recruiting for behavioral intervention trials that the selection procedures do not yield a sample that is biased toward one or more treatment or control conditions. For example, participants might have a preference for a particular behavioral intervention, and this preference might lead to nonadherence and increased dropout rate, and even affect treatment response if a person gets randomized to his or her nonpreferred condition (Holroyd, Powers, & Andrasik, 2005). Potential participants who do not want a particular treatment condition or the control condition might be more likely to refuse randomization, or may not be as motivated to put forth their best effort. No ready solution to this problem is available at this time. The investigator should assess and report reasons for refusal of randomization and treatment preferences, and identify these as possible confounds, even if they cannot be completely controlled (Holroyd et al., 2005).
Additional information about how study participants’ expectations and preferences impact treatment adherence, attrition, and outcomes should be collected routinely in RCTs, but seldom is. This failure to control for expectations is not a minor omission and may have serious consequences that may undermine any causal inference (Boot, Simons, Stothart, & Stutts, 2013). For example, Boot et al. (2013) examined the game-training literature and concluded that not controlling for expectations limits conclusions that can be drawn about the effectiveness of active videogame training in improving cognitive and perceptual abilities. Although they singled out videogame interventions for their review, they also pointed out that this is a broader problem affecting most behavioral interventions targeting mental health, education, and personal well-being. They recommend that researchers explicitly assess expectations, carefully choose outcome measures that are not influenced by differential expectations, and use alternative experimental designs that assess and manipulate expectation effects directly.