Owen, J., Drinane, J. M., Idigo, K. C., & Valentine, J. C. (2015). Psychotherapist effects in meta-analyses: How accurate are treatment effects? Psychotherapy, 52(3), 321-328.
One of the ongoing debates in the psychotherapy research literature has to do with the relative efficacy of psychotherapies. Is psychotherapy brand A (CBT, for example) more effective than psychotherapy brand B (psychodynamic therapy, for example)? The most common way to test this question is with randomized controlled trials (RCTs), in which clients are randomly assigned to treatment condition (brand A or B). This study design controls for systematic bias in the results that may be caused by differences between clients. But what about therapists? We know for example that therapist effects (i.e., differences between therapists) account for approximately 5% to 10% of client outcomes. Therapist effects are often larger than the effect of the empirically supported treatment that is being offered. Yet it is almost unheard of for therapists to be randomized to treatments, so therapist effects are not controlled in most psychotherapy trials. As a result the effects of the differences between therapists get statistically rolled into the treatment effects. As Owen and colleagues point out, the impact of not controlling for therapist effects is that some differences between treatments in an RCT will appear statistically significant when in fact they are not. One can control for the effect of therapist differences, thus providing a more accurate estimate of treatment effects, but this is rarely done in published RCTs. So, when these RCTs are summarized in a meta analysis, the meta analysis results are also affected by ignoring therapist effects. In their study, Owen colleagues did something very clever. They took data from 17 recent meta analyses of RCTs that found differences between two interventions. These included meta analyses of studies comparing: CBT vs alternative treatments, bona fide treatments vs non-bona fide treatments, culturally adapted treatments vs those that were not adapted, etc. There are many other meta analyses that show no differences between treatments, but the authors wanted to focus specifically on the 17 that did show differences. Owen and colleagues statistically estimated what would happen to the original study findings of significant differences between treatments if therapist effects on patient outcomes were controlled. They controlled for three different sizes of therapist effects that accounted for: 5% (small), 10% (medium), or 20% (large) of patient outcomes. Even small therapist effects (5%) reduced the number of significant differences between treatments from 100% to 80%. When psychotherapist effects were estimated to be medium (10% - which is the best estimate based on research), the number of significant differences between treatments dropped to 65%. For large therapist effects (20%), the number of significant treatment differences was only 35%.
I have argued previously that the psychotherapist matters. Placing more time and effort in developing good reflective practice based on quality information and developing therapist skills like empathy, progress monitoring, and identifying and repairing alliance ruptures will result in better patient outcomes. As Owen and colleagues note, when reading an RCT that claims to find significant differences between psychotherapies, ask yourself if they took into account the effects of differences between therapists.