Elsevier

Clinical Psychology Review

Volume 50, December 2016, Pages 95-107
Clinical Psychology Review

Review
Mechanisms of change in psychotherapy for depression: An empirical update and evaluation of research aimed at identifying psychological mediators

https://doi.org/10.1016/j.cpr.2016.09.004Get rights and content

Highlights

  • Little is known about the psychological mechanisms of psychotherapy for depression.

  • The mechanism question has motivated dozens of investigations of mediation.

  • We provide an empirical update and critical evaluation of this body of research.

  • Research is heterogeneous and unsatisfactory in methodological respect.

  • Psychotherapy might be too complex to be explained in simple models of psychological change.

Abstract

We present a systematic empirical update and critical evaluation of the current status of research aimed at identifying a variety of psychological mediators in various forms of psychotherapy for depression. We summarize study characteristics and results of 35 relevant studies, and discuss the extent to which these studies meet several important requirements for mechanism research. Our review indicates that in spite of increased attention for the topic, advances in theoretical consensus about necessities for mechanism research, and sophistication of study designs, research in this field is still heterogeneous and unsatisfactory in methodological respect. Probably the biggest challenge in the field is demonstrating the causal relation between change in the mediator and change in depressive symptoms. The field would benefit from a further refinement of research methods to identify processes of therapeutic change. Recommendations for future research are discussed. However, even in the most optimal research designs, explaining psychotherapeutic change remains a challenge. Psychotherapy is a multi-dimensional phenomenon that might work through interplay of multiple mechanisms at several levels. As a result, it might be too complex to be explained in relatively simple causal models of psychological change.

Introduction

Many researchers in the field of clinical psychology agree that gaining a better understanding of the mechanisms underlying psychotherapeutic change is crucial for optimizing treatment outcomes for patients suffering from psychiatric disorders such as depression (Kazdin and Nock, 2003, Kraemer et al., 2002). Knowledge about active ingredients of therapy can assist in the verification and refinement of theories of the disorder, and allows enhancement of elements that are crucial for therapeutic change, while dismissing those found to be redundant (Garratt et al., 2007, Longmore and Worrell, 2007).

An important first step towards examination of mechanisms of change is the identification of mediators (Kazdin and Nock, 2003, Kraemer et al., 2001, Kraemer et al., 2002). A mediator is a variable that statistically explains why and in what way a treatment has an effect on outcome, and can be seen as a potential mechanism: the actual process or event that is responsible for change (Baron and Kenny, 1986, Kazdin, 2007, Kazdin, 2009, Kraemer et al., 2001, MacKinnon et al., 2007). In other words, the mechanism is the phenomenon to reveal, the mediator can be the mean to this end. Mediators can be distinguished from moderators in the sense that they explain the relationship between an independent and dependent variable (i.e. they indicate whether treatment has an effect on outcome via the mediator), whereas moderators influence that relationship (i.e. they indicate when or under what conditions the relationship between treatment and outcome can be expected: Hayes, 2013).

Establishing a mediator involves several requirements. For a long time, mediation solely referred to statistical mediation: to statistically demonstrate that the effect of treatment on outcome is explained by a third variable: the mediator. The most well-known method to determine statistical mediation is indubitably Baron and Kenny's (1986) causal step method. With almost 60.000 citations, their paper is one of the most frequently cited articles in the field of psychology. According to Baron & Kenny, mediation is established when 1) there is a main effect of treatment (efficacy test), 2) treatment is related to change in the mediator (intervention test), 3) change in the mediator and change in outcome are related (psychopathology test), and 4) the effect of treatment on outcome is absent (full mediation) or significantly weakened (partial mediation) when statistically controlling for the mediator (mediation test). Subsequently, a Sobel test (Sobel, 1982) determines the amount of mediation – also called the indirect effect.

Influential as it has been, the Baron and Kenny (1986) model has significant limitations for application in social sciences and therefore also in clinical process research for disorders such as depression. For example, the method has low type I error rates and, in order to have sufficient power, requires large sample sizes and large treatment effects, both of which are not always available in this type of research (Hoyle and Kenny, 1999, MacKinnon et al., 2002, MacKinnon et al., 2004, Shrout and Bolger, 2002). The applicability of the model in this field is further limited by restrictions resulting from the first and fourth criterion. The first criterion (efficacy test) is formulated in a way that the ability to perform mediation analysis strongly depends on the presence of differential treatment effects. When two treatments turn out to be equally effective – a phenomenon that is not uncommon in the field of psychotherapy for depression (for more details see e.g. Cuijpers and van Straten, 2011, Cuijpers et al., 2008, Wampold et al., 1997) – this type of mediation analysis is not possible. This is an important drawback, because especially when two treatments turn out to be equally effective it is important to examine processes of change, since this can tell us more about whether the change that is observed is reached through similar or differential pathways (MacKinnon, 2008). Moreover, given the population (depressed patients) and the nature of treatments (psychotherapy), it is ethically and practically very difficult (if not impossible) to include a substantially less powerful treatment (such as a full waiting-list control group, or a placebo intervention) to increase the contrasts between groups. And even if a third ineffective control condition would be added, it is still not possible to test differential pathways between the two equally effective treatments. The fourth Baron and Kenny (1986) criterion (mediation test) has been criticised because the tests that have to demonstrate the reduction of the effect after statistically controlling for the mediator have shown to be underpowered (MacKinnon et al., 2007).

As a result of these limitations, the criteria for statistical mediation have been modified over time to make them more applicable and suitable for treatment research. For example, the MacArthur group (Kraemer et al., 2001, Kraemer et al., 2002) toned down the importance of the first criterion by stating that differential treatment effects are not required to establish mediation as long as there is an interaction between treatment and the mediator. This is particularly useful in clinical trials comparing two (equally) effective treatments that are likely to operate through different mechanisms. With regard to step 4, it was decided that it was sufficient to show that treatment has an effect on the mediator and that the mediator has an effect on the outcome, even after controlling for treatment, a procedure known as joint significance testing (MacKinnon et al., 2007). Furthermore, advances have been made in statistical methods to test the various mediation models (see developments by e.g. Arbuckle, 1999, Arbuckle, 2005, Kraemer et al., 2001, Kraemer et al., 2002, MacKinnon et al., 2002, MacKinnon et al., 2004, MacKinnon et al., 2007, MacKinnon, 2008: Muthén and Muthén, 2001, Muthén and Muthén, 2007, Preacher and Hayes, 2004).

Although statistical mediation still plays a central role in addressing whether a particular construct accounts for change (Hollon and DeRubeis, 2009, Kazdin, 2007, Kazdin, 2009), it is not sufficient to make a case for the operation of a mediator (e.g. Johansson and Høglend, 2007, Kazdin, 2007, Kazdin, 2009, Laurenceau et al., 2007). Probably the most important addition to statistical mediation is demonstrating the direction of causality. Conditions for inferring causal relations in scientific research have been outlined by e.g. Hill (1965), Kenny (1979), Schlesselman (1982), and brought to the psychotherapy literature by Kazdin (2003, Kazdin, 2007, Kazdin, 2009). Apart from a strong statistical association between treatment, mediator and outcome, Kazdin describes six requirements for adequate evidence for causal temporal relationships. First of all, it has to be demonstrated that the treatment causes the mediator variable to change, which in turn causes the outcome, and not the other way around (Kazdin and Nock, 2003, Kraemer et al., 2002). In order to get a clear view of the shape of change and the relation between mediator and outcome, it is important that both the mediator and outcome measure are assessed at multiple time points during treatment. The importance of demonstrating temporality is supported by many research groups (e.g. Collins and Graham, 2002, Hollon and DeRubeis, 2009, Johansson and Høglend, 2007, Kazdin, 2007, Kazdin, 2009, Kazdin and Nock, 2003, Kraemer et al., 2002, Laurenceau et al., 2007, Murphy et al., 2009), and has even been called the fifth step of statistical mediation analysis (Johansson et al., 2010). Second, alternative explanations for the observed relation between mediator and outcome should be ruled out. This can be done by using an experimental approach in which all variables are held constant across individuals in various conditions while changing only the proposed mechanism of change (Kazdin, 2007, Kazdin, 2009). Furthermore, Kazdin emphasizes the importance of specificity of the association among the intervention, proposed mediator and outcome. This means that it has to be demonstrated that the mediator plays a crucial role in one treatment, but not (or less so) in the other. In addition, inclusion of plausible processes, consistency across studies, and a gradient, in which larger changes in the mediator are associated with larger changes in outcome, should further enhance the evidence.

Kazdin (2007) emphasizes that each criterion is important, but that interpretations should be made based on their convergence. Examination starts with statistical tests for mediation. After that, one determines the value of the results by examining the extent to which a study meets the other criteria. Even though the satisfaction of each criterion increases the strength of the argument for the operation of a mediator – or even a mechanism – not all criteria are weighted equally important. According to Kazdin and Nock (2003), statistical association, temporality, specificity, and experiment are considered to be the most important, whereas the remaining three should further enhance the evidence.

The extended requirements and possibilities for identifying mediators also called for additional features of study designs. According to the latest standards, the extent to which a process meets the requirements for mediation can only be examined properly in a theoretically well planned RCT with carefully spaced repeated measures, sufficient power and an appropriate control group (Kazdin, 2007, Kazdin and Nock, 2003, Kraemer et al., 2002, Laurenceau et al., 2007). Furthermore, it is important to experimentally manipulate the proposed mediators, which requires an experimental study design. In addition, mediation analysis should be performed using up-to-date definitions and state-of-the-art statistical analyses techniques (Collins and Graham, 2002, Haaga and Stiles, 2000, Haubert and Dobson, 2007, Kraemer et al., 2002, Laurenceau et al., 2007, MacKinnon et al., 2007). Moreover, depending on what the theory stipulates about processes, assessment of a single mediator might not be sufficient. It is therefore recommended to include multiple mediators to examine rival hypotheses, test alternative explanatory models, and map out interactions between theorized processes.

The past decades, the interest for mediators in mechanism research in depression has grown, and several research groups worldwide have studied mediators of psychotherapy. In 2007, Johansson and Høglend identified 61 studies that performed mediational analyses to identify the active ingredients of psychotherapy for several psychiatric disorders. A closer look at the literature specific for depression indicates that the majority of studies has focused on the mediational role of cognitive processes, such as automatic thoughts, dysfunctional attitudes, attributional style, and other cognitive distortions. The cognitive mediation hypothesis was also the focus of the influential systematic review by Garratt et al. (2007). Garratt and colleagues summarized results of 31 studies on the role of cognitive change and concluded that research generally supports the cognitive mediation hypothesis, but that this does not necessarily need to be specific for interventions in which cognitions are actively targeted. This indicates that cognitive change, no matter how it occurs, might play a role in various treatment modalities. Even though Garratt et al. acknowledged that these findings increased knowledge about the relation between cognition and depression, they emphasized that their findings did not permit clear-cut answers about the exact role of cognitive change as a process that facilitates psychotherapeutic change in the context of psychotherapy. They provided several reasons for this. First of all, there was a large variety in research questions and methodology across studies, which made it difficult to compare results across studies and to integrate findings into broader knowledge. Second, many studies did not meet the criteria for reputable mechanism research, hereby limiting the interpretability of study findings. More specifically, Garratt et al. concluded that none of the studies that were identified in their review addressed the criteria for mediation in methodologically sound ways. Garratt and colleagues expressed their hope that this would change in subsequent years, in studies with e.g. larger sample sizes, up-to-date-statistical methods, and a broader array of measures. These issues are acknowledged by others in the field as well (e.g. Johansson and Høglend, 2007, Kazdin, 2007, Kraemer et al., 2001, Laurenceau et al., 2007). A third difficulty in interpreting results from studies in this field – not mentioned by Garratt et al. – is the fact that not every study that makes claims about mediators, actually performed statistical mediation analyses. Instead, some studies present correlations between changes in hypothesized process measures and depressive symptoms from pre- to post-treatment as evidence for mediation. Others make claims about mediators based on prediction analyses. This does not only further increase the heterogeneity in the field, but also leads to conclusions about mediators in studies where no statistical mediation analyses were performed. Garratt and colleagues did not differentiate between this in their review. Fourth, since most studies so far mainly focused on the role of cognitive factors, the influence of non-cognitive factors is still largely unknown.

Almost ten years have passed since the Garratt et al. (2007) review, and the question is whether and how the field has changed. The aim of the current review was therefore to provide an update and critical methodological evaluation of the current body of research on this topic. In a systematic literature search, we selected studies aimed at identifying psychological mediators in psychotherapy for depression. To get a comprehensive overview of the field, we included various forms of psychotherapy and included both cognitive and non-cognitive processes. We only selected studies that included an actual test of statistical mediation (Baron & Kenny (1986) or one of the more advanced methods). We summarize study characteristics and results of 35 studies and discuss the extent to which these studies meet the most important requirements for mechanism research that were mentioned earlier. With this we hope to learn more about the magnitude and relevance of the existing body of research and map out necessities for future research.

Section snippets

Data sources and data reduction

Three different approaches were used to identify relevant studies. First, five databases (i.e. PubMed, PsychInfo, Embase, Cochrane, and Cinahl) were systematically searched for potentially relevant papers that were published in English in peer reviewed journals until spring 2016. Key terms were Depression, Psychotherapy, Mechanisms and Mediation (a full key-term scheme can be found in Appendix A). The data search yielded a total of 617 unique studies. One of us (VM) carefully read through all

Study characteristics and results

Table 1 (left panel) gives an overview of study characteristics and results of 35 studies that were included in the review. The majority of studies was conducted in the USA (57.1% vs. 28.6% in Europe, and 14.3% in other parts of the world), and 48.6% was published in the past five years (2012–2016). Sample sizes ranged between n = 4 and n = 523, with a mean of n = 173 (SD = 145.3). Patients were adults (in 26 studies) and adolescents (in 9 studies) ranging in age from 12 to 68 years (M = 40.2 SD = 8.2 for

Discussion

We provided a systematic empirical update and critical evaluation of the current status of research aimed at identifying a variety of psychological mediators in various forms of psychotherapy for depression. With this we wanted to learn more about the magnitude and relevance of the existing body of research and map out necessities for future studies. We summarized study characteristics and results of 35 relevant empirical studies that were identified in a systematic literature search, and

References (0)

Cited by (128)

View all citing articles on Scopus
View full text