Mediators of outcome in adolescent psychotherapy and their implications for theories and mechanisms of change: a systematic review

Adolescents are a particularly important and vulnerable group with distinct mental health needs due to the developmental changes in this age period. According to Kessler and colleagues, 50% of lifetime diagnosable mental health disorders start before the age of 14, and this number increases to 75% before the age of 25 [1]. Ignoring young people's mental health needs can result in long-lasting adverse developmental outcomes for the individual and society. These include mental disorders in adulthood, impaired social and role functioning, difficulties with employment, as well as poorer general health outcomes [2]. In contrast, helping adolescents overcome mental health problems lays the foundation for their adult health and the health of their offspring. Thus, addressing adolescent mental health is of utmost importance.

Psychological treatment is one of the important keys for the promotion of youth mental health, given the extensive evidence on the general effectiveness of psychotherapy for treating mental disorders in adolescence [3, 4]. However, effect sizes are smaller when treating adolescents as opposed to adults, especially in the case of multiple simultaneous problems (e.g., comorbidity and social problems) [5]. This likely reflects at least partly the fact that psychotherapy with adolescents differs from the treatment of adults in multiple regards [6]. For example, in contrast to adults, adolescents are more often not self-referred but rather sent to therapy by others and may, therefore, be less intrinsically motivated for treatment. They may also present with age-specific symptoms, diagnoses, and target complaints [7]; their treatment more likely involves third parties (such as guardians, teachers, social workers, etc.); and they more likely have fewer shared interests and differ more in both age and value-systems with their therapists; all of which can compromise agreeing on therapy tasks and goals and achieving a good therapeutic alliance and, subsequently, outcomes [8, 9].

Furthermore, with more than 550 different posited psychotherapy models, potentially also applicable to young people, little apparent consensus exists on how psychotherapy works [10]. Thus, to increase and optimize treatment outcomes for young people, processes that facilitate successful therapeutic change should be empirically identified irrespective of therapeutic schools. By identifying transtheoretical treatment processes that facilitate successful therapeutic change and isolating those that are redundant and can be dismissed, it becomes possible to optimize treatment outcomes for this age group. This review summarizes all studies that have investigated mechanisms of change in the psychological treatment of adolescents by analyzing change mechanisms in relation to outcome in randomized controlled trials (RCTs).

An important first step towards examining mechanisms of change in psychotherapy is the identification of mediators [11]. While patient-related moderators (e.g., gender, age) serve to clarify what kind of treatment works for which kind of person, mechanisms of change define causal relationships between therapeutic change and psychological interventions. A mechanism of change explains how an intervention translates into a process that leads to an outcome, e.g., change in symptoms [12]. Thus, a mechanism is an explanatory concept that relies on identifying mediators, i.e., variables that explain changes between an intervention and one or more therapeutic outcomes statistically. Kazdin has formulated clear criteria on how to assess mediators of psychological treatments [13] as follows:

After 16 years of Kazdin’s suggestions on how to assess mechanisms of change in psychotherapy research, the extant lack of knowledge remains striking. In their review, Cuijpers et al. [14] concluded that despite more than 70 years of systematic empirical psychotherapy research, we still have no empirically validated mechanisms of change in psychotherapy, neither in terms of common nor specific factors (i.e., mechanisms operating across or solely within particular therapeutic approaches)—yet the evidence base is even smaller in psychotherapy for children and adolescents. However, there has been an increasing number of publications on mechanisms of change and recent attempts to synthesize the empirical findings. We have found 20 newer systematic reviews and/ or meta-analyses investigating change mechanisms in psychotherapy which used Kazdin’s criteria when discussing the robustness of the evidence. The majority of these reviews investigated studies with adults and focused on one specific change mechanism in one specific treatment model for one specific disorder, e.g., repetitive negative thinking in CBT for depression [15]; threat reappraisal [16], cognitive changes [17], or panic efficacy [18] in CBT for anxiety disorders; as well as mindfulness-based interventions for the treatment of depressive rumination [19]. Other reviews investigated several mediators with one specific therapy model related to one specific disorder, e.g., CBT for drug use disorder [20], CBT for insomnia [21], and CBT for irritable bowel syndrome [22]. Yet others examined proof for theory-driven change mechanisms in specific therapeutic approaches, e.g., Mindfulness-Based Cognitive Therapy [23], Motivational Interviewing [24] or Acceptance and Commitment Therapy [25]. A further category of reviews summarized evidence related to specific diagnoses across different therapeutic approaches, e.g., mechanisms in Internet- and mobile-based intervention for adult depression [26] or adult PTSD [27], or treatments for depression in adults, children and adolescents [28], and prevention of depression and anxiety [29], also across all ages. In addition, one systematic review and meta-analysis was dedicated to CBT for the treatment of anxiety in children and adolescents [30], one systematic review summarized studies on change mechanisms in psychological treatments for depressed adolescents [31], and another was dedicated to change mechanisms in externalizing disorders in adolescents and children [32]. A recent scoping review took a more comprehensive approach and examined predictors, moderators, and mediators associated with treatment outcome in randomized clinical trials among adolescents with depression [33]. Last, some reviews focused on specific mediators irrespective of diagnoses or treatment approaches, such as alliance [34] and insight [35]. Table 1 summarizes the reviews of studies that focused exclusively on adults and below, we will summarize only reviews of studies that included children and adolescents.

This article is based upon work from the COST Action TREATme (16102). The review was registered in Prospero (CRD42020177535) and follows the preferred reporting items for systematic review and meta-analysis (PRISMA) [39]. The patient/population, intervention, comparison and outcomes (PICO) model [40] was used to define the research question as follows: “In adolescents with psychological problems (P) receiving a psychological intervention (I), what mediators of outcome (O) have been evaluated and found to be significant, when compared to other interventions or control groups (C)“?

The search in MEDLINE and PSYCHINFO identified 5063 papers of which 4461 remained after removing duplicates (cp. Fig. 1). After screening of abstract and titles, 1006 publications were reviewed in full text. Exclusion criteria after reading full texts were mainly due to not fulfilling the age range criteria, not including a psychosocial intervention of any kind, not performing formal mediation tests and not including a RCT. Finally, 106 studies fulfilled the eligibility criteria of the present review and data from these were extracted.

Table 2 summarizes all 106 studies included with regard to main author, year of publication, country where the study was conducted, diagnoses or psychological problem, characteristics of the participants, treatment and comparator, mediation statistics and mediator categories as well as mediator significance. The summary table is grouped by diagnosis. The specific diagnoses were categorized in broader types if more than five studies addressed a specific disorder range. This yielded the followingsix diagnostic categories: anxiety (k = 19), depression (k = 20), externalizing disorders including offending and criminal behavior (k = 14), substance use disorders including alcohol, cannabis, cigarettes and other drug use (k = 25), posttraumatic stress disorders (k = 8) and other diagnoses (k = 22) for diagnoses that were only studied in one to three RCTs (e.g. eating disorders, OCD, ADHD, irritable bowel syndrome, etc.). Two studies were dedicated to both depression and anxiety equally [54, 55] and were thus counted for both diagnoses.

The mean age of participants ranged from 10.29 to 19.99 years. When mean age was not reported, the age range was used instead. In some cases, authors reported only school grades. We decided to include studies with school grades within the age range of 10 and 19, starting with grades five to six. Studies varied between 0 and 100% of female participants. Across all studies, a ratio of 57% of participants was female. Sample sizes ranged between 13 and 4277 participants with a mean of 170 participants (when the outlier of 4277 participants was excluded). In total, data for 19,407 participants were included in this review.

The included studies were from 11 different countries. A huge majority of studies were conducted in the USA (k = 73, 68%), followed by k = 8 from Australia (7.5%) and k = 6 (5.7%) each from the Netherlands and Norway and five from the UK; there were also two studies each from Canada, Germany and Sweden (one Swedish study was co-joint with the USA). One study each was included from Israel, Palestine and Hong Kong.

The interventions included in the studies were diverse and categorized into eight different approaches as folows: Cognitive-Behavioral Therapy (CBT with k = 54, 51%), Humanistic Therapies (HUM with k = 23, 22%), Systemic Therapies (SYS with k = 16, 15%), Educational Approaches (EDU with k = 14, 13%), Interpersonal Therapy (IPT with k = 2, 1.8%), third wave mindfulness therapies (MIN with k = 5, 5%) and other approaches (OTH with k = 3, e.g. Positive Psychology or culturally adapted programs). Five studies investigated an explicitly integrative treatment approach and one study included psychodynamic therapy (PDT). Settings differed in terms of individual, group or family therapy or a combination, respectively: 68 studies reported on treatments that included individual therapy for adolescents of which k = 41 offered only individual therapy, k = 22 offered individual therapy in conjunction with family sessions, and k = 5 offered individual therapy in conjunction with group sessions and/or family treatment. Group treatment was investigated in k = 31 studies in which k = 24 had group treatment only, and in the other studies, group treatment was combined with either individual and/or family treatment. Family therapy without additional settings was only conducted in k = 12 studies. All studies investigated outpatient treatment using a face-to-face (k = 88), online (k = 10) or blended (k = 8) mode of delivery. Treatment length in weeks varied from one to 54 weeks with a mean duration of 8.7 weeks. Session amount ranged from one to 52 sessions with a mean session count of 8.6 sessions across all studies. 24 studies reported on treatments equal to or shorter than four weeks and 12 studies reported on single-session interventions. Studies also differed in their choice of comparator conditions, which included no intervention (k = 14, 13%), waiting list (k = 13, 12%), treatment as usual (k  = 19, 18%), placebo (k  = 3, 3%), pharmacotherapy (k = 7, 6.6%) education (k= 24, 22.6%), guided self-help (k = 5, 4.7%) and active psychological treatments (k  = 31, 29%).

In terms of general statistical models for mediation tests, the following methods were extracted from the included studies: regression models (k = 43, 40.6%), structural equation modeling (SEM) (k = 20, 19.6%), path models (k = 13, 12.7%), hierarchical linear modeling (HLM; k = 14, 13.2%), general linear mixed model (GLM; k = 6, 5.8%), growth curve analysis (GCA; k = 4, 3.9%), multilevel regression analysis (MRA; k = 4, 3.9%) as well as cross lagged panel analysis (CLP; k = 2, 1.9%). Different estimation techniques and/or programs were reported to calculate the indirect effect of the mediator variable on outcome such as bootstrapping (k = 42, 39.6%), Baron & Kenny (B & K; k = 10, 9.8%), Sobel test or delta method (k = 7, 6.8%), PRODCLIN (k = 7, 6.8%), joint significance (k = 3, 2.9%), singular use of asymmetric distribution of products, marginal mediation models, and rank preserving models. However, k = 34 (32.1%) studies did not clearly report a specific estimation technique.

As far as we know, this is the first systematic review assessing mediators of psychotherapeutic changes in adolescents across treatments and diagnoses. By virtue of our transtheoretical and transdiagnostic approach, we extended a broader net than prior reviews to include a range of different therapeutic approaches, and found that mediator studies have involved mainly CBT, Systemic Therapy, Humanistic Therapy and Educational approaches. Regarding knowledge on therapeutic change mechanisms for adolescents, we found there is a striking lack of RCT mediator studies for IPT, third-wave therapies and above all—in view of their prevalence [62]—for psychodynamic therapies. All in all, 252 results on mediator-analyses in 106 studies were identified.

Concerning the question of which mediators have been studied in psychotherapy RCTs with adolescents, the biggest group of mediators were from the cognitive field, followed by family-related, behavioral, therapy-related, relational, and emotional mediators, respectively. Given the generally accepted importance of improved emotion regulation for successful therapies (e.g. [63, 64]), and how critical emotion regulation skills become during the sometimes tumultuous years of adolescence [65], the relatively small percentage of studies investigating emotional mediators is surprising. However, this may partly reflect methodological challenges, e.g., in reliably measuring emotional processes and separating change in them from change in ultimate outcomes, e.g., reduction in depressive symptoms. Overall, the 252 mediators were assessed with 181 different measures which comprised mainly self-report measures. This strongly suggests that there is little consensus on the central change mechanisms or related mediator measures between or even within different therapeutic approaches, irrespective of diagnoses. Thus, to create an orienting overview for future studies, mediators were subcategorized within the broader domains to identify putative transdiagnostic and transtheoretical change mechanisms that are supported by research. Within each category some mediators were identified to be promising for future investigations: changes in negative thoughts, dysfunctional beliefs and metacognitive skills in the cognitive category; family functioning and parenting skills among family-related mediators; and successful engagement in therapy activities as well as increased impulse control in the behavioral mediator category. This is partly confirming former reviews on adolescents that reported significant mediation results for changes in negative thoughts, family functioning, social engagement as well as parenting [31, 32]. Results on relational, therapy-related, and emotional mediators were inconclusive due to fewer studies, as well as a higher percentage of insignificant results, although peer-influence appears a promising mediator.

With regard to our second research question of whether there are age-, disorder- and treatment-specific mediators, we conclude that, in comparison to research in adult samples, age-specific mediators with age-adjusted measures have been studied. More importantly, some promising findings of significant mediators emerged especially in the domains of family-related changes (e.g., family functioning and parenting), specific adolescent behaviors (e.g., engagement in therapy) as well as relational changes (e.g., peer-relations). Further findings indicate that the studied constructs of cognitive mediators do not greatly differ from those in research on adult psychotherapy, although appropriately age-adjusted measures have been used with adolescent populations. Various disorders have been the target of mediator studies up until now, mainly focusing on anxiety, depression, externalizing disorders, trauma and substance abuse. Fewer studies have focused on OCD and eating disorders, singular studies have addressed other diagnoses, and no study in our review focused on change mechanisms in personality disorders among adolescents. Certain types of mediators have been investigated specifically in the context of certain diagnoses (e.g., phobic beliefs, post-traumatic cognitions, or antisocial peer influence). However, the majority of mediators have been studied across different diagnoses and could be characterized as non disorder-specific but rather relevant to common change processes.

The research evidence for particular mediators within the main diagnoses ranged between moderate and insufficient. None of the mediator categories (cognitive, family-related, behavioral, relational, therapy-related and emotional) could be classified as having strong evidence. No sufficient evidence was found for emotional mediators as these were rarely examined. Moderate evidence for the other categories was found in studies on anxiety for behavioral and cognitive mediators, in studies on externalizing disorders for family and relational mediators, in studies on substance abuse for all except relational mediators, in studies on trauma for cognitive mediators, and in studies on depression for all mediator categories. Perhaps unsurprisingly, we also found that certain therapeutic approaches were investigated more often in certain diagnoses: more CBT studies in anxiety and trauma, more humanistic (typically motivational interviewing) studies in substance abuse, and more systemic therapy studies in externalizing disorders. Given the tendency to mainly test treatment-specific change mechanisms without considering other possible change theories, the findings on posited diagnosis-specific mediators are unavoidably, if understandably, confounded by researcher bias. By investigating mainly mediators that are in line with specific therapeutic models, the commonly used research designs follow a logic of verifying rather than falsifying theoretical assumptions [66]. Thus, at present, the research appears insufficient for objectively and reliably identifying disorder-specific mediators.

More generally also, we note that only nine of 106 (8.8%) studies could be classified as having both relatively low risk of bias in design and proper rigor in testing mediation models. On the contrary, the majority of studies (k = 69) (67.6%) needed to be classified as being of poor quality, due to unclear reporting or limitations in one or both aspects. This highlights the importance of thoroughly reporting all aspects of study design for future mediation studies. More specifically, in terms of Kazdin’s criteria—i.e., recommendations about statistical power, the temporal criterion, reliable measures and tests, specificity, and the experimental manipulation of the mediator—virtually all studies had at least some shortcomings. With regards to statistical power, as the mediation tests comprised mostly secondary analysis, no study reported a power analysis for mediation testing in preparation for the trial. Yet, at least in the broad sense of having a group larger than 40 participants, fortunately almost 74% of the reviewed studies appeared sufficiently large for investigating mediation (cp., [28]). However, 64.7% of the studies did violate the temporal criterion, which has been noted as a concern before [21, 67, 68]. Perhaps the most alarming finding of this comprehensive review was the exceedingly large number of measures used to assess mediators; indeed, each mediator measure was only seldom employed in more than one or two studies. This heterogeneity should be kept in mind as it undermines construct validity and findings of significance within mediator categories should be therefore interpreted with caution. Furthermore, almost all of the used measures were self-reports which shows a methodologically limited perspective, with very few studies using observer-rated or objective measures (e.g., computer task).

With regard to statistical tests of mediation, our review identified 14 different methodologies; the majority of studies used bootstrapping or SEM methods. However, as far as we could make out, 33 studies did not report a specific estimation method for the indirect effect of the mediator on outcome. As there is currently no generally accepted gold standard for mediation analysis, these were all treated similarly in our BESRS rating system. Future advances in establishing such standards might nevertheless also provide more nuance to the current findings and there is a high need for consensus guidelines in reporting and planning of mediation studies overall [33, 69]. In our review, there was only a small number of studies showing specific and robust associations between mediator and outcome in the intervention group only (i.e. [70‐74])—although in these studies the overall evidence was slightly less strong, in terms of either experimentally manipulating the mediator or assessing alternative mediators. The main limitation shared by virtually all the studies in terms of study design was a lack of clear experimental manipulation of the mediator [75]. The same limitation was also identified by Lemmens and colleagues [28] in their review of studies on mechanisms of change in adulthood depression. They speculated that a possible reason might be that treatments in which only one isolated mediator is manipulated while keeping everything else constant would likely lack external validity. Considering the lack of experimental manipulation of the mediator in most study designs, it seems important to mention also a few clinically relevant studies whose designs seemed commendable for attempts to specifically influence the mediator and assess its impact on the outcome, such as the studies on at-risk adolescents for anxiety [76] or eating disorders [61]. These study designs thus show that, with sufficient creativity, disentangling plausible mediators can also be achieved in ecologically valid investigations of clinically relevant phenomena. The limited number of studies manipulating the mediator directly does not mean that such studies should not be attempted or are not possible. Indeed, recent literature has seen discussion of various alternatives in both experimental study design and statistical methodology to enhance the inferring of causality in mediation studies [75, 77, 78]. As noted already long ago [79], but discussed more recently also by Kazdin [80] and others (e.g., [81, 82]), causality can reasonably be inferred even in the absence of randomization—and, in some cases, has to be. Furthermore, multiple alternative designs and methods can be useful in shedding light on mechanisms of change, such as single-case experimental studies [80] and qualitative work embedded in RCTs (e.g., [83]).

The studies included in this review were primarily conducted in the western industrialized world which constitutes a limitation in terms of representativity. And while there is an excellent gender balance, studies failed to report if results differed with respect to gender, preventing conclusions on gender-specific mediators for adolescents.

Given the considerable heterogeneity between studies (most notably the multitude of mediator constructs and measures) and no agreed upon effect size in mediation studies, it was not feasible nor meaningful to estimate aggregated effect sizes for the identified mediators which is an important task for future mechanism research. Thus, no funnel plot test for evaluating publication bias was possible. While we can state that a third of the mediation effects were published with non-significant results, the extent of the file-drawer problem—i.e., non-significant results remaining unpublished—is difficult to evaluate.

Despite multiple limitations, our results align with other recent reviews [28, 31, 33], in calling for further investigation and greater methodological rigor in order to identify specific mediators leading to positive change in psychotherapy for adolescents. This identification of significant mediators will enable more reliable conclusions about the underlying mechanisms of change and the possible theoretical models supporting different forms of psychotherapy.

The findings of this review also allow for some tentative and preliminary observations to be considered in developing future treatment models. First, multiple mediators across mediator categories emerged as significant in the treatment of depression, across theoretical treatment models (Table 4). This fits well with the fact that numerous psychotherapeutic approaches, aiming to effect change through somewhat different pathways, have been found to be equally effective for treating depression [84]. The practical implication would be that these different approaches might benefit from further investigating mediators that are not restricted to their particular change theory, in order to clarify the prevalent specific vs. common factors debate [14]. Such future studies could incorporate both the study of transtheoretical mediators such as the ones identified in this review, as well as transdiagnostic mediators pertaining to the particular developmental tasks and challenges of adolescence (e.g., emotion regulation, self-image, peer influence, etc.).

In terms of identifying promising transdiagnostic mediators, most support was found for behavioral and cognitive mediators, followed by family-related mediators. That is, behavioral, cognitive and family-related mediators seemed to have a significant effect on the outcome in psychological interventions across diagnostic categories. While the findings of behavioral and cognitive mediators was perhaps not a surprise given the large number of studies on cognitive-behavioral treatments, the finding on family-related mediators does underline the importance of family and parental involvement in the treatment of adolescents. Accordingly, irrespective of diagnosis, future studies on the psychological treatment of adolescents might include measures of family-related mediators to clarify how families may facilitate or hinder adolescent mental health and recovery from mental illness. Network analyses may also prove useful in helping to elucidate how changes in one domain (e.g. family functioning) may affect changes in another domain (e.g. behavioral changes) or vice-versa [85].

Efforts to identify common, transdiagnostic factors that lead to positive therapeutic change are in line with a dimensional approach to diagnosis and psychopathology, which has been gaining attention (e.g. Caspi et al. [86]), as categorical approaches such as the DSM have come under severe scrutiny and criticism (e.g. [87]). In addition to not finding enough support for diagnosis-specific mediators, our results also indicate little evidence for treatment-specific mediators. While Kazdin’s criteria for mediators imply that true mediators should be treatment-specific, our results indicated that mediators were often significant both in the experimental treatment condition and in the control condition. These findings suggest that the search for specific mediators of specific psychotherapies—as they are currently defined—may often yield little, as potent curative elements may plausibly be shared across treatment and even control conditions [88].

Thus, in line with previous reviews, we advocate that both specific and non-specific mediators and processes should be investigated to validate as well as falsify particular theories of change, including non-psychological such as biological mediators. While the lack of biological mediators is not limited to adolescence research, with a recent review on brain-related functional mediators in adult psychotherapy failing to identify any studies on them [89], studying biological change mechanisms in adolescents’ psychotherapy could be particularly informative as adolescence is a time of rapid and pervasive biological maturation. This would also call for increased interdisciplinary collaboration and openness towards models of change on various interacting explanatory levels as opposed to those operating only on the psychological level, let alone from a particular conceptual model.

To summarize some conclusions for further research, in line with similar initiatives, we see that as each therapeutic approach may address numerous change mechanisms [14], and diagnostic categories may be at the same time too narrow and too broad when looking at change mechanisms [90, 91], a transtheoretical and transdiagnostic turn to investigating more concrete and specific mechanisms of change across diagnostic categories is called for. A similar approach has already been advocated e.g., by the National Institute of Mental Health’s Research Domain Criteria [92], recommending a dimensional rather than categorical approach to understanding mental illness and its early determinants on various levels, from the neural to the behavioral. Likewise, modular therapy approaches (e.g., [93]) with randomization at the level of modules addressing specific change mechanisms may offer better opportunities for identifying change mechanisms than comparing broad therapeutic orientations Another research initiative on specifically behavioral health problems (e.g., smoking, alcohol, substance abuse, poor diet, lack of exercise) has underlined the need to establish a core set of measures going beyond treatment-specific change theories and gauging various developmental challenges and observational perspectives [94]. From our point of view, this core set should also include biological change factors such as heart-rate-variability, cortisol, and other stress-related constructs.

To sum up our key findings, by not being restricted to specific theoretical models or specific diagnoses, our review identified general support for the existence of age-specific mediators pertaining to adolescents (i.e., the significance of family and peer-related mediators). This important finding suggests that rather than attempting to transfer new models from adult psychotherapy to treating adolescents, therapeutic outcomes might be better enhanced by focusing on developing sophisticated age-specific models of therapeutic change. However, research on change mechanisms investigated has rarely addressed the specific developmental challenges in adolescence. Our review suggests that adolescent models of psychotherapy would do well to adopt a developmental perspective which takes into account the many ways in which adolescent psychotherapy differs from the psychotherapy of adults, including the factors and mechanisms that may influence therapeutic change.

By extending a broad net across treatment models and diagnosis, this review has investigated and documented all the different kinds of mediators that have been studied to date in adolescent psychotherapy. This inclusiveness has been fruitful in recognizing that different treatment models may be identifying similar factors at work that lead to positive therapeutic outcomes. There are great benefits to taking a transtheoretical and transdiagnostic approach, as researchers can learn from different models about the variety of changes that can be observed in the process of psychotherapy. Both the field of psychotherapy research and the populations we serve stand to gain from the willingness of researchers to step outside model-specific comfort zones and include in their field of inquiry less familiar ways of thinking about how therapeutic change transpires.