Cognitive behavioural therapy and medication for treatment of adolescent depression: a network meta-analysis

Abstract

Background:

Cognitive behavioural therapy (CBT) and medication are widely accepted and useful interventions for individuals with depression. However, a gap remains in our current understanding of how CBT directly benefits adolescents with depression.

Aims:

The purpose of this study was to examine the short- and long-term effectiveness of CBT only, CBT+Medication, or Medication alone in reducing the duration of major depressive episodes, lessening internalizing and externalizing symptoms and improving global functioning.

Methods:

Data were extracted from 14 unique studies with a total of 35 comparisons. Network meta-analysis was conducted and p-scores, a measure of the extent of certainty that one treatment is better than another, were used to rank treatments.

Results:

There was no significant difference between any two treatments for depression, nor internalizing or externalizing symptoms. For global functioning, CBT had significantly greater effect at the longest follow-up than CBT+Medication. CBT+Medication had the highest p-score for depression, short- and long-term effects, and internalizing and externalizing symptoms long-term effects. No indication of publication bias was found.

Conclusions:

Neither modality, CBT nor medication, is superior for treating adolescent depression. However, CBT was superior in improving global functioning, which is essential for meeting developmental goals.

Introduction

In 2020, an estimated 17% of the U.S. adolescent population had a major depressive episode, with females having a prevalence of 25.2% and adolescents with two or more races having a startling 29.9% prevalence (National Institute of Mental Health, 2022). The prevalence of adolescents diagnosed with a major depressive episode (MDE) has increased significantly from 2005 to 2011, estimating that 1 in 11 reported a MDE (Mojtabai et al., 2016). Globally, adolescent depression has a high disease burden with 34% of adolescents globally at risk of developing clinical depression and females from the Middle East, Africa and Asia having the highest risk of developing depression (Shorey et al., 2022). Adolescent depression has been associated with increased morbidity (GBD 2017 Risk Factor Collaborators, 2018), increased risk of suicide (Maughan et al., 2013), difficulty with school performance (Clayborne et al., 2019), and poor social functioning (Kupferberg et al., 2016). The high disease burden of depression for adolescents is substantial, adversely affecting families and close friends. Half of adolescents who experience a MDE are more likely to have a recurrent episode within five years (Curry et al., 2011) and struggle with current or long-term employment (Clayborne et al., 2019). Thus, providing effective and appropriate therapy for depressed adolescents (behavioural and medication or a combination) is critical.

Cognitive behavioural therapy (CBT) is a widely accepted and useful intervention for adolescents with depression (Das et al., 2016) and is a recommended psychotherapy intervention for child and adolescent depression within the American Pschological Association (2019) and National Institute for Health and Care Excellence (2019) treatment guidelines. CBT has been used with varying effectiveness for adolescents as described in a 2007 meta-analysis (Klein et al., 2007). CBT remains the best psychological intervention for depression compared with interpersonal psychotherapy, for example (Weisz et al., 2017). Often CBT is combined with pharmacological intervention to improve treatment outcomes. However, with the increasing risk of suicide as a major side-effect of many medications (Cipriani et al., 2016), psychological interventions such as CBT are typically first-line treatment. There are many variations with regard to CBT protocols based on contextual factors such as setting (e.g. school, medical clinic, health department), use of pharmacological therapy, precise population of adolescents (e.g. incarcerated youth, high school vs middle school) and it may or may not include the family. A recent meta-analysis found that individual and group CBT are effective for anxiety disorders among children and adolescents (Sigurvinsdóttir et al., 2020). Another meta-analysis concluded that CBT is effective for youth with subclinical depression (Oud et al., 2019). However, a gap remains in our current understanding of how CBT directly benefits adolescents with MDE, and whether combining CBT with pharmacological interventions improves treatment outcomes.

Therefore, the aim of this network meta-analysis is to synthesize new evidence in order to quantify the effectiveness of CBT interventions for adolescents with MDE, with the goal of head-to-head comparison and ranking treatment options with a particular focus on pharmacological interventions. This meta-analysis is innovative and differs from prior reviews given its clear focus on adolescents and direct and indirect examination of treatment effects using an advanced statistical modeling (e.g. network meta-analysis). Specifically, we set out to identify if various treatment options (e.g. CBT only, CBT+Medication, or Medication alone) from current clinical trials have the same short- and/or long-term effects on global functioning, and internalizing and externalizing symptoms in adolescents with MDE.

Method

Literature search and study selection

We first searched the Database of Abstracts of Reviews of Effects (DARE) and the Cochrane Database of Systematic Reviews (CDSR) to identify the presence of similar reviews. Then, the databases PubMed, CINAHL, PsycINFO, Scopus, Cochrane Central Register of Controlled Trials, ClinicalTrials.gov and ProQuest Dissertations were chosen as the primary data sources for this review. Similar to prior meta-analyses, we used an exhaustive approach (Cooper, 2010) for the literature search using a variety of databases with multiple variations of search terms related to CBT in adolescent depression. An academic health centre reference librarian helped build a combination of index and MeSH terms, which was used according to the requirements of each database (see Table S1 in Supplementary material). In addition, we performed several snowball searches based on related previous reviews (Sigurvinsdóttir et al., 2020; Wang et al., 2017). We also conducted ancestral searches in retrieved studies and consulted experts in the field for leads on relevant studies.

To be eligible, the study needed to include (i) an RCT design; (ii) a sample of adolescents aged between 9 and 18 years; (iii) a CBT intervention for adolescents; (iv) a comparison group to be CBT+Medication, or Medication only, (v) feature keywords in the title and/or abstract of full-length publications; and (vi) published in English. No restrictions were applied on publication status or date. The last search was updated in July 2022.

We adhered to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement (Moher et al., 2009) to guide the process of identification, selection and appraisal of the included studies. Each included study was reviewed twice against the study inclusion criteria by two independent reviewers. We developed a structured matrix for abstracting selected studies and building a database including extensive details regarding each research record.

In total, 7043 studies were identified from all databases. Of these, 3605 duplicates were removed, 2982 studies were excluded after title/abstract screening, and 609 studies were removed after full text review due to not meeting inclusion criteria, leaving a total of 83 studies. With the study purpose of comparing: (a) CBT for adolescents (CBT-A) vs CBT-A+Medication, (b) CBT-A vs Medication, and (c) CBT-A+Medication vs Medication, we further excluded studies that included alternative therapy (n=55), attention control (n=17), usual care (n=30) and CBT+placebo (n=1) as the control group. We also excluded another 10 studies that focused on CBT for both adolescents and parents. In case of multiple studies that used the same dataset, we selected the primary study that included the most complete data with the longest follow-up and was usually published in a later year. For example, TADS 2007, not TADS 2004, was selected. Thus, we did not miss important information about long-term follow-up outcomes.

A total of 14 studies were included in the final analyses. Figure S1 in the Supplementary material shows the PRISMA diagram of sample size evolution. These records were also re-reviewed by two independent researchers to ensure they met study inclusion criteria. The final 14 unique studies (with 35 comparison trials) with their descriptive characteristics are presented in Table 1.

Table 1. A summary of characteristics of the 14 studies and their participants

Variables and coding

The effect size analysed in this network meta-analysis was based on Cohen’s d (Cohen, 1988) a standardized mean difference of the outcome between the treatment and comparison groups. Because the research designs in all the 14 included studies (Brent et al., 2008; Byford et al., 2007; Clarke et al., 2005; Clarke et al., 2016; Goodyer et al., 2007; Hilton et al., 2013; Jacobs et al., 2010; Kennard et al., 2008; Kim et al., 2012; March et al., 2007; Melvin et al., 2006; Melvin et al., 2017; Riggs et al., 2007; Wilkinson and Goodyer, 2008) were pre-test–post-test–control group designs, we followed Morris’s alternative approach to obtain an estimate of Cohen’s d using the pooled pre-test standard deviation (Morris, 2008). The cognitive behavioural outcome measures in this network meta-analysis were depression, internalizing and externalizing symptoms (Internal & External), and global functioning (Global). Internalizing symptoms refer to problems of withdrawal, somatic complaints and anxiety, while externalizing symptoms exhibit themselves in delinquent and aggressive behaviour (Levesque, 2011). Global functioning refers to the level of general functioning for adolescents in all areas (at home, at school, and with peers) (Aas, 2011). The lower the outcome measure score, the better the cognitive behaviour. Thus, a negative value of d indicates a greater improvement of the cognitive behavioural outcome from pre-test to post-test in treatment group than that in comparison group. For calculating the effect size, the quantitative information about means and standard deviations was extracted from tables and figures of descriptive statistics reported in the included studies. In case of figures the quantitative information was digitized using the WebPlotDigitizer, version 3.9 (Rohatgi, 2015).

Besides the quantitative information for calculating the effect size, we coded several variables related to characteristics of the trials (e.g. year, authors, location), samples (e.g. age, gender, ethnicity, diagnosis), and methods (e.g. design, setting, measures, comparisons, treatment characteristics). Variables were initially coded by two independent reviewers. The coded variables were then cross-reviewed by two independent reviewers to identify and resolve any conflicts. The study team made several review rounds on the final coded sheet and ensured all studies were coded using the same standards. For example, while initial codes included several measures, researchers aggregated them into three major categories (depression, internalizing and externalizing symptoms, and global functioning).

Data analysis

All the extracted data were entered onto a Microsoft Office Excel spreadsheet for analysis in R. The variance of the effect size was estimated using Morris’s formula (2008) as follows:

$${\sigma ^2}\left( d \right) = 2\left( {c_p^2} \right)\left( {1 - \rho } \right)\left( {{{{n_T} + {n_C}} \over {{n_T}{n_C}}}} \right)\left( {{{{n_T} + {n_C} - 2} \over {{n_T} + {n_C} - 4}}} \right)\left( {1 + {{{d^2}} \over {2\left( {1 - \rho } \right)\left( {{{{n_T} + {n_C}} \over {{n_T}{n_C}}}} \right)}}} \right) - {d^2}$$

where ρ is the population correlation between the pre-test and post-test scores. The value of ρ is usually unavailable in the studies, so it was set as 0.45 as suggested by Morris (2008).

Network meta-analyses were conducted for head-to-head comparisons among three treatment comparators: CBT, Medication, and CBT+Medication. Their effects on three outcome measures: Depression, Internal & External, and Global, were presented in forest plots and tested using fixed-effects models or random-effects models if there was a substantial amount of heterogeneity among effect sizes across the studies tested by Higgins’ I ² (Higgins and Thompson, 2002; Higgins et al., 2003). According to Borenstein et al. (2009), I ² =0–40% may suggest low, 30–60% moderate, 50–90% substantial, and 75–100% considerable heterogeneity. In addition to forest plots of head-to-head comparisons among treatment effects, the treatment ranking p-scores were also estimated for measuring the probability that a treatment is better than the competing treatments (Rücker and Schwarzer, 2015). A treatment with a p-score of 1 is ranked as the best treatment among all the competing treatments, and 0 ranked the worst. All the network meta-analyses within a frequentist framework were conducted using an R package, netmeta (Rücker et al., 2019).

To quantify the overall heterogeneity and inconsistency across the whole network, the DerSimonian–Laird’s τ ² (1986), Higgins’ I ² (2002), and Cochran’s Q _total (1950) were calculated. The Q _total can be further decomposed into Q _{within designs} for assessing the heterogeneity between studies with the same design (i.e. the subset of treatments compared in a study) and Q _{between designs} for assessing the design inconsistency. According to Borenstein et al. (2009), τ ²=0.04 indicates low, 0.09 moderate, and 0.16 high heterogeneity; while I ²=0–40% may suggest low, 30–60% moderate, 50–90% substantial, and 75–100% considerable heterogeneity. A statistically significant Q _total, Q _{within designs} or Q _{between designs} at the α=0.05 level was also used to indicate heterogeneity and inconsistency. In addition to forest plots of head-to-head comparisons between treatment effects estimated from network meta-analysis, the treatment ranking p-scores (Rücker and Schwarzer, 2015) were also estimated for measuring the probability that a treatment is better than the competing treatments. A treatment with a p-score of 1 is ranked as the best treatment and 0 is ranked as the worst, and the mean is always 0.5.

Publication bias

The publication bias was tested using funnel plots (Light and Pillemer, 1984). A symmetric funnel plot with a non-significant Egger test (Egger et al., 1997) suggests no publication bias among the included studies.

Results

The 14 studies with 35 comparison trials involving 2216 unique participants were included in the network meta-analysis. Table 1 shows a summary of the characteristics of the 14 studies and their participants. The mean age of participants in the 14 studies was 15 and 53% of the participants were female. Six studies were non-U.S.-based, and most samples were recruited from clinical settings (93%). Racial and ethnic backgrounds of study samples were not provided in 36% of the studies, and in the remaining nine studies, five included a sample with 25% or greater racial and ethnic minorities. Suicidality was addressed in half of the studies.

Table 2 displays the effect size of each outcome for the 35 trials within the 14 studies. Three pairs of network meta-analyses were conducted, one for each of the three outcomes: Depression, Internal & External, and Global. For each outcome, its short-term effect d ₀ (i.e. immediate effect at the end of intervention) and long-term effect d ₁ (i.e. lasting effect at the longest follow-up) were analysed separately.

Table 2. The effect sizes of each outcome for the 35 trials within the 14 studies

^a d ₀, short-term effect size at the end of intervention;

^b d ₁, long-term effect size at the longest follow-up.

Depression

There were 14 comparison trials within 10 studies (Brent et al., 2008; Clarke et al., 2005; Goodyer et al., 2007; Kennard et al., 2008; Kim et al., 2012; March et al., 2007; Melvin et al., 2006; Melvin et al., 2017; Riggs et al., 2007; Wilkinson and Goodyer, 2008) regarding treatment effect on depression. Ten of the 14 trials also had long-term follow-up ranging from 1 to 12 months. Supplementary Figure S2 displays two networks of direct comparison trials – one for short-term effect the other for long-term effect.

Short-term effect

Figure 1A displays the forest plot of the short-term effects of treatments on depression. The estimated treatment effects were based on a random-effects model because there was substantial overall heterogeneity among the effect sizes across the 14 studies (τ ²=0.18; I ²= 81.9%; Q _total=55.10, d.f.=10, p<.001). There was also significant heterogeneity within designs (Q _{within designs}=46.03, d.f.=8, p<.001) and inconsistency between designs (Q _{between designs}=9.07, d.f.=2, p=.01). The forest plot shows that there was no significant difference between any two treatments because all the 95%CIs covered zero although the treatment ranking p-scores indicated that CBT+Medication had the highest p-score (0.97) followed by Medication (0.34) and CBT (0.18). From the symmetrical funnel plot with a non-significant Egger test (p=.56) (Supplementary material Fig. S3a), we found no indication of publication bias among the 14 short-term trials on depression.

Figure 1. Forest plots of treatment effects on depression. MD, standardized mean difference (d). A, short-term effect (end of intervention); B, long-term effect (longest follow-up).

Long-term effect

Figure 1B displays the forest plot of the long-term effects of treatments on depression. The estimated treatment effects were based on a random-effects model because there were substantial overall heterogeneity among the effect sizes across the 10 studies (τ ²=0.12; I ²=73.6%; Q _total=22.77, d.f.=6, p<.001). There was also significant heterogeneity within designs (Q _{within designs}=17.77, d.f.=4, p<.001) and marginal inconsistency between designs (Q _{between designs}=4.99, d.f.=2, p=.08). The forest plot shows that there is no significant difference between any two treatments because all the 95% CIs covered zero although the treatment ranking p-scores indicated that CBT+Medication had the highest p-score (0.88) followed by CBT (0.42) and then Medication only (0.20). From the symmetrical funnel plot with a non-significant Egger test (p=.69) (Supplementary Fig. S3b), we found no indication of publication bias among the 10 long-term trials on depression.

Internalizing and externalizing symptoms

There were 10 comparison trials within eight studies (Brent et al., 2008; Byford et al., 2007; Clarke et al., 2005; Clarke et al., 2016; Hilton et al., 2013; Kim et al., 2012; Melvin et al., 2006; Melvin et al., 2017) about treatment effect on internalizing and externalizing symptoms. Eight of the 10 trials also had long-term follow-up ranging from 1 to 23 months. Supplementary Fig. S4 displays two networks of direct comparison trials. One network is for short-term effect at the end of intervention and the other network is for long-term effect at the longest follow-up.

Short-term effect

Figure 2A displays the forest plot of the short-term effects of treatments on internalizing and externalizing symptoms. The estimated treatment effects were based on a fixed-effects model because there was no overall heterogeneity among the effect sizes across the 10 studies (τ ²=0; I ²=0%; Q _total=4.47, d.f.=7, p=.72). There was also no heterogeneity within designs (Q _{within designs}=4.33, d.f.=5, p=.50) or inconsistency between designs (Q _{between designs}=0.14, d.f.=2, p=.93). The forest plot shows that there was no significant difference between any two treatments because all the 95% CIs covered zero. Although the treatment ranking p-scores indicated that CBT had a slightly higher p-score (0.68) than CBT+Medication (0.45) and Medication only (0.37), they all cluster around the p-score mean of 0.5 suggesting similar efficacy (Rücker and Schwarzer, 2015). From the symmetrical funnel plot with a non-significant Egger test (p=.18) (Supplementary Fig. S5a), we found no indication of publication bias among the 10 short-term trials on internalizing and externalizing symptoms.

Figure 2. Forest plots of treatment effects on internalizing and externalizing symptoms. MD, standardized mean difference (d). A, short-term effect (end of intervention); B, long-term effect (longest follow-up).

Long-term effect

Figure 2B displays the forest plot of the long-term effects of treatments on internalizing and externalizing symptoms. The estimated treatment effects were based on a fixed-effects model because there was no overall heterogeneity among the effect sizes across the eight studies (τ ²=0.002; I ²=6.4%; Q _total=5.34, d.f.=5, p=.38). There was also no heterogeneity within designs (Q _{within designs}=5.01, d.f.=3, p=.17) or inconsistency between designs (Q _{between designs}=0.34, d.f.=2, p=.85). The forest plot shows that there was no significant difference between any two treatments because all the 95% CIs covered zero, and the treatment ranking p-scores close to the mean of 0.5 also indicated similar efficacy: CBT+Medication (0.62), Medication (0.54), and CBT (0.34). From the symmetrical funnel plot (Supplementary Fig. S5b), we found no indication of publication bias among the eight long-term trials on internalizing and externalizing symptoms. The Egger test was unavailable due to the insufficient number of trials.

Global functioning

There were 11 comparison trials within seven studies (Byford et al., 2007; Clarke et al., 2005; Goodyer et al., 2007; Jacobs et al., 2010; Kim et al., 2012; Melvin et al., 2006; Melvin et al., 2017) about treatment effect on global functioning. Seven of the 11 trials also had long-term follow-up ranging from 1 to 12 months. Supplementary Fig. S6 displays two networks of direct comparison trials. One network is for short-term effect at the end of intervention and the other network is for long-term effect at the longest follow-up.

Short-term effect

Figure 3A displays the forest plot of the short-term effects of treatments on global functioning. The estimated treatment effects were based on a random-effects model because there was moderate overall heterogeneity among the effect sizes across the 11 studies (τ ²=0.04, I ²= 50.5%; Q _total=14.13, d.f.=7, p=.05). There was also heterogeneity within designs (Q _{within designs}=11.85, d.f.=5, p=.04) although no inconsistency between designs (Q _{between designs}=2.28, d.f.=2, p=.32). The forest plot shows that there was no significant difference between any two treatments because all the 95% CIs covered zero although the treatment ranking p-scores indicated that CBT+Medication (0.65) and Medication (0.65) had higher p-scores than CBT (0.20). From the symmetrical funnel plot with a non-significant Egger test (p=.91) (Supplementary Fig. S7a), we found no indication of publication bias among the 11 short-term trials on global functioning.

Figure 3. Forest plots of treatment effects on global functioning. MD, standardized mean difference (d). A, short-term effect (end of intervention); B, long-term effect (longest follow-up).

Long-term effect

Figure 3B displays the forest plot of the long-term effects of treatments on global functioning. The estimated treatment effects were based on a random-effects model because there was moderate overall heterogeneity among the effect sizes across the seven studies (τ ²=0.11; I ²=69.4%; Q _total=13.08, d.f.=4, p=.01). There was also heterogeneity within designs (Q _{within designs}=8.76, d.f.=2, p=.01) although no inconsistency between designs (Q _{between designs}=4.33, d.f.=2, p=.11). The forest plot shows that CBT had a significantly more long-term effect on global functioning than both CBT+Medication (d=–0.89, 95% CI=[–1.51, –0.27]) and Medication (d=–0.94, 95% CI=[–1.61, –0.28]) although there was no significant difference between CBT+Medication and Medication (d=–0.06, 95% CI=[–0.45, 0.33]). Such findings echoed the treatment ranking p-scores which indicated that CBT (1.00) had a much higher p-score than CBT+Medication (0.31) and Medication only (0.19). From the symmetrical funnel plot (Supplementary Fig. S7b), we found no indication of publication bias among the seven trials on global functioning. The Egger test was unavailable due to the insufficient number of trials.

Discussion

The prevalence of adolescent depression is rising (Mojtabai et al., 2016) and its impact on adolescents’ school performance, social functioning, suicide risk, as well as adulthood psychosocial and health outcomes (Clayborne et al., 2019; Keenan-Miller et al., 2007) are far reaching. Therefore, we conducted this systematic review and meta-analysis to evaluate and compare the effectiveness of CBT only, CBT+Medication, or Medication only in reducing duration and frequency of MDE in the adolescent population. This meta-analysis indicates there is no significant difference between any two treatments (CBT only, CBT+Medication, or Medication) for MDE, internalizing and externalizing symptoms in the short- or long-term effects. It is noteworthy that the CBT+Medication p-score (measure of certainty that one treatment is better than the other) for depression short-term effects and long-term effects was 50% larger than the p-scores for CBT or Medication alone. For global functioning, CBT had significantly longer lasting effect at the longest follow-up than CBT+Medication or Medication alone. These findings suggest the importance of CBT in improving global functioning but raises questions about the role of CBT over medication for treating adolescent depression, internalizing, and externalizing symptoms.

Overall, the results were clear. There were no difference in effectiveness of the three comparators for improving internalizing and externalizing symptoms, both in the short and long term. The absence of heterogeneity in effect sizes and quite similar p-scores further strengthens this observation that CBT alone, medication alone, or combination therapy all have similar effectiveness for internalizing and externalizing symptoms. However, there is more uncertainty in the results for MDE, with considerable heterogeneity in effects sizes and substantial differences in the p-scores for combination therapy (CBT+Medication) vs either alone. The uncertainty in our findings is consistent with the literature which shows that response rate for depression in youths treated with psychotherapy is 39% compared with 24% for control groups (Cuijpers et al., 2021), indicating a relatively low response rate. Pharmacological treatment of depression in adolescents is also challenging. A recent meta-analysis examined comparative efficacy of anti-depressants in youth and found that most were not significantly better than placebo (Cipriani et al., 2016), which could re-ignite conversation about the most appropriate first-line treatment for adolescent depression. Our findings suggest that there is room for both modalities in the treatment of adolescent depression, and a need for other therapeutic techniques. Also, a focus on longer term outcomes, such as functioning, may be an important consideration for measuring treatment outcomes.

The results indicate that CBT may not be better in the short term, but there are long-term benefits of CBT over medication for overall functioning. This is somewhat consistent with what could be expected, as CBT focuses on teaching techniques to challenge cognitive distortions (e.g. catastrophizing) and enhance behavioural adaptation (e.g. identifying and coping with difficult emotions) that the adolescent can use in the present and carry forward into the future. As a result, after depressive symptoms, measures of global functioning are the most frequently measured in clinical trials because of their importance in predicting future outcomes. In addition, the growing focus on patient-centred research indicates a need for more attention to general wellbeing outcomes beyond the disease pathology (Krause et al., 2019). Data from 80 youth seeking care in the UK National Health Service identified that the types of goals that young people set when in therapy are often related to functioning such as being able to attend social events and to improve school performance (Bradley et al., 2013). In addition, improving global functioning in adolescents with depression has meaningful implications for meeting future developmental milestones (e.g. establishing career and independent living) and possibly future quality of life (Peters et al., 2016). Follow-up with 196 adolescents in the TADS study (March et al., 2007), found that randomization to any of the three clinical arms (CBT, CBT+Medication, or Medication alone) was associated with improved functioning and success with developmental targets four years later (Peters et al., 2016). However, most meta-analytic work has necessarily focused on symptom improvement and not functioning. Our findings add insight into the benefits of CBT for both the short and longer term, and the broader, whole-person perspective.

The long-term effects of CBT for global functioning suggest that pediatric healthcare providers should establish strong connections with youth-focused psychotherapy services in their area, and consider collaborative models of care that integrate psychotherapy into the primary care setting. Telehealth services could be advantageous for reaching communities with reduced access to traditional mental health services, but research is needed to identify potential barriers (e.g. internet access, mistrust, establishing a therapeutic relationship). Additionally, evidence on models of care that will bring services to adolescents such as mental health models in school are needed. While CBT is the most studied psychotherapy modality, other therapies focused on interpersonal therapy and behavioural activation are promising. CBT that includes components of behavioural activation is associated with better outcomes than CBT alone (Oud et al., 2019) and behavioural activation interventions are more economical to deliver than CBT (Richards et al., 2016). Future research could explore innovative delivery models such as telehealth which may increase access to psychotherapy, and the delivery of behavioural activation interventions using community health workers. Lastly, there remains a need to determine the effectiveness of other psychotherapies compared with medication and combination therapy.

Limitations

Several limitations of this study should be noted. First, we recognize that family-based CBT (parent and adolescent) may be meaningfully different from CBT focused only on adolescents. We originally included evaluation of this additional approach as a study aim. However, due to the lack of needed comparisons to run the analysis, we decided to exclude the studies that focused on CBT for both adolescents and parents. There is a need for future research to examine the benefits and differences in outcomes based on a family-based CBT approach. Similarly, we did not compare other psychotherapies such as interpersonal therapy or behavioural activation. As CBT has the largest literature base, we made the decision to focus on this modality. Second, a lack of recent publications on this topic makes it difficult to assess if depression experiences may differ for the contemporary cohort of adolescents. Third, we were unable to further examine the important contextual factor of suicidality due to the lack of consistent reporting. Incomplete reporting on suicidality could reflect that the publications analysed within this study had a lower burden of symptoms. Fourth, due to the scope of this review study, we needed to exclude some studies with other control treatments (e.g. attention control, usual care, placebo, etc.). This could introduce potential selection bias and might limit the interpretation of our findings. Lastly, there were substantial heterogeneity and inconsistency in studies on depression and global functioning; thus, further investigation is warranted to explain this. One such further investigation would be meta-regression in multivariate meta-analysis with a larger number of studies.

Conclusions

Adolescent depression is on the rise and innovative efficacious treatment models are critically needed. Our findings emphasize the important role of CBT for improving global functioning, but we did not find CBT to be superior to medication for MDE and other outcomes. Improving the overall functioning of adolescents with depression supports their ability to meet future developmental milestones. More research is needed to develop or refine other psychotherapeutic techniques for adolescents. Ideally, newer treatment modalities would translate to community and clinical settings easily to support equitable access for all adolescents with depression.