Introduction

Depression continues to present as a significant concern within mental health across the globe (Patel et al. 2007). Its detrimental effects appear to be especially pronounced for women, for those with lower levels of education, and for individuals resident in poor, low- and middle-income countries (LAMICS) (Tomlinson et al. 2009). While reports on prevalence rates for depression in LAMICS is generally limited, the available data range from 9.7 to 20.5 % (cf., Ali et al. 2002; Hartley et al. 2011; Kleintjies et al. 2006; Nduna et al. 2013). Moreover, the emergence of depressive symptomology in these resource-deprived contexts appears to be associated with poor developmental outcomes for children of depressed mothers (Hartley et al. 2011), increased disability (Kleintjies et al. 2006), drug and alcohol use, and sexual violence (Nduna et al. 2013), factors whose effects are often compounded by contextual circumstances related to poverty, such as overcrowding, inadequate sanitation, and poor nutrition (Rochat et al. 2011).

Some prevalence reports, however, show results that are substantially higher than those just indicated. Studies from South Africa (Hartley et al. 2011; Kleintjies et al. 2006; Manikkam and Burns 2012; Nduna et al. 2013; Rochat et al. 2011), India (Johnson et al. 2015), Bangladesh (Gausia et al. 2009), and Tanzania (Kaaya et al. 2010) report prevalence rates for depression ranging from 33 to 45.5 % in samples of poor, rural or peri-urban-based individuals, which is around four times higher than the estimated prevalence of 7 % to 12 % in high-income countries (HICS).

Thus, there is a clear need for the development and implementation of systematic interventions that address the needs of those suffering from depression in LAMICS. Until recently, though, few efforts in this regard have been advanced, and it also seems that the majority of individuals in these contexts presenting to primary care settings do not receive evidence-based treatment for their depression (Patel et al. 2004). This is in part because most LAMICS have limited resources, human or financial, dedicated towards the advancement of mental health, and their national treatment policies often neglect to make sufficient provision for meeting the mental health needs of its populations.

Additionally, although the evidence supporting the effectiveness of various interventions is robust, the vast majority of these are developed in HICS, and, consequently, matters of cross-cultural validity have been questioned (Patel 2000). Specifically, practitioners have cited concerns about the uncritical adoption of treatments from HICS contexts, without critical considerations of their cultural appropriateness, an awareness of varying health system factors, and a consideration of the cost-effectiveness and feasibility of delivering such treatments, given the particular socio-cultural, economic, and historical contexts in LAMICS (Kagee and Lund 2012).

Cognitive behavioural therapy (CBT), in particular, on the back of extensive empirical evidence, is now promoted as the treatment of choice for depression. Most meta-analytic studies of cognitive-behavioural therapies for adult depression in HICS have found moderate to large effect sizes (Hofmann et al. 2012), but it has been suggested that this may be a considerable overestimation resulting from publication bias (i.e., a bias in regard to positive results being more likely to be published) (Cuijpers et al. 2010).

Additionally, the obtained effect sizes tend to vary depending on the comparison condition/s employed. Specifically, it has been shown that CBT is more effective than waitlists and treatment-as-usual conditions. For example, van Straten et al. (2010), in their meta-analytic investigation, reported a Cohen’s d of 0.42, while Beltman et al. (2010) reported a standard mean difference of 0.83 for their investigation, both effect sizes of moderate magnitude. Oei and Dingle (2008), on the other hand, reported large effects (d ranging from 1.1 to 1.3) for both controlled studies and those with single-group, pre- to post-treatment designs. However, when CBT has been compared to active control conditions, the results have been rather mixed, with some finding it to be superior to other approaches (cf., Jorm et al. 2008) while others have found it to be only equally efficacious to its counterparts (Beltman et al. 2010; Pfeiffer et al. 2011), including when compared to pharmacological treatment (Hofmann et al. 2012).

While CBT’s effectiveness is seemingly well-established in the developed world, there remains a distinct lack of CBT intervention studies in LAMICS. Unlike in the developed world, where international health agencies - such as the National Institute of Clinical Excellence, the American Psychiatric Association, the Movement for Global Mental Health, and the National Institutes of Health - make strong recommendations for the use of interventions with clear empirical evidence to assist practitioners in making treatment decisions, most LAMICS have no such locally-derived guidelines.

Despite this, there is preliminary evidence that CBT is being used in resource-poor contexts with increasing frequency (e.g., Kagee et al. 2004). However, published accounts of its evaluation in these locales appear limited. Recently though, some LAMICS have endeavoured to deliver appropriate, feasible and cost-effective interventions based on evidence generated locally.

The aim of this present study was to investigate, systematically, the current evidence base for the impact of CBT interventions on depression involving interventions specifically developed or modified from existing developed-world treatment protocols for implementation in LAMICS, and to understand the feasibility of applying them in these resource-constrained locales.

Method

Identification and Selection of Studies

A comprehensive search of MEDLINE, Web of Science, PsycINFO, Sabinet, and Academic Search Premier was undertaken to systematically identify literature relating to the implementation and evaluation of CBT in developing countries. Databases were searched from inception until March 2015. The search was restricted to studies that met pre-defined criteria, particularly involving studies that evaluated depression as a primary objective. Where studies assessed a number of other outcomes (e.g., anxiety or somatic symptoms) only the data for depression were extracted. For inclusion, studies were required to implement an intervention program that was primarily cognitive and behavioural in nature, or, when a multi-component intervention package was used, a cognitive therapeutic strategy must have been the primary input.

Both individual and group-administered intervention protocols were included. Importantly, the search was specifically limited to studies conducted with samples from low-income, lower-middle-income, and upper-middle-income countries, as defined by the World Bank criteria. All studies published in peer-reviewed journals in English, and that compared a CBT intervention to one or more control conditions (treatment as usual, a waitlist, or an active comparison condition) were included in the analysis.

Studies with child and adolescent samples (below 18 years of age) were excluded. Unpublished dissertations and case studies were also excluded, as were studies in which patients were not randomly assigned to conditions. Consequently, studies that employed a pre-post study design of a single experimental group with no comparison condition were excluded.

Search terms included combinations of the following: ‘cognitive-behavioral therapy’, ‘cognitive-behavioural therapy’, ‘cognitive therapy’ or ‘CBT’, in combination with the terms ‘depression’, ‘major depression’, or ‘major depressive disorder’ and ‘low-income’, ‘low and middle-income’, ‘LMIC’, ‘LAMIC’, ‘poor’, or ‘developing country’. Additionally, the bibliographies of all retrieved articles were manually searched to identify additional potentially eligible studies. Lead authors of prominent studies were contacted to determine whether there were possible eligible studies that were currently in press and had not yet been published.

The resultant studies were examined by two persons working independently who assessed them for quality and whether they met the inclusion criteria. Any differing decisions regarding particular articles were discussed between both those persons and the first author, and a consensus decision arrived at regarding inclusion or not. The selection and inclusion of studies at each stage of the process is illustrated in Fig. 1.

Fig. 1
figure 1

The identification and selection of studies included in the meta-analysis

Full size image

Data Extraction

Studies assessed to be eligible for inclusion were perused and the following information extracted: authors, country of implementation, sample sizes, type of comparison condition/s, content of treatment program (whether delivered according to an established manual or some other form in which cognitive restructuring was the primary emphasis), treatment format (whether group, individual, or self-help), length of treatment program, the means used to measure change in depression (whether based on clinical interview or a self-report instrument), reported effect sizes (or when these were not reported, means and standard deviations for the index and control groups at pre- and post-intervention).

Estimation of Effect Sizes

For each study, a Cohen’s d effect size was calculated for differences in treatment effect between the index and control conditions. The resultant effect sizes were examined using Cohen’s (1977) criteria for determining magnitude. As such, effect sizes of 0.2 and less are regarded as small, from 0.3 to 0.7 are deemed moderate, and those from 0.8 and above are said to be large. In calculating the effect sizes, if more than one instrument was used to measure the outcome variable (i.e., depression), the mean of the multiple effect sizes was calculated, such that each study contributed only one effect size to the analysis.

When means and standard deviations were not reported, the available statistics (e.g., t, F) were used to calculate effect sizes. Where the statistics reported were not sufficient to conduct the calculation, the respective authors were contacted to provide the data that would facilitate this calculation. If attempts at contacting such authors proved unsuccessful, the articles under question were excluded.

To assess the heterogeneity of effect sizes, the Q-statistic was calculated. A significant Q-value indicates that the null hypothesis of homogeneity can be rejected. An I 2-statistic was also computed as an indicator of heterogeneity in percentages. Increasing values indicate increasing heterogeneity, with values of 0 % indicative of no heterogeneity, 50 % indicating moderate heterogeneity, and 75 % indicating high heterogeneity (Cuijpers et al. 2010).

Considerable heterogeneity between studies was expected. Therefore, the pooled mean effect size was calculated using the random effects model. Random effects models are recommended when employing data from a series where it is assumed that the effect size varies from one study to the next, and where it is unlikely that studies are functionally equivalent. Additionally, random effects models allow statistical inferences to a population of studies beyond those included in the meta-analysis (Field and Gillett 2010).

Finally, a potential limitation of meta-analyses is that overall effect sizes may be overestimated due to a publication bias that favours significant findings. Therefore, a number of analyses [i.e., fail-safe N; a funnel plot; Duval and Tweedie’s (2000) trim and fill procedure; Egger’s (Egger et al. 1997) test of the intercept; and Begg and Mazumdar’s (1994) rank correlation test] were conducted to assess the possibility of publication bias.

All meta-analyses and publication bias analyses were carried out with the Comprehensive Meta-Analysis software developed by Biostat (Borenstein et al. 2009).

Results

Study Characteristics

The initial search using the pre-defined criteria yielded 417 results. An additional 3 articles with publication pending were retrieved from a primary author in the field. The titles and abstracts of the resulting studies were examined. Duplicates and studies that were clearly irrelevant, as well as those that did not meet the inclusion criteria were excluded. Where a decision regarding inclusion could not be made on the basis of the information contained in the abstract, the full-text article was retrieved.

This process resulted in 33 results whose full-text articles were retrieved and examined in greater depth. Twenty-two (n = 22) studies were then excluded for the following reasons: not an empirical paper (n = 11); no comparison condition (n = 5); assessed non-psychological outcome measure (n = 2); and included cognitive psychotherapeutic techniques but only as a minor component of a larger intervention (n = 4). These procedures resulted in a total of 11 eligible studies being identified, which included 842 subjects who received a CBT intervention and 760 subjects who were assigned to control conditions (overall total N = 1602).

The findings from these studies were used for the meta-analysis (see Table 1 for a summary of the essential features of the included studies). Pakistan contributed four studies (Naeem et al. 2011; Naeem et al. 2014; Naeem et al. 2015; Rahman et al. 2008); while South Africa contributed two (Chetty and Hoque 2013; Tshabalala and Visser 2011), and one each from Colombia (Arango-Lasprilla et al. 2014), Turkey (Hamamci 2006), China (Wong 2008), Iran (Faramarzi et al. 2008), and Mexico (Lara et al. 2003).

Table 1 Characteristics of studies contained in the meta-analysis
Full size table

All eleven studies employed randomized and controlled methodologies, with the exception of the Mexican-based study by Lara et al. (2003) where this was not possible. The majority of studies used treatment-as-usual control measures, while two employed active control conditions. Outcomes were assessed using a variety of symptom checklists: the Beck Depression Inventory (Beck et al. 1961) was used in five studies, the depression sub-scale of the Hospital Anxiety and Depression Scale (Zigmond and Snaith 1983) in three; and in one study each, the Hamilton Depression Rating Scale (Hamilton 1960), a modified version of the Patient Health Questionnaire (Kroenke et al. 2001), and the Centre for Epidemiologic Studies Depression Scale (Radloff 1977).

Intervention Characteristics

The interventions varied in content and structure, mode of implementation and treatment length. All treatment protocols were modifications of existing interventions developed and used in HICS. These protocols had been suitably modified to account for cultural and regional needs, appropriate socio-cultural concepts were integrated into the programmes, and linguistic differences were considered in the delivery of treatments. In six studies, treatments were administered as individual interventions and, in four, in a group-administered format, whilst in Lara et al. (2003) a comparison was made between a group and individual format of treatment. The majority of treatment interventions (n = 9) were delivered within the context of primary healthcare centres by trained mental health professionals (e.g., psychologists, social workers, nurses, or psychiatrists) and, for a minority (n = 2), within the community by well-trained lay community workers. Treatment length ranged from 7 to 16 contact sessions.

Meta-Analysis

Pre- to post-treatment effect sizes ranged from 0.05 to 3.05. The pooled effect size of the 11 comparisons between CBT and a comparison condition was 1.12 (95 % CI 0.78–1.46, p < 0.001), which sits within the large range. Heterogeneity was high (Q = 72.34, p < 0.001, I 2 = 86.2 %). Pre- to post-treatment effect sizes for individual studies are displayed in Table 2.

Table 2 Culturally-adapted cognitive-behavioural therapy interventions in LAMICS, effect sizes for pre- to post-treatment change in depression
Full size table

The observed effect sizes corresponded to a z value of 14.92 (p < 0.001) indicating that it would require 627 studies with null results for the two-tailed p value to reach non-significance. Adjustment for publication bias according to Duval and Tweedie’s (2000) trim and fill procedure estimated the adjusted effect size to be smaller than the initial estimate (d = 0.78, 95 % CI 0.44–1.12). Egger’s (Egger et al. 1997) test of the intercept was significant (t = 2.38, df = 9, p = 0.02), but Begg and Mazumdar’s (1994) test was not (τ = 0.33, p = 0.08).

Inspection of the funnel plot for precision (depicted in Fig. 2) indicates that the effect sizes of most of the studies, bar one, were concentrated towards the bottom of the graph and were dispersed across a range of values. This is unsurprising given that the majority of the studies included in the meta-analysis involved relatively small sample sizes, in which greater sampling variation in effect size estimates is likely. While a pair of studies clustered closely around the estimated mean, a larger number of studies were concentrated more on one side of the mean than on the other, which is suggestive of likely publication bias.

Fig. 2
figure 2

Funnel plot of precision by Cohen’s d for depression

Full size image

An additional analysis of effect sizes was conducted with the sample stratified according to treatment format (i.e., whether treatments were delivered in an individual or in a group format) as variations here might explain some of the heterogeneity. This analysis yielded large numerical and statistically different overall effect sizes in favour of individual (d = 0.82, 95 % CI 0.71–0.94, p < 0.001) compared to group treatment (d = 0.55, 95 % CI 0.33–0.78, p < 0.001).

Discussion

The evidence supporting the use of CBT in treating depression in adulthood is well-established in the developed world, despite some contentions that the magnitude of these effects may, to some extent, be overestimated. Meanwhile, in the developing world, the substantial prevalence of depression and the extent of its debilitating consequences, for both the individual and society at large, necessitate the consideration of culturally-adapted CBT treatment protocols for addressing the needs of sufferers in these impoverished locations. Specifically, the extent to which CBT is currently being used in LAMICS is unknown. Where interventions have been adapted and utilized in LAMICS, evaluations have either been absent or lacking in rigor and so a conclusive determination of its overall effectiveness remains undetermined. This review summarized, following a systematic search and meta-analysis, the preliminary evidence for the treatment effectiveness of CBT for adult depression in impoverished, developing world settings.

The pooled overall effect size was relatively large (d = 1.12), indicating that adapted CBT interventions in comparison to control conditions are associated with substantial reductions in depressive symptomology in developing world samples. After adjustment for likely publication bias, the effect size was reduced, albeit still being of large magnitude (d = 0.78). When effect sizes were examined, stratified according to individual and group-administered treatment protocols, comparative results showed individual counselling to be far superior to group-delivered formats.

These effect sizes are of the same order as those found in treatment studies of CBT for depression in the developed world (e.g., Hofmann et al. 2012; Oei and Dingle 2008; Pfeiffer et al. 2011). However, caution should be exercised in comparing these results with those found in HICS, where meta-analytic studies have included a far greater number of trials of high methodological quality. Additionally, we found high heterogeneity in our analyses. This indicates that the studies under review were most likely not functionally equivalent. Potential sources of heterogeneity could be variations in the format of intervention (for example, our analysis suggests that individual counselling yields larger effects than groups), intervention content, study design, choice of control measure, or differing sample sizes.

We believe that the success of these adapted interventions lies in the thoughtful preparation of the adapted treatment programs. A number of interventions were modified, following extensive qualitative study including both mental health professionals and clients with depression diagnoses, to ensure that the planned interventions were acceptable for the context. Naeem et al. (2015), for example, utilized the findings of these feasibility investigations to anticipate potential barriers to engagement with therapy, identify likely helpful and unhelpful techniques, determine the local acceptability of talking therapies, ascertain the extent to which CBT was consistent with the prevailing cultural, religious, familial, and social values, and identify potential obstacles in its delivery. Future studies should take heed of these recommendations and attempt to similarly consider relevant adjustments.

There are a number of indications that the large effect yielded here may perhaps be overestimated. Inspection of the asymmetrical funnel plot clearly shows a conglomeration of studies with smaller sample sizes yielding larger effects positioned to one side of the overall mean, while the few larger studies centred around the mean. This is a clear indication that publication bias is likely, especially when considering the results of the Egger’s test and the Begg and Mazumdar’s rank correlation test. Despite the latter not being statistically significant, this would most likely have reached significance if this present study was sufficiently powered by a larger sample size. These results collectively suggest that publication bias is likely and that the true effect size, following adjustment for this bias, would be closer to the lower end of the large magnitude range.

The results of this analysis, despite being based on a relatively small number of studies, are encouraging in so far as it provides an indication that preliminary efforts to adapt CBT have been successful in reducing the burden of depression in LAMICS and may further contribute to reducing the treatment gap. However, further investigation is required. In particular, studies are needed in which the relative efficacy of individual- versus group-administered counselling is evaluated, as positive evidence related to the latter would be particularly useful in impoverished settings where demand for services is high and resources are low. Our preliminary results, in concurrence with Cuijpers et al. (2011), Huntley et al. (2012), and Lara et al. (2003), suggest that individual counselling may be superior to group formats. This may be because fewer people drop out of individual therapy than from groups (OR = 0.56). In other words, the odds of dropping out of individual therapy are approximately half that of dropping out of group therapy (Huntley et al. 2012).

There is evidence that counselling interventions facilitated by lay counsellors with limited training are effective in meeting the treatment demands in countries where the numbers of professional mental health counsellors are limited (e.g., Bolton et al. 2003), but these kinds of studies were not included in this study as most tend to be multi-component in nature and employ uncontrolled designs. Importantly, in LAMICS settings where large-scale roll-out of mental health interventions is dependent on lay counsellors, it is imperative that effective ways of ensuring competence and assessing fidelity are developed.

Conclusions

We conclude, on the basis of this review and analysis, that CBT interventions whose characteristics and format are suitably adapted to meet the prevailing socio-cultural demands of the context in which they are implemented, but retain as its essence the restructuring of distorted cognition, poses a substantial mechanism for the alleviation of depressive disorders in LAMICS. Specifically, its time-limited nature and cost-effectiveness render it particularly suited to meeting the demands of resource-constrained contexts, where decisions about the allocation of scarce mental health resources is dependent on high quality data about treatment efficacy.

Importantly, treatment decisions in these locales are best made when based on locally-derived evidence. While the number of evaluations of treatment interventions for both depressive and anxiety disorders in developing countries continues to grow, urgent evaluations are also needed for interventions that are frequently used but for which evidence is scant; for example, community interventions are commonly used to intervene with a wide range of psycho-social issues in LAMICS but no evidence-base exists. Additionally, studies are required in which the comparative efficacy of CBT and other psychological approaches shown to be efficacious in treating depression are evaluated; for example, mindfulness-based cognitive therapy; interpersonal psychotherapy; and problem-solving therapy (Linde et al. 2015). While we acknowledge that further investigation of the efficacy of CBT with populations from developing countries is required, the evidence to date, including that of the present study, provide sufficient preliminary evidence in support of the implementation of CBT in reducing the burden of one of the most debilitating disorders in developing countries.