Effects of mindfulness-based interventions on depressive symptoms in patients with substance use disorders: a systematic review and meta-analysis

A medical librarian-assisted literature search was performed in 8 English databases including Ovid Medline (1946⁺), Ovid PsycINFO (1967⁺), CINAHL (1937⁺), PubMed (1809⁺), Scopus (1788⁺), ProQuest Dissertation and Theses (1996⁺), Cochrane Library (1995⁺), and Clinical Trials (2000⁺). Eligible studies were identified by searching the databases for the literature collected from their inception till November 2021. We used the Medical Subject Headings (MeSH) term, and search terms are as follows: (mindfulness OR mindful* OR meditation OR meditat*) AND ((substance use disorder*) OR (substance use) OR (drug* addict*)). Truncating these terms with an asterisk allowed for plural and singular results and other related terms. Reference lists were also searched for relevant studies (ancestry studies) and previous literature reviews. Table 1 depicts an example of the search strategy used in this systematic review and meta-analysis. Figure 1 displays the flowchart of eligible studies.

Figure 1.

We included studies that (1) reported MBIs directed only at SUD patients; (2) assessed depression as a quantitative outcome; (3) included a control group to the treatment group (waitlist, usual care group); and (4) were written in English language.

We excluded studies that (1) involved MBIs plus an additional or alternative intervention that potentially affected depression (e.g., Tai Chi, qigong, transcendental meditation [TM], dialectical behavioral therapy [DBT], or acceptance and commitment therapy [ACT]), (2) lacked complete data, and (3) did not provide sufficient data for computing ES.

Two independent researchers (CR and SP) screened titles and abstracts of retrieved citations. We obtained full texts for citations judged as potentially eligible by at least one reviewer. Each reviewer independently assessed full texts against the specified eligibility criteria, and any disagreements regarding eligibility were resolved through discussion additional discussions including the author team (SB and SO).

After reviewing all eligible primary studies, we developed the codebook. Two researchers (CR and SP) revised the codebook after pilot testing with 5 studies. All researchers (CR, SP, SB, and SO) independently extracted and coded information from each study. Data from each primary study included the source of information (e.g., country, year of publication, and publication status), methods included quality of study (e.g., assignment into group, concealed allocation, blinded, intention-to-treat, power of sample, and fidelity of intervention), interventions (e.g., type of MBIs, number of session/week, and duration of each session), participant characteristics (e.g., mean age, sample size of each group, and health conditions), and outcome (e.g., depression instrument, reliability of scale, and mean and standardized depression score). As with the inclusion criteria, all discrepancies were resolved through discussion to achieve consensus.

Two researchers separately coded studies. We compared data for discrepancies. Differences in coding were discussed, and a third researcher was consulted when consensus could not be reached. When primary studies had more than two MBI groups, we coded to not duplicate participants by dividing control participants as equally as possible across comparisons. Thus, from 19 primary studies (s = 19), we coded 20 comparisons (k = 20).

We used SPSS to analyze data to describe the overall characteristics of studies. Then, meta-analysis was performed using Comprehensive Meta-Analysis version 3.0 (CMA). Because depression was measured using multiple ways across the primary studies, we used standardized mean differences of depression scores between the postintervention and control groups using Hedges’ g with 95% CIs. We used Hedges’ g because Cohen’s d has been shown to demonstrate a potential bias, which tends to overestimate the ES with small numbers of primary studies.²⁶ We assumed the value of ESs was normally distributed with heterogeneity due to varying participant, method, intervention, and outcome characteristics. We used the random-effects model—in which CMA weights each study by the inverse of the within- and between-studies variance—to estimate the mean of the true effects. Studies with a higher precision were weighted more heavily than studies with a lower precision.^26,27

We used a forest plot, Q statistic (weighted sum of squares reflecting total dispersion), and the I² statistic (a ratio of real variability across the values of ES) to explore heterogeneity across studies. A forest plot is a graph comparing the values of ESs across studies.²⁶ This graph visually reflects heterogeneity across studies. A significant Q statistic value indicates that heterogeneity exists. The benchmarks for I² are 25%, 50%, and 75%, indicating low, moderate, and high heterogeneity, respectively. If an I² value is near 100%, it means that the most observed variance is true variance.²⁶ When heterogeneity was observed, a moderator analyses, that is, subgroup analyses based on source, participant, method, and intervention characteristics, was used to explore how these various characteristics influenced ES. A meta-analog of analysis of variance (ANOVA) was used for categorical moderators, while meta-regression was used for continuous moderators.²⁶

Primary study quality assessment is essential for performing meta-analysis. Low-quality studies may distort the summary effect estimate.²⁸ In this systematic review and meta-analysis, we used quality indicators as moderators and examined the difference of the values of ES for studies with and without quality indicators. Quality indicators included concealed allocation, random assignment, data collector masking, a priori power analysis, comparison of groups, and intention-to-treat analyses.²⁸ These indicators were analyzed as dichotomous moderators, while the sample size and attrition were analyzed as continuous moderators.^29,30 Table 2 lists the quality indicators for each primary study.

We used the funnel plot, Begg and Mazumdar rank correlation test, and Egger’s bias value for estimating publication bias. A symmetrical funnel-shaped distribution represents the absence of publication bias. A significant value of the Begg and Mazudar rank correlation test and Egger’s bias were used to suggest the presence of publication bias (one tail).²⁹

The 19 primary studies that met our inclusion criteria (s = 19) reported data on 19 between-group comparison (k = 20) and 14 pre-/post-test/control group comparison. All primary studies were published between 2010 and 2021. A total of 1352 participants were included in this systematic review and meta-analysis; among them, 686 participants underwent MBIs and 578 served as controls. Ten primary studies were conducted in the United States,^31-40 3 studies in Iran,^41-43 two studies in Brazil,^44,45 and one study in Turkey,⁴⁶ Canada,⁴⁷ India,⁴⁸ and Thailand, respectively.⁴⁹ Table 3 depicts the demographics of the 19 studies. The mean age of participant ranged from 28.4 years to 50.4 years (mean = 38.6 ± 7.0, s = 15). Female participants represented between 0 and 50.0% of the participants (mean = 24.6%). Primary researchers used 7 different depressive instruments: (1) Depression Anxiety and Stress Scale (DASS-21, s = 2), (2) Beck Depression Inventory (BDI-II, s = 9), (3) Center for Epidemiological Studies Depression (CES-D, s = 4), (4) Overall Depression Severity and Impairment Scale (ODSIS, s = 1), (5) Hamilton Depression Rating Scale (HAM-D, s = 1), (6) Hospital anxiety and Depression Scale (HAD, s = 1), and (7) Brief Symptom Inventory (BSI, s = 1). Higher scores reflect higher levels of depression. The reliability of these scales ranged from 0.66 to 0.93 (s = 8). Eleven primary studies were funded (s = 11). Nine research teams employed adapted mindfulness-based interventions (adapted MBIs), 8 employed MBRP, and one each employed MBSR. Table 4 shows more information about the intervention.

Overall, MBIs had moderate ES to reduce depressive symptoms in patients with SUD (g = 0.67, 95% CI: 0.29, 1.05, I = 89%). Figure 2 displays the forest plot of individual studies; 12 primary studies showed significant values of ES. MBIs grouped into pre-/post-comparisons demonstrated significant reduction in depressive symptom, with a value of ES of.95 for correlated (r = 0.8), and g was 0.90 for uncorrelated groups. The value of ES with the control group showed improvement of depressive symptoms, with a value of 0.28 for correlated (r = 0.8), and g was 0.29 for uncorrelated groups, which suggests that spontaneous recovery of depression was likely to exist. Table 5 depicts the ES of MBI versus control comparisons.

Figure 2.

After we found moderate heterogeneity (I² = 89%), we conducted subgroup analyses. Only one moderator significantly influenced ES: providing a MBI using a mixed format (i.e., individual plus group intervention) had a greater effect (g = 2.13) on reducing depressive symptoms in patients with SUD than providing MBI as a group intervention (g = 0.64) or an individual intervention (g = 0.33, P = 0.034). Recruiting participants using concealed allocation tended to reduce depressive symptoms (g = 1.22) as compared to not using concealed allocation (g = 0.48, P = 0.086). No other quality indicators—such as blinded data collection, intention-to-treat, and fidelity of intervention—were demonstrated to have a moderating influence on the value of the ES. Tables 6 and 7 show subgroup analyses.

The result of the funnel plot was visually asymmetrical, which suggests the possibility of publication bias^26,27 (Figure 3). The Begg and Mazumdar rank correlation test resulted in a Kendall’s tau of 0.391 (P = 0.009), which further suggests the existence of publication bias. In addition, the value of Egger’s regression intercept was 5.66 (95% CI: 2.98, 8.34, P< .001). While the weight of evidence strongly suggests the presence of publication bias, the small number of included primary studies indicates that caution should be used in the interpretation of the potential of publication bias.^26,29

Figure 3.

We highlight the following strengths of our systematic review and meta-analysis. First, we performed a comprehensive database search, which included both published and non-published studies in an effort to identify as many possible sources of relevant data. Second, we conducted a rigorous quality assessment, and we used the results of this assessment for moderator analyses to understand how these variables impacted the value of the ES. Third, we performed subgroup analyses to explore the source of heterogeneity, which has not been reported in prior studies.

We highlight the following weaknesses of our systematic review and meta-analysis. First, we included only primary studies written in English, which may contribute to the risk of publication bias. Second, several primary studies provided only limited information, which indicates that we only had a small number of primary studies upon which we base our analyses. Based upon what we observed in the literature, we urge future researchers to pay attention to possible influence factors (such as types of MBI, guided meditation, and gender of participants) so that the correlation between those factors and the effectiveness of MBIs on depression among patients with SUD can be evaluated.

This systematic review and meta-analysis provides evidence for the use of MBIs for patients with SUD. Specifically, nurses and health care providers might consider using the mixed format intervention, which includes both individual and group interventions. Given the value of the ES we observed, MBIs may be used as adjunctive or alternative complementary treatment to decrease depressive symptoms in patients with SUD, especially when staffing availability limits the number of health care providers. With regard to research implications, we recommend that the future researchers should consider enhanced methodological strategies—such as using concealed allocation to provide the greater reliability in the value of ES. Interestingly, we note that only a few researchers measured depressive symptoms in long-term effects of MBIs, so we would suggest that the further researcher might examine the long-term effects of MBIs on depression.