Examining theoretical mediators in an evidence-based and culturally adapted CHW-led intervention for depression and anxiety

Abstract

Objective:

Emerging research suggests that mental health interventions led by community health workers (CHW) are effective and increase access to services for minoritized groups. However, the question of how these interventions work has rarely been tested. In this secondary analysis of data from a randomized clinical trial of a CHW-led intervention that included training in mindfulness and encouraged active engagement and self-efficacy in managing mental health care, we aimed to examine several mediators of change based on theoretical mechanisms of action.

Method:

A sample of 1,044 racially and ethnically diverse adults (mean age 42.6; 83.8% female-identifying) with moderate to severe depression or anxiety symptoms were randomly assigned to the Strong Minds, Strong Communities psychoeducational intervention or an enhanced control condition. Models tested the direct effects of the intervention on outcomes (depression, anxiety, functional impairment) as well as the indirect effects of proposed mediators (mindfulness, patient activation, self-efficacy, and working alliance).

Results:

Multiple mediator models showed consistent, significant indirect effects of mindfulness across all outcomes, suggesting partial mediation. Patient activation was a significant partial mediator for depression only. Results for working alliance were consistently nonsignificant. Exploratory analyses examined moderation by racial/ethnic group (Latinx, Asian, Non-Latinx Black, and Non-Latinx White), finding no significant effects.

Conclusion:

These findings suggest that structured, short-term CHW-led interventions are operating through similar therapeutic processes as those observed in studies using licensed professionals.

Involving lay professionals in mental health services is an efficient approach to expanding the mental health workforce and addressing the shortage of providers for traditionally underserved populations (Barnett, Lau, et al., 2018). Studies demonstrate the benefits of mental health interventions led by lay professionals (for reviews, see (Barnett, Gonzalez, et al., 2018; Gustafson et al., 2025; White et al., 2020), and emerging evidence shows equivalent symptom improvement in peer- versus licensed professional-delivered interventions (Crisanti et al., 2019). There is a paucity of research examining how these interventions work, raising questions about the extent to which interventions previously developed with more intensively trained professionals retain the theoretical “active ingredients” when delivered by lay professionals. Thus, researchers have recommended that studies carefully consider theory-driven mechanisms of action in lay professional-led interventions (White et al., 2020) to allow for further refinement of interventions, balancing the need for fidelity with potential adaptations (Gustafson et al., 2025; Lakin et al., 2023).

A distinct advantage of CHWs is that they share the same language and culture of the population they serve, positioning them to more effectively combat barriers to care such as stigma, language, social determinants, or mistrust of health care providers (Barnett, Lau, et al., 2018). CHWs and other lay professionals have been integrated into mental health service delivery in different ways including outreach and screening (Snell-Rood et al., 2019; Wennerstrom et al., 2015), psychoeducation, and stigma reduction (Hernandez & Organista, 2013), prevention (Tran et al., 2014), and supporting adherence to ongoing mental health care (Allen et al., 2017). While lay professionals in the mental health domain have previously been described as using more informal strategies and providing more general social support (Joo et al., 2016), the landscape is changing. With extensive training, supervision, and attention to fidelity, lay professionals can also provide interventions rooted in evidence-based psychotherapies such as problem-solving therapy (Myers et al., 2019), interpersonal psychotherapy (Patel et al., 2010), and cognitive-behavioral therapy (Stanley et al., 2014). There is a growing interest in “task shifting” in the U.S. as a potential means for increasing access to care, reducing disparities in service receipt, and freeing up resources for those who need more specialized care. However, given the historical focus on “supportive” roles for lay professionals and questions about their perceived credibility (Kanzler et al., 2024), it is important to clarify the mechanisms underlying their effectiveness in delivering evidence-based interventions. This paper is the first, to our knowledge, to examine potential mediators of symptom change in a randomized controlled trial (RCT) of a short-term intervention for depression and anxiety delivered by community health workers (CHWs).

The CHW-led Strong Minds, Strong Communities (SMSC) is a culturally-adapted intervention developed for racial and linguistic minority adults with moderate to severe depression and/or anxiety symptoms that draws on empirically-based coping strategies. CHWs delivered the intervention with high fidelity, and participants reported a significant reduction in symptoms when compared to an enhanced control condition (Alegría et al., 2025). The core components of the program informed the selection of proposed mediators, which included mindfulness, patient activation, and patient self-efficacy. While additional skill modules varied by session (e.g., goal setting, unhelpful thoughts, healthy habits), across the sessions there was a consistent focus on these three core components. While acquisition of skills like cognitive restructuring could also underlie symptom improvement, the brevity with which each of these skills is covered may lessen the impact in comparison to the core components that are present across many sessions.

Proposed mediators

Mindfulness has been shown to be effective in reducing depression and anxiety symptoms both as a stand-alone intervention (Blanck et al., 2018) and when integrated with CBT (Goldberg et al., 2019), and evidence supporting its efficacy extends to Black (Dawson et al., 2022), Chinese (Chan et al., 2020), and Latinx populations (Castellanos et al., 2020). A recent review of 33 studies found that changes in mindfulness significantly mediated the effects of mindfulness-based interventions for anxiety and depression (Johannsen et al., 2022), suggesting that changes in mindfulness are an important “active ingredient.” However, the service providers in these cited studies were highly trained mental health professionals or doctoral students in training; none involved lay professionals. Although the samples in that review were predominantly White (when racial/ethnic composition was described), lay professionals have been trained to deliver mindfulness-based interventions (with favorable outcomes) in Latinx samples (Pérez-Solorio et al., 2025; Wagner et al., 2016). A recent paper (using expert mental health professionals) showed that mindfulness significantly mediated intervention effects on depression and anxiety for Latinx immigrants (Ornelas et al., 2025). While this suggests that the mediation effects may replicate in more diverse samples, the extent to which mediation is observed in lay professional-delivered programs is unknown.

Patient activation refers to being an informed participant in one’s own health care and includes knowledge about self-management of one’s symptoms and active engagement with treatment providers (Hibbard et al., 2004). Increases in patient activation are associated with reduced health care costs and better physical and mental health (Greene et al., 2015; Hibbard et al., 2015), and a review of 29 intervention studies using mental health samples showed consistent pre-post improvements in patient activation levels (Keen et al., 2021). Self-efficacy refers to self-perceptions of mastery and confidence in one’s abilities and is related to, but separable from, self-esteem (Sherer et al., 1982). In the healthcare context, patient self-efficacy refers to confidence in one’s ability to carry out health behaviors or manage a physical or mental health condition. Interventions designed to increase self-efficacy are associated with better physical (Rajati et al., 2014) and mental health outcomes (Zhang et al., 2014). Lay professionals are ideally suited to facilitate patient activation and self-efficacy in the context of behavioral health care, which is seen as a key role of CHWs (Kanzler et al., 2024; Liana & Windarwati, 2021). Therefore, across all SMSC sessions there was a focus on encouraging active engagement in participants’ mental health care and increasing self-efficacy via psychoeducation, learning self-management strategies, facilitating continuity of care with medical providers, and communication training.

Researchers have recommended both patient activation and self-efficacy as important intermediaries of health service outcomes (Hibbard & Greene, 2013). In a randomized trial testing an intervention designed to increase patients’ engagement and communication with their behavioral health provider, brief training in patient-provider communication (e.g., via role plays) led to increases in self-reported patient activation (Alegría et al., 2008), suggesting that health care self-management and engagement is modifiable via intervention and is a promising candidate mediator. Similarly, a randomized trial aimed at increasing self-efficacy in depressed patients with poorly controlled diabetes or coronary heart disease found that improvements in perceived self-efficacy predicted subsequent improvements in depression (Ludman et al., 2013). Self-efficacy significantly mediated the relationship between medical staff support and mental health outcomes in a longitudinal study of SARS patients (Mak et al., 2009). Thus we expected that participants who showed greater improvements in activation and self-efficacy would have better outcomes.

Symptom improvements seen in the SMSC program also may be attributable to the development of a strong working alliance between the participant and CHW, which is considered a core element in the work of lay professionals (Joo et al., 2016; Mundorf et al., 2017). In a recent review of 37 in-person, individual psychotherapy studies (not necessary RCTs) involving a variety of clinical diagnoses, 70.3% found that working alliance was a significant mediator (Baier et al., 2020). While there are fewer studies of working alliance with lay professionals, a study of peer support coaching for social anxiety found that working alliance ratings were equivalent to those of professional therapists (Bautista et al., 2022). A study in which health navigator CHWs were paired with low-income, pregnant women found no relationship between working alliance and reductions in depression or anxiety, but alliance was associated with adherence to study-related visits (Mundorf et al., 2017). However, the extent to which the quality of the working relationship in lay professional-led interventions may be an important mechanism of action is largely unexplored.

The current study

In this secondary analysis of data from the SMSC study, we tested the predicted mediators (mindfulness, patient activation, perceived self-efficacy, and working alliance) in a large and ethnically diverse sample of adults with depression or anxiety symptoms who were randomly assigned to the SMSC program or an enhanced control condition. We hypothesized that increases in all four mediators would be associated with improvements in symptoms and functioning. If there were significant indirect effects associated with more than one of the mediators, we planned to subsequently test a multiple mediator model to determine their significance when examined simultaneously. In addition to examining general mediation effects across the entire sample, we also examined the extent to which mediation effects varied by racial/ethnic subgroup. These moderation effects were explored post-hoc, without a priori hypotheses about subgroup differences, given the lack of prior research in this area.

Method

Participants

All participants provided informed consent and the study was approved by an Institutional Review Board. The sample included 1,034 adults whose primary language was Cantonese, English, Mandarin, or Spanish recruited from 2 U.S. sites (Boston, MA and Greensboro, NC) with a mean age of 42.6 years (SD = 13.3, range from 18 to 86). They predominantly identified as female (n = 867, 83.8%) and were racially and ethnically diverse (Asian n = 137, Non-Latinx Black n = 149, Latinx n = 656, Non-Latinx White n = 92). Participants were required to have moderate to severe symptoms of depression or anxiety; participants with symptoms of bipolar disorder or substance abuse, or those receiving ongoing non-pharmacological mental health treatments, were excluded. More details about participant recruitment and eligibility (including the Consort diagram) can be found in Alegría et al. (2025).

Measures

As noted below, all measures were available in English and Spanish. When Mandarin or Cantonese versions were not available, professional translators were used and multiple iterations were tested for accuracy.

Hypothesized Mediators (Assessed 3 months after starting the intervention or enhanced control)¹

The Mindful Attention Awareness Scale (MAAS) is a 15-item measure of mindful attentional focus and awareness of internal experiences (Brown & Ryan, 2003). High scores are associated with a greater ability to change emotional states and lower self-consciousness and rumination. The MAAS is sensitive to individual differences in mindfulness training (Brown & Ryan, 2003). Data support the validity of the Spanish (Buz et al., 2022; Soler et al., 2012) and Chinese translations (Deng et al., 2012). In the current sample, the MAAS had good internal consistency (omega = .91).

The 9-item Patient Activation Scale (PAS) has been used in previous studies to assess self-management of mental health problems and active engagement with providers (Alegría et al., 2008; Chiang et al., 2019). The PAS has been translated into Spanish and Mandarin with internal consistencies equivalent to that of the English version (in the current study, omega = .88).

The Perceived Efficacy in Patient-Physician Interactions (PEPPI) scale, available in English and Spanish and translated into Chinese for the current study, contains 5 items concerning perceived self-efficacy and confidence in getting one’s needs met when interacting with a health care provider (Maly et al., 1998). In this study, the “provider” referred to the CHW for intervention participants or their primary care provider or physician for control participants. In the current sample, the PEPPI had good internal consistency (omega = .93).

The Working Alliance Inventory (WAI) is a widely-used measure of the therapeutic alliance (Horvath & Greenberg, 1989); a short, 12-item version with a primary general factor (Tracey & Kokotovic, 1989) was used in the current study. The WAI has been adapted and validated for use in Spanish populations (Andrade-González & Fernández-Liria, 2016) and has been translated into Chinese. Items were completed in reference to the CHW (for intervention participants) or their primary care provider or physician (for control participants). The WAI had good internal consistency (omega = .93).

Outcomes (Assessed 6 and 12 months after starting the intervention or enhanced control)

The Hopkins Symptom Checklist (HSCL-25) contains 25 self-report items assessing anxiety (10 items) and depression symptoms (15 items) in the past two weeks (Derogatis et al., 1974). The HSCL-25 has sound psychometric properties in other languages (e.g., (Rodríguez-Barragán et al., 2021), and in the current sample omega = .90 for the anxiety scale and .93 for the depression scale.

The World Health Organization Disability Assessment Scale (WHODAS 2.0) is a measure of health and disability across different domains with higher scores indicating greater disability or functional impairment. It is available in multiple languages and is commonly used to assess the clinical efficacy of interventions in the general population (Üstün et al., 2012). In the current sample, the measure showed good internal consistency (omega = .91).

Procedures

Participants completed baseline sociodemographic information (including age, gender, race, ethnicity, primary language, socioeconomic status, educational level, and employment status) along with other baseline measures and were randomized to either the 10-session Strong Minds, Strong Communities (SMSC) intervention with provider referrals for social determinants (delivered via phone, in-person, or video conference) or an enhanced control condition (4 phone calls over 6 months from a Care Manager to monitor symptoms and provide referrals for social determinants). To increase accessibility and compliance, all contact (including assessments) with participants was in their preferred language (Spanish, 53.7%; English, 35.3%; Mandarin, 6.9%; Cantonese, 4.1%). The SMSC intervention was manualized and consisted of a CHW guide and a participant workbook. CHWs completed 80 hours of intensive training including didactics, role plays, and “practice” cases with feedback from supervisors on all recorded training cases. Demonstration of high fidelity and competence was required before CHWs could be approved for working with study participants. They worked under the close supervision of licensed psychologists and counselors but were otherwise responsible for delivering the intervention. Fidelity was monitored on an ongoing basis throughout the trial (all sessions were recorded), with results showing good adherence to study protocols (Alegría et al., 2025).

All measures were administered orally and responses were recorded by the interviewer (who was blind to the study condition) on a tablet. Mediators were assessed at 3 months and outcomes were assessed at 3, 6, and 12 months (after starting the intervention or enhanced control), and we use the 6-month outcomes for the current analyses. See Alegría et al. (2025) for more information about the study design and procedures.

Sample size and data analytic plan

An a priori power analysis was conducted based on estimated effects of the intervention on both the mediator and outcome of 0.25 (standardized coefficients) and an indirect effect of .0625 (the a and b paths multiplied together). These effect sizes would be considered between small and medium effect sizes (Fritz & MacKinnon, 2007; Kenny, 2025). A sample size of 1044 balanced with equal sample sizes across groups yielded > 80% power to detect these target effect sizes. The analyses of moderation by racial-ethnic group are exploratory and were not considered in the a priori power analysis.

Assumptions of mediation analysis were met. First, the temporal ordering of variables implies in our model in that we regress 6-month outcomes on 3-month mediators, which are then, respectively regressed on baseline outcomes and the binary intervention variable. In addition, tests for linearity suggested no to small departures from linearity in main variable associations. There was also no evidence of multicollinearity as VIF scores among the mediators were in the range of 1.00 to 2.5 (under 5 indicates no multicollinearity).

All analyses were conducted in R (R version 4.4.2 http://www.r-project.org) using the Lavaan package (Rosseel, 2012). We estimated separate models in which anxiety symptoms, depression symptoms and functional impairment were regressed on each of the three-month mediators as well as an effects-coded dummy variable representing the intervention and respective baseline scores for each outcome. Three-month mediators were regressed on the intervention variable. We examined links from the intervention to the mediators (‘a’ paths), links from each mediator to outcomes (in separate models; ‘b’ paths), and indirect effects by creating products of the ‘a’ and ‘b’ paths. Bootstrapped confidence intervals were generated around each indirect effect to evaluate statistical significance with indirect effects having 95% confidence intervals that did not include zero indicative of indirect effects or mediation. Missing data were handled via full information maximum likelihood (FIML). In all models, we specified covariances between outcome scores and the mediators. The mediators were tested first in separate models for each outcome, allowing for more straightforward comparisons of our results with those from other studies. Significant mediators were then examined in multiple mediator models (again, for each of the 3 outcomes) to detect unique indirect effects, controlling for the other mediators. While these data are nested both within site and within CHWs, preliminary analyses indicated no variation across site and very small intraclass correlations and design effects. Consequently, we did not account for nesting in these analyses.

To examine variation in mediation effects across race-ethnicity, we conducted multigroup analyses to test if ‘a’ and ‘b’ paths varied across the Asian American, Black, Latinx, and White participants. We conducted χ² difference tests across models where effects were freely estimated in each race-ethnic group to models where those same effects were constrained to equality (i.e., no group differences). Significant increases in χ² values would indicate that there is variation in intervention and mediator effects across the race-ethnic groups.

Openness and Transparency

The clinical trial from which the secondary data were drawn was registered in Clinicaltrials.gov as NCT04092777. R code used for this paper is posted in PsychArchives (https://doi.org/10.23668/psycharchives.21377). The measures for the mediator and outcome variables are cited. Requests for the SMSC intervention manual can be sent to the corresponding author; all requests will be reviewed by the study team and should include a detailed description of how the manual would be used and a rationale for the request. Because of the sensitive nature of the data collected, raw data are not publicly available. Data may be requested from the corresponding author and may be shared, pending approval by the principal investigators of a rationale and preregistered analysis plan.

Results

Table 1 shows baseline demographic information across the two conditions. Note that a previous paper tested for site differences in intervention effectiveness and there were no significant differences for any of the outcomes (Alegría et al., 2025). The mediators were moderately intercorrelated, with the highest correlations between patient activation (PAS) and self-efficacy (PEPPI; r = .68), working alliance (WAI) and patient activation (r = .58), and working alliance and self-efficacy (r = .63). Mindful attention (MAAS) was modestly correlated with both patient activation (r = .37) and self-efficacy (r = .31) but only weakly with working alliance (r = .18; all correlations were statistically significant at p <.001).

Table 1.

Participant Descriptive Statistics by Experimental Condition

	Intervention (N=524)	Control (N=520)	p-value
Gender
Male	78 (14.9%)	91 (17.5%)	0.29
Female	446 (85.1%)	429 (82.5%)
Race/Ethnicity
AAPI	46 (8.8%)	46 (8.8%)	0.75
Black	69 (13.2%)	80 (15.4%)
AI	2 (0.4%)	1 (0.2%)
White	69 (13.2%)	68 (13.1%)
Latinx	333 (63.5%)	323 (62.1%)
Mixed	3 (0.6%)	2 (0.4%)
Other	2 (0.4%)	0 (0%)
Language of Interview
English	174 (33.2%)	197 (37.9%)	0.36
Spanish	290 (55.3%)	269 (51.7%)
Mandarin	40 (7.6%)	32 (6.2%)
Cantonese	20 (3.8%)	22 (4.2%)
Employment Status
Employed	342 (65.3%)	328 (63.1%)	0.37
Seeking Work	170 (32.4%)	185 (35.6%)
Missing	12 (2.3%)	7 (1.3%)
Education Level
Less than HS	207 (39.5%)	194 (37.3%)	0.05
HS Diploma and Technical School	163 (31.1%)	147 (28.3%)
Some Post HS	58 (11.1%)	89 (17.1%)
College and Post Secondary	93 (17.7%)	90 (17.3%)
Missing	3 (0.6%)	0 (0%)

Note: AAPI = Asian American and Pacific Islander; AI = American Indian; HS = high school

Single mediator models

Single mediator models estimated the effect of the randomized intervention condition (SMSC program or enhanced control) on the mediator variable (MAAS, PAS, PEPPI, or WAI) at 3 months and on anxiety, depression, and functional impairment (HSCL-A, HSCL-D, and WHODAS) at 6 months. Models also included baseline scores on the outcome variable as covariates. In all models, there was a significant effect of the intervention on the outcomes (indicating improvements in depression, anxiety, and functional impairment), and the intervention significantly impacted the mediators in the expected directions (increases in mindfulness, patient activation, working alliance, and self-efficacy). Mindfulness and patient activation scores consistently significantly mediated the intervention effect for symptoms of depression (b = −.030 and −.042, respectively) and anxiety (b = −.031 and −.035, respectively), and functional impairment (b = −.428 and −.436, respectively), while working alliance scores consistently did not. Self-efficacy (PEPPI) scores were significant for depression (b = −.026) and anxiety (b = −.020) but not for functional impairment. Full detailed results from the single mediator models can be found in Supplementary Materials.

Multiple mediator models

The multiple mediator models simultaneously tested all 4 mediators for each of the 3 outcomes (see model illustration in Figure 1 and results in Table 3), again with baseline scores on the outcome variable as covariates. Regarding indirect effects on the three outcomes, results differed slightly across the outcomes. For functional impairment and anxiety, only mindfulness scores emerged as a significant mediator of the intervention effect. For depression, both patient activation and mindfulness scores were significant. In all cases, there were also significant direct effects of the intervention condition on outcomes, suggesting partial rather than full mediation. As with the single mediator models, working alliance was not a significant mediator, and self-efficacy scores were no longer significant when examined in the context of the other proposed mediators. Overall, the intervention demonstrated small-to-medium effects sizes in predicting 6-month outcomes (Table 3 βs between .−.21 and −.27) and medium effect sizes in predicting 3-month mediators (small effect size for mindfulness). Effect sizes associating the mediators and outcomes were mostly small, except for mindfulness which demonstrated medium effect sizes.

Figure 1. Example Multiple Mediators Model for Anxiety Outcome.

Note: Mediators were assessed at 3 months; outcomes (in this figure, anxiety) were assessed at baseline and at 6 months.

Table 3.

Intervention Direct and Indirect Effects in Multiple Mediator Models for Anxiety, Depression, and Functioning

Outcome	Predictor	b	95% Confidence Interval	β
Direct Effects
Anxiety	Intervention	−0.139*	[−0.200, −0.079]	−0.268
Depression	Intervention	−0.159*	[−0.226, −0.094]	−0.272
Functioning	Intervention	−1.808*	[−2.839, −0.852]	−0.208
Mindfulness	Intervention	0.193*	[0.058, 0.321]	0.182
Patient activation	Intervention	0.833*	[0.599, 1.065]	0.475
Patient efficacy	Intervention	0.826*	[0.588, 1.072]	0.462
Working alliance	Intervention	0.612*	[0.459, 0.772]	0.523
Anxiety	Mindfulness	−0.165*	[−0.202, −0.127]	−0.336
Anxiety	Patient activation	−0.028	[−0.059, 0.002]	−0.095
Anxiety	Patient efficacy	0.013	[−0.017, 0.043]	0.044
Anxiety	Working alliance	0.014	[−0.029, 0.059]	0.032
Depression	Mindfulness	−0.198*	[−0.237, −0.159]	−0.360
Depression	Patient activation	−0.037*	[−0.067, −0.006]	−0.110
Depression	Patient efficacy	0.020	[−0.009, 0.049]	0.062
Depression	Working alliance	0.000	[−0.040, 0.042]	0.001
Functioning	Mindfulness	−2.335*	[−2.915, −1.750]	−0.284
Functioning	Patient activation	−0.378	[−0.832, 0.065]	−0.076
Functioning	Patient efficacy	0.276	[−0.141, 0.712]	0.057
Functioning	Working alliance	0.078	[−0.553, 0.744]	0.011
Indirect Effects
Intervention → Mindfulness → Anxiety		−0.032*	[−0.055, −0.009]	−0.061
Intervention → Mindfulness → Depression		−0.038*	[−0.066, −0.011]	−0.066
Intervention → Mindfulness → Functioning		−0.451*	[−0.790, −0.132]	−0.052
Intervention → Patient activation → Anxiety		−0.023	[−0.053, 0.002]	−0.045
Intervention → Patient activation → Depression		−0.031*	[−0.060, −0.005]	−0.052
Intervention → Patient activation → Functioning		−0.315	[−0.719, 0.053]	−0.036
Intervention → Patient efficacy → Anxiety		0.010	[−0.014, 0.036]	0.020
Intervention → Patient efficacy → Depression		0.017	[−0.007, 0.043]	0.028
Intervention → Patient efficacy → Functioning		0.228	[−0.113, 0.613]	0.026
Intervention → Working alliance → Anxiety		0.009	[−0.018, 0.036]	0.017
Intervention → Working alliance → Depression		0.000	[−0.026, 0.025]	0.000
Intervention → Working alliance → Functioning		0.048	[−0.352, 0.460]	0.005

Note: b = unstandardized coefficients;

^*indicates a 95% CI that does not include zero, β = standardized coefficients

R-Square for each outcome was: Working Alliance = 0.068, Patient Activation = 0.056, Patient Efficacy = 0.053.

Mindfulness = 0.008, Depression at 6 months = .453, Anxiety at 6 months = .371, Functioning at 6 months = .411

Multiple mediator by race/ethnicity

Finally, we tested whether there was moderation of the mediation effects by racial/ethnic group (Latinx, Asian, Non-Latinx Black, and Non-Latinx White). Two models were compared using a Chi-square difference test. The first model specified that the ‘a’ pathway (from the intervention to mediator) and ‘b’ pathway (from the mediator to the outcome) were constrained to be equal across the four racial/ethnic groups. In the second model, the ‘a’ and ‘b’ pathways were unconstrained. If the two models differ significantly, this would suggest that there are differences in mediation across the groups. However, results showed that the Chi-square difference test was nonsignificant, indicating that mediation processes did not vary by racial/ethnic subgroup (see Supplementary Materials).

Discussion

Prior results from an RCT showed that the SMSC intervention was more effective than an enhanced control condition in a large sample of racial and linguistic minority adults with depression and anxiety (Alegría et al., 2025). This secondary analysis aimed to investigate whether there were significant indirect effects of core intervention targets. While all the hypothesized mediators were in the expected direction (the intervention was associated with higher patient activation, self-efficacy, mindfulness, and working alliance ratings), only two emerged as significant partial mediators of the intervention effect when examined simultaneously, mindfulness and patient activation. Although there were indirect effects associated with self-efficacy (PEPPI) in the single mediator models, it dropped out in the multiple mediator models, perhaps due in part to its strong correlation with patient activation and working alliance (PAS and WAI).

We found that the improvements in symptoms and overall functioning experienced by participants in the SMSC program are partly attributable to higher mindful attention/focus as well as increased self-efficacy and taking initiative in managing their mental health concerns. The mediation effects were small and indicated partial mediation, which is probably not surprising given that the intervention had several other skill-based components. The use of mindfulness exercises was integrated throughout the 10-session intervention and, across all analyses, mindfulness stood out as the most consistent significant mediator. These results are consistent with a recent meta-analysis showing small to medium mediation effects of mindfulness in interventions for anxiety or depression led by professionals with more years of training in mental health treatment (Johannsen et al., 2022), suggesting (at least indirectly) the integrity of the mindfulness component of the intervention was retained when implemented by CHWs.

Although prior work has underscored the importance of approaches to increase patient activation and self-efficacy to improve health outcomes, it has not previously been tested as a mediator. In our sample, patient activation levels were only associated with depression outcomes, but not with anxiety or overall impairment. Higher patient activation is operationalized in this study as a combination of increased symptom self-management and increased active engagement with mental health care providers. One of the 9 scale items asks specifically about the use of pleasurable activities to improve mood and, while patient activation is not a measure of behavioral activation per se, increases in patient activation may reflect a similar mechanism of action. Thus, our results suggest that CHW encouragement of participants to engage with healthcare professionals about their mental health concerns and to use active coping strategies may be particularly helpful in reducing depression. These results are consistent with substantial literature demonstrating the effectiveness of behavioral activation-based interventions (including culturally adapted approaches) for depression (Cuijpers et al., 2007; Lehmann & Bördlein, 2020).

Given that CHWs have historically been viewed as having a more supportive role, with an emphasis on unique characteristics of CHW (shared language, culture, and community), it is perhaps surprising that the working alliance was not a significant mediator, even when tested in single mediator models. Although the intervention (versus control) was associated with higher working alliance ratings, those ratings were not associated with the outcomes at 6 months. The literature on working alliance as a mediator in treatment outcome studies has been mixed (Baier et al., 2020), and the lack of mediation effect for the working alliance scale in our sample is consistent with the proposal that therapeutic modalities emphasizing the therapeutic relationship are more likely to show mediation (Huibers & Cuijpers, 2015). In a study of a CHW-led program on perinatal care, Mundorf and colleagues (Mundorf et al., 2017) found that alliance was not related to changes in depression or anxiety but was related to program adherence. One possibility is that the importance of the working alliance in CHW-led interventions is not in the intervention efficacy per se (which is driven by the intervention active ingredients) but rather in the individual’s willingness to initiate the intervention (e.g., combating stigma) or in reducing premature discontinuation.

An exploratory question was whether mediation results would differ by racial/ethnic group. Due to lack of prior research examining differential mediation by race-ethnicity, there was no a priori reason to predict specific patterns among subgroups. Still, we recognized the possibility of group differences in the response to certain program components. For example, a recent scoping review showed that among Black American adults, there was a general perception that mindfulness can be helpful, but participants frequently cited religious/spiritual barriers (Haidar et al., 2023).

Our moderation analyses did not reveal group differences in the observed mediation effects. Caution, however, is warranted in interpreting these results. Our sample was unbalanced in terms of race and ethnicity; it was largely Latinx (63.4%), with modest sample sizes (< 15% each) for the other 3 subgroups (Asian, Non-Latinx Black, Non-Latinx White). Therefore, these exploratory analyses were unplanned and not powered to detect subgroup effects, and future research and replication is warranted. Another factor to consider is that the SMSC program involved systematic cultural adaptations, including cultural translation of the manuals and measures for the target ethnic and racial groups and adaptations to account for differences in mental health literacy and sociocultural factors (Alegría et al., 2025). Importantly, CHWs were matched to participants based on their native language and interventions (or enhanced control calls) were delivered in their preferred language. When such cultural adaptations are absent, group differences may emerge.

Overall, our findings suggest that the SMSC program is exerting its effects via theoretically-based mechanisms of action, but again caution in the interpretation of these findings is warranted. The mediation effects were modest, and it should be noted that a statistical mediator is not necessarily a mechanism (Kazdin, 2007). For example, while our mediation results for mindfulness are consistent with prior research, dismantling studies (designed to isolate single intervention components) raise questions about the specific contribution of mindfulness in integrated programs (Blanck et al., 2018). Since the SMSC program involved multiple components, which is common in CHW-led interventions (Gustafson et al., 2025), and our analyses cannot address whether mindfulness capacities changed in direct response to skill acquisition or application, we cannot directly attribute the mediation effect to the mindfulness exercises. Furthermore, the measure of mindfulness used in the present study assessed general state-like mindful awareness in daily life (Brown & Ryan, 2003) as opposed to mindfulness strategy use or home practice.

Despite these limitations, this secondary analysis of mediator effects contributes to the small but growing literature examining the efficacy of lay professional-led interventions for mental health problems. This is one of the first studies to examine theoretically-based mediators in a CHW-led intervention for traditionally underserved participants. These mediation results complement outcome studies of CHW-led interventions and begins to shed light on the important question of whether CHW effectiveness is attributable to specific intervention components. Our findings instill confidence in task shifting models of mental health care, supporting the notion that structured, short-term CHW-led interventions are operating via similar therapeutic processes as those observed in studies using more advanced professionals.

As lay professional-based models for expanding mental health services continue to expand, examination of mediators should be an important research priority. As noted in a recent review of lay professional mental health interventions, understanding which intervention components should be included is an important question that impacts program development and implementation (Gustafson et al., 2025). Identification of key mediators will aid in refining the content and structure of intervention programs and could also help shape lay professional training and supervision in the mental health field. It will also be important to examine outcomes beyond symptom improvement (such as those noted above, stigma endorsement or premature termination), as there may be differential effects of mediators on different outcomes. Finally, we note that the effects of intervention “active ingredients” depends heavily on fidelity, but monitoring fidelity is perhaps one of the most labor-intensive aspects of implementing a program like SMSC. As these programs transition from the research context to the community, efforts to streamline fidelity monitoring are needed.

Table 2.

Study Variable Means, Standard Deviations, and Correlations

Measures	M (SD)	Range	1	2	3	4	5	6	7
1. WAI	5.71 (1.17)	1.75 – 7.00	--
2. PAS	8.04 (1.76)	1.00 – 10.00	0.58**	--
3. PEPPI	8.61 (1.79)	1.00 – 10.00	0.62**	0.68**	--
4. MAAS	4.42 (1.06)	1.33 – 6.00	0.19**	0.39**	0.32**	--
5. HSCL-A	1.61 (0.52)	1.00 – 3.70	−0.12*	−0.26**	−0.16**	−0.47**	--
6. HSCL-D	1.77 (0.58)	1.00 – 3.93	−0.16**	−0.33**	−0.19**	−0.54**	0.79**	--
7. WHODAS	20.31 (8.71)	12.00 – 58.00	−0.12*	−0.25**	−0.14**	−0.44**	0.65**	0.71**	--

Note. WAI = Working Alliance Inventory; PAS = Patient Activation Scale; PEPPI = Perceived Efficacy in Patient-Physician Interactions; MAAS = Mindful Attention Awareness Scale; HSCL = Hopkins Symptom Checklist (A = Anxiety, D = Depression); WHODAS = World Health Organization Disability Assessment Scale.

^*p < .01;

^**p < .001

Public health significance:

CHW-led interventions have the potential to reduce disparities in access to mental health services, but little is known about how these interventions work. In a large sample of ethnically and racially diverse adults, this study showed that a skills-based, CHW-led intervention was effective in increasing mindfulness as well as active engagement in mental health self-management, resulting in subsequent improvements in depression, anxiety, and overall functioning.