This review was preregistered on PROSPERO (Prospective Specific Evaluation of Reviews) (CRD42025641863 [22]). The literature search took place from February 2024 and ended in July 2024. To ensure comprehensiveness, we used three search strategies: database searches and manual searches of reference lists of relevant articles.

The review followed the guidelines of the PRISMA (Preferred Reporting Items for Systematic Reviews and Meta-Analyses) [23]. A comprehensive search was conducted in the following electronic databases: (1) PsycINFO, (2) PubMed, and (3) ScienceDirect. The search strategy was designed to identify both quantitative and qualitative studies focusing on culturally adapted mental health interventions delivered via digital platforms (eg, e-mental health, mobile applications). Search terms were developed using combinations of keywords related to cultural adaptation (eg, “culturally appropriate,” “adapted intervention”) and intervention modality (eg, “digital health,” “mobile app,” “e-mental health”). To ensure breadth and sensitivity, the search strategy incorporated a wide range of related terms. The complete search strategy is provided in Multimedia Appendix 1 [24]. Given that previous reviews included studies up to 2014, only studies published between January 2014 and April 2024 were included. The systematic search process and the rationale for study inclusion and exclusion were documented in accordance with PRISMA standards (see Checklist 1). Two lead authors independently reviewed articles for inclusion, with disagreements resolved through discussion and consensus.

The population, intervention, control, and outcomes model served as the foundation for the creation of the inclusion criteria [25]. People from non-WEIRD societies were referred to as part of the population [12]. This systematic review focuses on DMHIs adapted for non-WEIRD populations, with clearly defined inclusion and exclusion criteria.

Exclusion criteria were excluded if they (1) involved interventions that lack cultural adaptation, (2) targeted individuals under 18 years, (3) were nondigital interventions, (4) were observational studies, case reports, and qualitative studies, and (5) were articles not published in English or outside the 10-year timeframe. By adhering to these criteria, the review will evaluate the impact of cultural adaptations on reducing drop-out rates and the overall effectiveness of these interventions.

After duplicates were removed using EndNote (version 20.3; Clarivate), the remaining records were screened manually using Microsoft Excel. The titles and abstracts were independently screened by two lead authors, based on pre-established inclusion and exclusion criteria. The level of agreement between the screeners was 85% across title/abstract screening, data extraction, and quality assessment stages. Discrepancies were resolved through discussion until consensus was reached.

Two authors independently conducted the quality assessment of all included quantitative studies using the Appraisal Tool for Cross-Sectional Studies [29]. Disagreements were resolved by discussion. Each item was rated as “yes,” “no,” or “do not know,” with scores assigned according to conventions used in previous reviews [30,31]: yes or not applicable (N/A)=1 point; no or do not know=0 points. Total scores ranged from 0 to 20, with studies rated as good (≥15), fair (10-14), or poor (<10).

A total of 184,047 records were identified through the database search. After removing duplicates (n=180,371), 3676 articles remained for title and abstract screening. Of these, 3641 were excluded based on the predefined inclusion and exclusion criteria. The remaining 35 articles were assessed for eligibility by the authors. Eleven studies were excluded at this stage because they were study protocols or review articles, and one study met all inclusion criteria but was excluded from the final review due to inaccessibility; attempts to obtain the full text through institutional subscriptions and direct author contact were unsuccessful. Ultimately, 23 articles were included in the final review. The study selection process is summarized in Figure 1, and detailed study descriptions are available in Multimedia Appendix 2 and Table 1.

Participants (n=4656; see Multimedia Appendix 2) represented diverse cultural and demographic backgrounds. Several studies focused on specific subpopulations like pregnant women who were targeted in two studies [47,60,61], while young adults aged 18 to 30 years were the focus of 39% (8/23) [33,34,44,48,55,57,61,62]. Chinese participants were the most frequently represented cultural group, included in 26% (6/23) of studies [32-34,43,47,57]. University students from various countries were the focus in 21% (5/23) of studies [44,48,55,57,62]. Studies involving Middle Eastern or Arabic-speaking populations accounted for 17% (4/23) [13,54,56,61]. A significant proportion of studies examined Hispanic/Latinx participants (7/23, 30%) [40,46,48,49,58,60,62] and only two studies (2/23, 9%) included Black American participants [50,53], and one study (1/23, 4%) evaluated a DMHI among Indigenous communities [35], highlighting ongoing underrepresentation of these groups in culturally adapted digital mental health research, and one study was conducted on Japanese students [55].

Among the 23 included studies, most investigated DMHIs among participants residing in urban settings (8/23, 35%), typically located near metropolitan areas [32,35,43,46,49,50,55,58]. Most of the studies (14/23, 61%) often relied on community-based recruitment methods such as advertisements, mailing lists, and outreach through community centers [32,35,43,44,46,48,50,53-55,57,58,61,62]. Internet-based recruitment was the second most common strategy, used in 22% (5/23) of studies, primarily through platforms such as social media [44,54,55,57,60]. Seven studies (7/23, 30%) recruited participants directly from outpatient clinical settings located in urban areas [32-35,43,46,53]. A notable proportion of studies (7/23, 30%) recruited through universities and online platforms—to enhance sample diversity and reach [32,34,47,48,53,54,62]. For six studies (6/23, 26%), recruitment settings were not clearly reported, although some recruitment strategies (eg, convenience or snowball sampling) were described [45,47,50,51,56,61].

Across the 23 included studies, participant engagement varied substantially. Attrition rates—defined as loss to follow-up—ranged from 5.3% to 87%, with a median attrition rate of approximately 18.4% among studies reporting this outcome. While some studies demonstrated relatively low attrition (eg, <15%) [46,47,56,57,61], some reported notably high rates (>35%) [44,60,62], and five studies did not state attrition rates [33,34,43,54,58], limiting comprehensive comparison. Attrition rates were reported in 61% (14/23) of studies. Dropout rates, reflecting noncompletion of the intervention, also varied widely, from 0% to 66%, with a median dropout rate of 18.7%. Dropout rates were reported in 17 studies (17/23). Adherence rates, or the proportion of sessions or modules completed by participants, ranged from 26.3% to 100%, with a median adherence rate of approximately 71% in studies that reported these data. Adherence rates were reported in 61% (14/23) of studies. High adherence was reported in programs such as internet-delivered cognitive behavioral therapy (Sleep Healthy Using the Internet for Black women) [50], where over 60% of participants completed all modules. However, 39% (9/23) of studies did not report adherence rates [43,48,51,53-58]

Across the included studies, a wide range of cultural adaptation strategies were employed to enhance the relevance and effectiveness of DMHIs for diverse populations. One of the forms of cultural adaptation used in the studies was language translation, implemented in 57% (13/23) of studies to improve linguistic accessibility [32,33,44,46-48,50,51,54,56-58,60]. The other forms of cultural adaptation included content and imagery adaptations that were mainly used in 70% (16/23) of studies to align with cultural norms, such as visuals and metaphors tailored for specific populations [33,35,44,46-48,50,51,53,54,56-58,60-62]. In 70% (16/23) of studies, cultural values and local practices were integrated into the intervention design, including the incorporation of traditional healing methods for Indigenous groups [33,35,44,46-48,50,51,53,54,56-58,60-62]. Stakeholder involvement—including collaboration with cultural experts, local communities, and leaders—was reported in 48% (11/23) of studies [33,35,44,46-48,50,51,53,54,56-58,60-62]. Iterative feedback and refinement processes—using focus groups, cognitive interviews, and pilot trials—were used in 43% (10/23) of studies to adjust the interventions based on user responses [33,35,44,46-48,50,51,53,54,56-58,60-62]. Only three studies (3/23, 13%) employed the Ecological Validity Framework (EVF), guiding systematic adaptation across multiple cultural dimensions [51,56,61]. Similarly, surface- and deep-structure adaptations—which modify both observable aspects like language and deeper cultural constructs—were applied in 9% (2/23) of studies [44,58]. Technology adaptation to locally preferred platforms (eg, WeChat in China) was reported in 35% (8/23) of studies [32,33,47,56-58,60,61]. However, only one study (1/23; 4%) included cultural competency training for providers to ensure culturally sensitive delivery [53].

The studies included in this systematic review reported various clinical outcomes, focusing on improvements in mental health symptoms, quality of life, and other relevant measures. Most commonly, the studies targeted insomnia and sleep-related issues as primary clinical outcomes (4/23, 17%) [32,50,54,62], followed by depression (12/23, 52%) [32-34,46-49,54,57,60-62] and anxiety (8/23, 35%) [32,45,47-49,54,56,61,62]. Other notable outcomes included significant reductions in pregnancy-related anxiety among pregnant women in China using a digital guided self-help mindfulness-based intervention [47] an automated bilingual digital health tool in the United States significantly reduced binge drinking episodes [49].

Overall, most studies (21/23, 91%) were of high methodological quality. Overall, most studies demonstrated clear research aims and employed study designs that were appropriate and well justified in relation to their objectives. The target populations were clearly defined across all studies. Statistical methods were generally well described. Additionally, the key findings of the studies were usually presented clearly, with discussions and conclusions that were largely justified. Most studies also acknowledged their limitations, enhancing transparency. However, improvements are needed in two studies [58,62] by ensuring representative sampling, justifying sample sizes, addressing nonresponse bias, and transparently reporting dropout data.

This systematic review synthesized findings from 23 RCTs examining dropout, attrition, and adherence in culturally adapted DMHIs among non-WEIRD adult populations. Participant engagement varied widely, with median dropout and attrition rates around 18% and mean adherence of 71%. Interventions using deep, participatory forms of cultural adaptation—combining translation with locally meaningful content, stakeholder involvement, and iterative refinement—showed the highest adherence (often >75%) and lowest dropout (typically <11%). In contrast, interventions limited to surface-level adaptations such as language translation alone frequently exhibited higher dropout (up to 56%) and lower adherence.

Dropout rates ranged from 6% to 87% and appeared to vary depending on adaptation depth [47,60]. Studies integrating multiple culturally grounded elements (eg, language, imagery, values, and delivery context) reported greater retention and engagement. For instance, Zhang et al [32] integrated culturally specific sleep concepts into a CBT-I intervention, achieving a dropout rate of only 6.09%, while Silva et al [40] used culturally resonant telenovela-style content and reported dropout at 8.4%. These findings indicate that culturally resonant content may be linked to greater trust, relevance, and sustained participation.

By contrast, interventions that employed surface-level adaptations—such as translation without deeper contextual integration—or lacked explicit cultural adaptation tended to show higher dropout. Spanhel et al [62], for example, provided a non-adapted English CBT intervention to a diverse population and observed a dropout rate of 56%. Similarly, Zeng et al [33] and Guo et al [34] implemented basic linguistic and platform-level adaptations but reported dropout rates of 49.2% and 41%, respectively. These findings suggest that surface-level efforts were typically associated with lower sustained engagement in culturally diverse populations.

The role of participatory design processes emerged as another important determinant of adherence. Studies like Zhou et al [50] and Lindegaard et al [39] used stakeholder input such as including collaboration with cultural experts, local communities, and leaders and iterative design such as using focus groups, cognitive interviews, and pilot trials, which corresponded to relatively low dropout rates (10.5% and 28%, respectively). However, participatory adaptation alone did not guarantee low attrition, as seen in Barrera et al [60], where despite iterative feedback mechanisms, dropout peaked at 86.97%, possibly due to high geographic and contextual diversity or technological barriers. This highlights the need to complement participatory design with context-sensitive implementation strategies.

An overall pattern emerged in which studies with lower dropout rates were often observed alongside stronger clinical outcomes. For instance, Zhang et al [32], Silva et al [40], and Sun et al [57] demonstrated both high retention and significant reductions in insomnia, depression, or anxiety. However, some studies with moderate to high dropout (eg, Refs [44,48]) still reported clinical improvements among completers, indicating that while adaptation enhances effectiveness, it may not be sufficient to retain all users without additional strategies to address barriers to access and sustained use.

Despite engagement data from 23 RCTs, a meta-analysis was not possible due to heterogeneity in interventions, populations, outcomes, and definitions of engagement, as well as limited extractable data (≈60%) and small moderator subgroups. We therefore share the descriptive synthesis, and future work using standardized metrics may allow meta-regression.

The reviewed studies highlight several strengths of culturally adapted interventions in supporting engagement and clinical outcomes. Many adapted programs demonstrated higher completion and retention rates, such as tailored versions of Sleep Healthy Using the Internet for Black women [50] and Project Using Practice and Learning to Increase Favorable Thoughts for Hispanic adults [46]. Several interventions also reported improved clinical outcomes, including reduced substance use, greater abstinence, and enhanced sleep or mood symptoms. These positive patterns were most often observed in studies incorporating language congruence and cultural values such as communal participation and sensitivity to race-based stressors.

Despite these successes, several limitations emerged. Many pilot trials had small samples, limiting generalizability and highlighting the need for larger, multi-center validation [32,39,46,53,54]. Perceived cultural relevance also varied within target groups; for instance, in [53], although 84% of participants found the adaptation relevant, some viewed it as “overdone,” reflecting within-group diversity and the importance of facilitator racial matching [53]. Some culturally adapted DMHIs faced challenges in sustaining engagement, such as the online mindfulness course for Pakistani students with high attrition [44] and the Egyptian post-traumatic stress disorder intervention whose participants desired more “human” interaction [56]. A further limitation is that a quantitative meta-analysis or meta-regression was not performed. Considerable heterogeneity in study design, intervention type, and outcome measures—along with inconsistent definitions of adherence and dropout and limited extractable numerical data—made statistical aggregation inappropriate. Subgroup counts were also too small for stable moderator modeling. Moreover, standardized mean differences or confidence intervals for clinical outcomes could not be reported, as most studies used heterogeneous measures and lacked sufficient statistical detail. Consequently, clinical outcomes were synthesized narratively to reflect overall improvement of trends across interventions. Future research should standardize engagement metrics and reporting to enable robust meta-analytic and meta-regression approaches that can better quantify determinants of adherence and attrition. Finally, many studies relied solely on self-reported outcomes and unblinded data collection, increasing the risk of bias [53,56]

Drawing from these implications, several key recommendations emerge for future research and practice. First, it is essential to prioritize comprehensive cultural adaptation, moving beyond superficial changes to genuinely embed content and delivery methods within the target culture’s values and sociocultural realities [53]. This includes ensuring language congruence [46] and actively involving community members and cultural experts in the design process to ensure adaptations are relevant and address within-group heterogeneity [53]. Second, to support [22] engagement and retention, hybrid models integrating human support should be considered, as noted by users of a culturally adapted web-based post-traumatic stress disorder intervention for Egyptians who desired more “human” interaction and personalization [56]. Proactive monitoring of engagement metrics is also vital to enable timely re-engagement strategies [33]. Third, given that the perceived cultural relevance can be influenced by the race of the intervention facilitator [53], comprehensive cultural competence and implicit bias training for facilitators are recommended to build trust and address potential microaggressions [53].