Systematic review and meta-analysis of effects of standalone digital mindfulness-based interventions on sleep in adults

Abstract

Sleep disturbances and mental health issues represent escalating global health challenges, exacerbated by the COVID-19 pandemic and increased digital media consumption. Digital mindfulness-based interventions (DMBIs) have emerged as a scalable and accessible alternative, but no prior meta-analysis has rigorously isolated the effects of standalone DMBIs on sleep and mental health. We conducted a systematic review and meta-analysis of randomized controlled trials (RCTs) from MEDLINE, Embase, PsycINFO, Web of Science, and the Cochrane Library. Eighteen eligible RCTs were identified, involving a total of 4870 participants for sleep outcomes and 4489 participants for mental health outcomes. Standalone DMBIs significantly improved sleep health with a moderate effect size (Hedges’ g = 0.38, p < 0.001; very low-certainty evidence) and mental health with a moderate effect size (Hedges’ g = 0.33, p < 0.01; very low-certainty evidence). Sensitivity analyses confirmed the robustness of the findings, and meta-regression demonstrated a dose-response relationship between intervention dose and outcomes. Nevertheless, high heterogeneity and publication bias reduced the certainty of evidence to very low levels. Our findings support the potential of standalone DMBIs as a scalable and cost-effective approach to improve sleep and mental health across diverse adult populations. However, further high-quality research is essential to improve evidence certainty, address adherence challenges, and optimize intervention strategies based on delivery format and population characteristics.

Introduction

Sleep is defined in the occupational therapy practice framework as an essential occupation for daily functioning and well-being¹. It supports physical restoration, cognitive processing, and emotional regulation, whereas sleep disturbance compromises these functions and causes broad negative consequences^2–6. A meta-analysis of over 250 studies with about 500,000 participants reported that approximately 40% of adults experienced sleep disturbance during the COVID-19 pandemic⁷. In community-dwelling older adults, the pooled prevalence of poor sleep quality measured with the Pittsburgh Sleep Quality Index (PSQI) was 45% across 41 studies⁸. Beyond prevalence, sleep disturbance doubles the risk of depression in cohort studies and elevates risk for anxiety and cognitive decline, while also contributing substantially to healthcare costs and productivity losses^9,10. These trend, exacerbated by the COVID-19 pandemic, lifestyle changes, chronic stress, and excessive digital media use¹¹, highlights sleep disturbance as a pressing global public health challenge.

Pharmacological interventions are effective for sleep problems¹², but long-term use is constrained by risks such as cognitive impairment, dependency, and adverse effects¹³. Consequently, non-pharmacological interventions have emerged as first-line therapies. Cognitive behavioral therapy for insomnia (CBT-I) is the gold standard with robust evidence across populations¹⁴. Nonetheless, adherence is often poor due to burdensome components like sleep restriction (<3.64 h a day) and difficulties among those with severe comorbidity or low motivation^15,16. These barriers highlight the need for more flexible alternatives.

Mindfulness-based interventions (MBIs) such as mindfulness-based stress reduction (MBSR) and acceptance and commitment therapy (ACT) have emerged as promising alternatives¹⁷. Mindfulness is typically defined as non-judgmental, present-moment awareness including thought, emotion, and bodily sensations. Mindfulness may improve sleep by reducing pre-sleep hyperarousal, lowering rumination, and enhancing emotion regulation^18,19. Meta-analyses of MBIs have reported small-to-moderate improvements in sleep quality (standardized mean difference = 0.30–0.40) across populations¹⁷, supporting mindfulness as a viable non-pharmacological option.

Digital mindfulness-based interventions (DMBIs) deliver training via web or mobile platforms, increasing accessibility²⁰. In prior research, approximately ten randomized controlled trials (RCTs) showed that DMBIs improve sleep quality and reduce distress compared with waitlist or usual care controls (effect size = 0.16–1.14)^21,22, with attrition often lower than in-person programs. However, most studies combined DMBIs with adjunctive elements (e.g., sleep hygiene or cognitive therapy)^23,24, obscuring their independent effects. Although one systematic review exists²², comprehensive meta-analyses (CMAs) are lacking, and prior reviews rarely assess heterogeneity by delivery mode (app vs web), population (clinical vs non-clinical), or comparator type. Consequently, the independent efficacy of standalone DMBIs remains unclear.

To address these limitations, the present study conducted a systematic review and meta-analysis focusing exclusively on RCTs focusing exclusively on standalone DMBIs for sleep disturbance and mental health outcomes in adults. By isolating the digital mindfulness component, we aimed to provide a rigorous evaluation of its independent effects on sleep and mental health outcomes. Furthermore, subgroup and sensitivity analyses were performed to explore potential sources of heterogeneity, while certainty of evidence grading was undertaken to enhance the transparency and reliability of our conclusions. This study, therefore, seeks to generate high-quality evidence that can inform clinical practices and future research regarding the use of standalone DMBIs to address the sleep health crisis.

Results

Study selection

A total of 1473 literature records were identified from the databases. After removing duplicate records, title/abstract screening was conducted on 846 records. Full-text screening for eligibility assessment was conducted on 61 records, and finally, 18 articles were included in this systematic review and meta-analysis (Fig. 1).

Fig. 1. PRISMA flow diagram of included studies.

The diagram displays the number of records identified, screened, assessed for eligibility, and included in the systematic review and meta-analysis.

Characteristics of studies

Table 1 provides characteristics of the 18 included studies. All studies were published between 2012 and 2024. 4870 participants (62% female) were included in the 13 trials, and the total sample sizes per study ranged from 14 to 1255 participants (Table 1). Across studies, the study-level mean age ranged from 17 to 53 years, and the proportion of female participants ranged 33.5%–96.2% Eleven studies targeted populations with clinical conditions—including insomnia, tinnitus, and occupational stress—while the remaining seven studies involved healthy or non-clinical samples.

Table 1.

Characteristics of studies included in the meta-analysis

Authors, year	Sample size	Mean age (year)	Gender (female%)	Population	Mindfulness intervention	Recommended minimum dose	Delivery mode	Control condition	Intervention duration	Follow-up
Boettcher et al.49	E = 44 C = 46	38	71.4%	Tinnitus	MBSR	24 min per day	Website	Online discussion forum	8-week	6-month
Chapoutot et al.50	E = 15 C = 15	48	80.0%	Insomnia with hypnotic dependence	ACT	12 min per day	Video conferencing	Waitlist	8-week	6-month
Hesser et al.51	E = 35 C1 = 32 C2 = 32	48	43.4%	Tinnitus	ACT	24 min per day	Website	C1: Cognitive behavioral therapy C2: Online discussion forum	8-week	12-month
Huberty et al.52	E = 124 C = 139	44	74.9%	Insomnia	MBSR	10 min per day	Mobile app (Calm)	Waitlist	8-week	None
Huberty et al.53	E = 585 C = 444	NR	50.6%	Healthy adults	MBSR	10 min per day	Mobile app (Calm)	Waitlist	8-week	None
Kennett et al.54	E = 14 C = 13	29	96.2%	Insomnia	Mindfulness training	15–30 min per day	Website	Waitlist	6-week	None
Kirk & Axelsen55,	E = 30 C1 = 30 C2 = 30	36	70.0%	Healthy young adults	Mindfulness training	20–30 min per day	Mobile app (Headspace)	C1: music intervention C2: waitlist	10-day	None
Kirk et al.56	E1 = 34 E2 = 34 E3 = 28 C1 = 29 C2 = 38	25	38.6%	High perceived stress (PSS-10)	Mindfulness training	15 min per day	E1-3: Mobile app (Headspace)	C1: audiobook C2: Waitlist	4-week	None
Lahtinen & Salmivalli57,	E = 365 C = 507	17	85.7%	Secondary education students	Mindfulness training	1–5 min per day	Mobile app (Tita)	Waitlist	8-week	3-month
Lappalainen et al.58	E = 41 C = 36	53	63.9%	Insomnia	ACT	Nor reported	Website	Waitlist	7-week	6-month
Li et al.59	E = 167 C = 166	42	78.6%	Insomnia	MBSR	Nor reported	Mobile app (WhatsApp)	Waitlist	3-week	1 and 3-month
Low et al.60	E = 12 C = 11	Nor reported	86.9%	Insomnia	Mindfulness training	10 min per day	Mobile app (Headspace)	Progressive muscle relaxation	40-day or 60-day	None
Mak et al.61	E = 604 C = 651	36.3	74.3%	Healthy young adults	Mindfulness training	7 min per day	Website	Cognitive behavioral therapy	8-week	3-month
Messer et al.62	E = 11 C = 10	51.0	76.0%	Cancer survivors	MBSR	5 min per day	Website	Usual care	6-week	None
Purdie et al.63	E = 27 C = 39	Nor reported	78.7%	Resident physicians	Mindfulness training	24 min per day	Mobile app (UCLA CHORUS)	Waitlist	6-week	None
Querstert et al.64	E = 60 C = 58	40.6	80.5	Adults with work-related affective lamination	MBSR	10 min per day	Website	Waitlist	4-week	3 and 6-month
Remskar et al.65	E = 155 C = 145	27.0	35.5%	Healthy adults	Mindfulness training	10 min per day	Mobile app (Medito)	Audiobook	30-day	2-month
Smith et al.66	E = 7 C = 7	Nor reported	78.5%	Physician assistant students	Mindfulness training	1 min per day	Mobile app (10% Happier)	Waitlist	8-week	None

The DMBIs implemented across the studies varied in content and format. Interventions involved mindfulness training (MT) (n = 9), MBSR (n = 6), and ACT (n = 3). Recommended minimum doses ranged from 1 min to 30 min per day. In terms of delivery platforms, mobile applications were the most common (n = 10), followed by websites (n = 7), and a single study employed video conferencing (Table 1).

Control groups included waitlist (n = 12), active control conditions such as cognitive behavioral therapy, music therapy, or progressive muscle relaxation (n = 4), and non-specific active control conditions such as online forums or audiobooks (n = 5). Intervention durations ranged from 10 days to 8 weeks, with 8-week programs being the most frequently used format (n = 8). On average, intervention length was 6.4 weeks. The average timing of follow-up assessment was 4.7 months after baseline (Table 1).

As detailed in Table 2, adherence data were inconsistently reported, and adherence was typically supported (n = 12) via email (n = 5), short message service (n = 3), video conferencing (n = 2), telephone (n = 1), or app reminders (n = 1). Sleep health outcomes measured across studies included sleep quality (n = 13), insomnia severity (n = 7), sleepiness (n = 3), pre-sleep arousal (n = 3), and sleep time (n = 2) assessed using an objective measure (e.g., actigraphy), validated instruments such as the Insomnia Severity Index (ISI), PSQI, and Epworth Sleepiness Scale (ESS). Mental health outcomes—including depression (n = 11), anxiety (n = 9), stress (n = 4), mood (n = 3), and mental well-being (n = 2)—were measured using tools such as the Beck Depression Index (BDI), Generalized Anxiety Disorder-7 (GAD-7), and Positive Affect Negative Affect Schedule (PNAS). Across studies, the average dropout rate was 15.4% (Table 2).

Table 2.

Adherence and measures characteristics

Author, year	Adherence	Adherence support	Dropout rate	Sleep domain	Sleep measure	Mental health domain	Mental health measure
Boettcher et al.49	Nor reported	None	7.7%	Insomnia severity	ISI	Depression Anxiety	BDI BAI
Chapoutot et al.50	Nor reported	Video conferencing	6.0%	Sleep quality Sleepiness Sleep time	ISI ESS Sleep diary	Depression Anxiety	QD2A STAI
Hesser et al.51	Nor reported	Video conferencing	3.0%	Insomnia severity	ISI	Depression Anxiety	HADS
Huberty et al.52	Average of 14.98 min per day and 6.36 sessions each week	Short message service	4.5%	Sleepiness	ESS	NR	NR
Huberty et al.53	Nor reported	Short message service	26.8%	Insomnia severity Sleepiness	ISI ESS	Depression Anxiety	DASS-21
Kennett et al.54	79%	Email	15.6%	Insomnia severity Sleep arousal	ISI PSAS	Mood	PNAS
Kirk and Axelsen55	>80%	None	9.0%	Sleep quality	D3SQI(s)	Stress	PSS
Kirk et al.56	Nor reported	None	27.5%	Sleep quality	PSQI(s)	Stress	PSS
Lahtinen and Salmivalli57,	At least once per week: 72.5% Near-daily: 39% No practice on almost: 14.1%	App alarm	29.7%	Sleep quality	5-point Likert scale(s)	Depression Anxiety	R-BDI GAD-7
Lappalainen et al.58	Under 1 h per week: 65% 1–2 h per week: 35% 2–3 h per week: 21% Over 3 h per week: 7%	Email	7.2%	Sleep quality Sleepiness	BNSQ ESS	Depression	BDI
Li et al.59	Nor reported	Short message service	24.2%	Insomnia severity Pre-sleep arousal	ISI PSAS	Depression Anxiety Mental well-being	PHQ GAD-7 WHO-WBI

Author, year	Adherence	Adherence support	Dropout rate	Sleep domain	Sleep measure	Mental health	Mental health measure
Low et al.60	Nor reported	None	0%	Insomnia severity Sleep time Pre-sleep arousal	ISI Actigraphy PSAS	Mood	PNAS
Mak et al.61	Nor reported	Telephone	15.8%	Sleep quality	MOSS	Stress Mental well-being	MHI WHO-WBI
Messer et al.62	Multiple times weekly: 64% Three or more: 36% Twice weekly: 28%	Email	8.6%	Sleep quality	PSQI	Depression Mood	HADS POMS
Purdie et al.63	Nor reported	None	4.5%	Sleep quality	PSQI	Depression Anxiety Stress	BDI BAI PSS
Querstret et al.64	Nor reported	Email	26.2%	Sleep quality	PSQI	NR	NR
Remskar et al.65	Nor reported	Email	75.9%	Sleep quality	10-point Likert scale(s)	Depression Anxiety	DASS-21
Smith et al.66	Nor reported	None	0%	Sleep quality	NIH-PROMIS(s)	Depression Anxiety	DASS-21

Effects of standalone DMBIs on sleep health

A total of 46 effect sizes from 18 studies involving a total of 4870 participants were included in the analysis of sleep outcomes. The multilevel meta-analysis demonstrated a statistically significant improvement with a small-to-moderate effect size (Hedges’ g (95% CI) = 0.38 (0.20–0.56; p < 0.001). The heterogeneity test was statistically significant (Q = 65.5; p < 0.05), indicating that variability across studies was not fully attributable to sampling error. Variance decomposition indicated that approximately half of the variance was due to sampling error (49.0%), with the remaining variance distributed across within-study heterogeneity (25.9%) and between-study heterogeneity (25.1%). These results suggest moderate heterogeneity in sleep health, which was further explored through moderator analyses (Table 3 and Fig. 2).

Table 3.

Effects of DMBI on sleep and mental health

Outcomes	Sleep health						Mental health
Main analysis
Overall effects	k	Hedges’ g (95%CI)	Q	I²(%) (Var.1)	I²(%) (Var.2)	I²(%) (Var.3)	k	Hedges’ g (95% CI)	Q	I²(%) (Var.1)	I²(%) (Var.2)	I²(%) (Var.3)
	46	0.38 (0.20–0.56)***	65.5*	49.0	25.9	25.1	41	0.33 (0.10–0.48)**	20.3	100	0	0
Subgroup analysis
1) Population
- Clinical populations	38	0.45 (0.27–0.63)***	61.9**	100	0	0	27	0.38 (0.18–0.57)***	17.5	100	0	0
- Healthy populations	8	0.25 (−0.06 to 0.56)	2.78	100	0	0	14	0.26 (0.04–0.47)*	2.35	100	0	0
2) Delivery method
- Apps	27	0.32 (0.15–0.49)**	18.8	100	0	0	27	0.27 (0.10–0.44)**	14.7	100	0	0
- Websites	11	0.40 (0.11–0.69)**	2.17	100	0	0	12	0.47 (0.19–0.76)**	3.38	100	0	0
- Video conferencing	Not applicable as only one study remained
3) Control group
- Active controls	10	0.34 (−0.11 to 0.78)	12.8	70.0	0	30.0	7	0.10 (−0.18 to 0.38)	1.60	100	0	0
- Non-specific active controls	6	0.39 (0.01–0.77)*	2.10	100	0	0	10	0.55 (0.19–0.92)**	7.26	100	0	0
- Waitlist controls	30	0.40 (0.24–0.56)***	50.5**	57.0	0	43.0	24	0.34 (0.17–0.51)***	7.39	100	0	0
Sensitivity analysis without studies at high risk of bias
Overall effects	35	0.41 (0.25–0.58)***	51.8*	100	0	0	33	0.41 (0.24–0.57)***	14.9	97.8	2.2	0

^*p < 0.05, ^**p < 0.01, ^***p < 0.001, Var.1: sampling error variance, Var.2: within-study variance, Var.3: between-study variance

Fig. 2. Effect of standalone DMBIs on sleep health in adults.

Black squares represent individual study estimates, and the horizontal black line shows 95% confidence intervals. DMBIs digital mindfulness-based interventions, CI confidence interval.

Effects of standalone DMBIs on mental health

For mental health outcomes, 41 effect sizes from 16 studies comprising 4489 participants were analyzed. The multilevel meta-analysis revealed a small but significant benefit of standalone DMBIs (Hedges’ g (95% CI) = 0.33 (0.18–0.48); p < 0.01). The heterogeneity test was not significant (Q = 20.3; p > 0.05), indicating highly consistent effects across studies. Variance decomposition showed that nearly all variability was attributable to sampling error (100%), whereas within-study and between-study heterogeneity were negligible. These results suggest that the effects of standalone DMBIs on mental health were highly consistent across studies, with little residual heterogeneity (Table 3 and Fig. 3).

Fig. 3. Effect of standalone DMBIs on mental health in adults.

Black squares represent individual study estimates, and the horizontal black line shows 95% confidence intervals. DMBIs digital mindfulness-based interventions, CI confidence interval.

Subgroup analyses

The first subgroup analysis investigated population type. For sleep health, standalone DMBIs showed a moderate effect in clinical populations (Hedges’ g (95% CI) = 0.45 (0.26–0.63); p < 0.001), with significant heterogeneity (Q = 61.9; p = 0.006). In healthy populations, the effect was smaller and not significant (Hedges’ g (95% CI) = 0.25 (−0.06 to 0.56); p = 0.111). For mental significant effects were observed in both clinical populations (Hedges’ g (95% CI) = 0.37 (0.18–0.56); p < 0.001;). No heterogeneity was detected (Q = 17.5; p = 0.89). In healthy populations, significant effects were also found (Hedges’ g (95% CI) = 0.25 (0.04–0.47); p = 0.02), with no evidence of heterogeneity (Q = 2.35; p = 0.99) (Table 3).

The second subgroup analysis focused on delivery mode. Studies delivered via video conferencing (one study each for sleep and mental health) were excluded due to insufficient data. For sleep health, website-delivered DMBIs yielded a moderate effect (Hedges’ g (95% CI) = 0.40 (0.11–0.68); p = 0.007), with no evidence of heterogeneity (Q = 2.17; p = 0.99). Mobile apps showed a small-to-moderate effect (Hedges’ g (95% CI) = 0.32 (0.15–0.48); p < 0.001). No heterogeneity was detected (Q = 18.8; p = 0.84). For mental health, website-based interventions produced a moderate effect (Hedges’ g (95% CI) = 0.47 (0.18–0.76); p = 0.001), with no heterogeneity (Q = 3.38; p = 0.98). Mobile app-based interventions showed a small-to-moderate effect (Hedges’ g (95% CI) = 0.27 (0.09–0.44); p = 0.002), again with no significant heterogeneity (Q = 14.7; p = 0.96) (Table 3).

The third subgroup analysis investigated the type of control group. For sleep health, no significant difference was found compared to active controls (Hedges’ g (95% CI) = 0.33, (−0.10 to 0.78); p = 0.13). In contrast, significant effects were detected when compared to non-specific active controls (Hedges’ g (95% CI) = 0.38 (0.01–0.76); p = 0.04), with no evidence of heterogeneity (Q = 2.10; p = 0.83). Significant effects were also observed compared with waitlist controls (Hedges’ g (95% CI) = 0.39 (0.21–0.58); p < 0.001), accompanied by moderate heterogeneity (Q = 50.5; p = 0.008). For mental health, no significant effects were found relative to active controls (Hedges’ g (95% CI) = 0.09 (−0.17 to 0.36); p = 0.47). In contrast, significant effects were found compared to non-specific active controls (Hedges’ g (95% CI) = 0.55 (0.18–0.91); p = 0.004) with no evidence of heterogeneity (Q = 7.26; p = 0.61). Significant effects were also observed relative to waitlist controls (Hedges’ g (95% CI) = 0.33 (0.16–0.51); p < 0.001), with no heterogeneity (Q = 7.39; p = 0.99) (Table 3).

Sensitivity analyses

Among the 18 studies, five were classified as high risk of bias, while the remaining 13 were judged as either low risk or raising some concerns (Fig. 4). Sensitivity analyses were performed by excluding high-risk-of-bias studies from the main analysis that yielded significant results. All significant findings remained robust after exclusion, with effect sizes even increasing (sleep health: from 0.38 to 0.41; mental health: from 0.33 to 0.41) (Table 3). Additionally, Cook’s distance analysis identified 6 outliers in sleep health outcomes, whereas no outliers were observed for mental health outcomes. When these 6 outliers were excluded from the sleep health analysis, the overall effect size remained significant (Hedges’ g (95% CI) = 0.49 (0.28–0.59); p < 0.001) without significant heterogeneity.

Fig. 4. Risk of bias summary for included studies.

The figure displays the overall risk of bias assessments across all included studies using the Cochrane Risk of Bias tool.

The Egger-MLMA regression results prove significant publication bias (sleep health: standard error = 1.641; p = 0.001; mental health: standard error = 0.971; p = 0.048), and the funnel plot exhibited asymmetrical data distribution. However, their effect estimate remained significant even after Trim-and-Fill adjustment, although effect sizes decreased (sleep health: from Hedges’ g = 0.38–0.19; mental health: from Hedges’ g = 0.33–0.21) (Supplementary Figs. 1 and 2).

Certainty of evidence assessment

Based on the Grading of Recommendations Assessment, Development and Evaluation (GRADE) of each significant effect estimate, this study concludes that standalone DMBIs provide a very low-certainty benefit for sleep and mental health. The certainty of evidence for both outcomes was downgraded due to publication bias and a serious risk of bias (Table 4).

Table 4.

Summary of GRADE assessment for each outcome

Outcomes	Hedges’ g (95% CI)	Q	Risk of bias	Inconsistency	Indirectness	Imprecision	Publication bias	Certainty of evidence
Sleep health	0.38 (0.20–0.56)	65.5	Serious (high risk: 5/18)	Serious	Not serious	Not serious	Serious	Very low
Mental health	0.33 (0.10–0.48)	20.3	Serious (high risk: 4/17)	Serious	Not serious	Not serious	Serious	Very low

Meta-regression

Meta-regression analysis showed that the recommended dose significantly impacts both sleep (β (95% CI) = 0.01 (0.00–0.02); p = 0.02) and mental health (β (95% CI) = 0.01 (0.00–0.02); p < 0.001).

Discussion

This systematic review and meta-analysis evaluated the effectiveness of standalone DMBIs on sleep and mental health outcomes among adults. The results demonstrated small-to-moderate effects (SMD = 0.31 for sleep health and 0.32 for mental health) for standalone DMBIs. While these effects were modest, their public health relevance is considerable. Prior research suggests that even small effect sizes, when applied at scale, can lead to substantial population-level health improvements²⁵. Therefore, standalone DMBIs offer a highly scalable and cost-effective solution, particularly important in contexts where access to traditional therapies is limited.

This study is the first meta-analysis to systematically isolate the independent effects of standalone DMBIs, free from the confounding influences of adjunctive interventions like cognitive behavioral therapy or pharmacologic treatment. By exclusively including RCTs, employing rigorous sensitivity analyses, and evaluating the certainty of evidence using the GRADE assessment, the current study provides a more precise and methodologically robust evaluation of the impact of standalone DMBIs on sleep and mental health.

The findings of this study align with a prior systematic review examining standalone DMBIs for sleep health improvement. Previous research systematically reviewed 11 studies and concluded that standalone DMBIs were more effective than non-specific active controls or waitlist controls, but found no significant differences compared to active controls²². However, the prior review did not conduct a meta-analysis and thus could not quantitatively estimate between-group differences. In contrast, the present study provides a quantitative evaluation of the effect of standalone DMBIs through meta-analysis. Furthermore, while the prior review conducted subgroup analyses only based on control group types, this study extended the subgroup, thereby contributing to a more comprehensive understanding and generalizability of standalone DBMIs’ effectiveness. This study also examined adherence strategies across studies. Although only six studies explicitly reported adherence rates, 12 studies described additional strategies to support participant engagement. Prior research has shown that initial engagement with standalone digital interventions is often strong, but maintaining long-term adherence remains a challenge^26,27. In addition, the average dropout rate across studies was 15.4%, with some trials reporting substantially higher attrition. Such variability underscores that adherence difficulties are not isolated to a few studies but represent a systematic barrier to optimizing intervention effectiveness. High dropout may attenuate observed intervention effects, bias outcomes toward more adherent participants, and limit the generalizability of findings. Therefore, a previous study has suggested the potential value of artificial intelligence (AI)-driven personalization and gamification techniques to improve adherence and promote sustained engagement²⁸.

Another novel contribution of this study is its simultaneous evaluation of both sleep health and mental health outcomes. While previous studies have primarily focused on either sleep or mental health outcomes in isolation^17,29, this meta-analysis demonstrates that standalone DMBIs can provide meaningful improvements across both domains. Specifically, standalone DMBIs produced moderate improvements in sleep quality as well as in mental health indicators, including depression, anxiety, stress, and psychological well-being. This finding suggests that improvements in sleep health may positively influence mental health, supporting prior theoretical and empirical models highlighting the bidirectional relationship between sleep disturbance and psychological distress³⁰. The inclusion of both sleep and mental health as outcomes broadens the potential applicability of standalone DMBIs as comprehensive behavioral health interventions. Therefore, this study reinforces the need to consider integrated intervention strategies that target both sleep and mental health simultaneously to maximize clinical and public health benefits.

Subgroup analyses provide further insights into the conditions under which standalone DMBIs are effective. First, population characteristics moderated intervention effects, with individuals with clinical conditions experiencing greater benefits (sleep: Hedges’ g = 0.45; mental health: Hedges’ g = 0.37) than healthy populations (sleep: Hedges’ g = 0.25, not significant; mental health: Hedges’ g = 0.25). This finding aligns with previous research suggesting that individuals with greater severity of baseline health-related symptoms may have a larger potential for measurable improvement^31,32. However, heterogeneity was high in the clinical group, likely because these trials encompassed diverse conditions (e.g., insomnia, tinnitus, and other chronic health problems) with different symptom trajectories. Such variation across the clinical groups may explain the wide dispersion of effects observed in this category³³.

Second, the delivery mode of standalone DMBIs significantly influenced outcomes. Website-based interventions yielded larger effects for both sleep (Hedges’ g = 0.40) and mental health (Hedges’ g = 0.47) compared to mobile app-based interventions (sleep: Hedges’ g = 0.32; mental health: Hedges’ g = 0.27). These results may reflect differences in program comprehensiveness, content depth, and user engagement between web-based and app-based platforms. Specifically, web-based platforms may enhance intervention efficacy by offering more structured content, including multimedia materials, which can promote sustained user engagement. Additionally, web-based interventions generally provide greater technical stability and are accessible across devices, potentially reducing dropout rates and improving overall adherence compared to mobile apps^34,35.

Third, the type of comparator affected effect sizes. Standalone DMBIs produced greater improvements relative to waitlist (sleep: Hedges’ g = 0.39; mental health: Hedges’ g = 0.33) and non-specific active control groups (sleep: Hedges’ g = 0.38; mental health: Hedges’ g = 0.55), whereas effects were non-significant compared with active control conditions. This pattern suggests that some observed improvements may stem from general participation and engagement effects rather than specific intervention components, a phenomenon documented in prior behavior change studies²⁷. Notably, heterogeneity was elevated in comparisons with waitlist groups for sleep outcomes. This pattern is plausible as the waitlist is not a uniform counterfactual: across trials, it ranged from no-contact waiting to minimal-contract check-ins, and the waiting period length also varied. Such differences shape participants’ expectations and behavior (e.g., resentful demoralization, spontaneous help-seeking), which can widen outcome variability between studies. Previous research has shown that waitlist comparators often yield inflated and more variable effects than usual care or active comparators in behavioral trials, underscoring their heterogeneity-inducing nature³⁶.

Taken together, these subgroup findings emphasize that standalone DMBIs are not uniformly effective but context-dependent. Greater benefits appear in clinical populations, web-based delivery formats, and trials using weaker comparator conditions. At the same time, heterogeneity highlights the influence of diverse participant characteristics and comparator design on observed outcomes. Future research should aim to reduce variability by stratifying analyses by specific clinical conditions and employing more consistent comparator conditions. Addressing these sources of heterogeneity will strengthen the precision and applicability of evidence on standalone DMBIs.

Sensitivity analyses strengthen the robustness of this study’s findings. Excluding high-risk-of-bias studies not only maintained the statistical significance of the main results but also reduced heterogeneity. This reduction in heterogeneity suggests that study quality may have contributed to variability in effect sizes and that the main findings are highly stable across lower-risk trials. Cook’s distance analysis further supports the robustness of our findings. Although a small number of studies exerted disproportionate influence on sleep health outcomes, their removal did not alter the overall conclusions. Instead, the consistency and strengthening of the observed effects highlight that the benefits of DMBIs on sleep are reflect a stable pattern across the broader evidence base. In addition to sensitivity testing, this study conducted meta-regression analysis to examine the relationship between the recommended minimum dose and outcomes. The analysis revealed a significant positive association, with higher recommended doses yielding greater improvements for both sleep and mental health outcomes. This finding highlights the importance of intervention dose in the design of DMBIs and suggests that greater exposure and practice time may lead to larger therapeutic effects³⁷. Future research should further investigate optimal dose-response relationships and consider strategies such as adaptive content delivery and AI-driven personalization to maximize adherence and intervention success.

This study has several strengths that enhance its contributions to the literature. First, by exclusively focusing on standalone DMBIs, it provides clear evidence of their potential as scalable and cost-effective strategies to improve sleep and mental health. Second, by concurrently examining both sleep and mental health outcomes within a single analytic framework, this study offers a more comprehensive understanding of the broad impacts of DMBIs. Third, the inclusion of extensive subgroup analyses—exploring population characteristics, delivery modes, and control group types—helps identify conditions under which DMBIs are most effective. Finally, the exclusive inclusion of RCTs, combined with rigorous sensitivity analyses and the GRADE assessment, ensures a high level of methodological quality, setting this study apart from prior studies in the field.

Despite these strengths, some limitations must be acknowledged. First, substantial heterogeneity was observed in the main analysis and some subgroup analyses, suggesting that differences in outcome measurement tools may have influenced the observed effect sizes. Second, the overall certainty of evidence was rated as very low, which indicates that caution is warranted when considering the clinical applicability of these findings. Third, the limited duration of follow-up in most included trials restricts conclusions regarding the long-term sustainability of standalone DMBIs-induced improvements in sleep and mental health outcomes. Finally, factors such as adherence and engagement, which were variably reported and difficult to quantify across studies, may have influenced intervention effectiveness and warrant further investigation.

Future research should address these gaps by conducting long-term follow-up trials to determine the durability of standalone DMBI effects, exploring AI-driven personalization and adaptive intervention delivery strategies to enhance engagement and adherence, and conducting high-quality studies to increase the certainty of evidence. Additionally, the development of standardized frameworks for reporting and evaluating standalone DMBIs will be critical to improve comparability and synthesis across future studies.

In conclusion, this systematic review and meta-analysis provide evidence that standalone DMBIs are effective in improving sleep and mental health in adults, although the certainty of evidence remains very low due to methodological limitations across studies. By isolating the effects of standalone DMBIs and employing rigorous methodological approaches—including RCT inclusion, sensitivity analyses, subgroup analyses, and the GRADE assessment—this study advances the evidence base for DMBIs. Addressing challenges such as long-term adherence, optimizing intervention dose and delivery formats, and improving standardization of intervention reporting will be essential to fully realize the public health potential of standalone DMBIs in promoting sleep and mental well-being.

Methods

This systematic review and meta-analysis follow the Preferred Reporting Items for Systematic Reviews and Meta-analyses (PRISMA) guidelines and were prospectively registered with PROSPERO (ID: CRD420251040033).

Literature search

The electronic databases were searched by the corresponding author on April 28, 2025, using MEDLINE, Embase, PsycINFO, Web of Science, and the Cochrane Library to search for peer-reviewed articles in English, with no start date and no language restrictions. For search terms, three main subject-heading domains were combined with the AND operator in accordance with a previous study: (online OR internet OR digital OR e-health OR m-health OR app OR web OR smartphone OR mobile OR computer OR software) AND (mindful* OR meditate* OR Vipassana OR ACT) AND (sleep OR insomnia)²². In addition, gray literature was searched manually by entering the same keywords into Google and Bing on May 2, 2025. During the preparation of the meta-analysis, the search was repeated three times to identify newly published trials. The last search was conducted on June 15, 2025. After completing the literature search, two independent reviewers screened the titles and abstracts of all retrieved records to assess eligibility. Full-text articles were subsequently obtained for trials deemed potentially relevant and reviewed in detail for inclusion. In addition, the reference lists of all included articles and related review articles were manually examined to identify any supplementary studies.

Study selection

Studies were included if they met the following criteria: (1) participants regardless of age, with either healthy or clinical conditions, (2) the intervention involved MBIs delivered exclusively via digital platforms or technologies, including mobile applications, computer software, websites, video conferencing, or short message service; (3) the intervention placed mindfulness at the core of the program, consistent with the definitional criteria proposed by Crane et al.³⁸, rather than as an ancillary or supportive element. Specifically, eligible interventions were required to provide structured mindfulness training (MT), incorporate sustained meditation practice, be informed by contemplative traditions, and emphasize the development of present-moment awareness with a non-judgmental attitude³⁸; (4) a non-DMBIs control group was used for comparison such as an active interventions, usual care, education, or waitlist conditions; (5) sleep outcomes were assessed through objective measures (e.g., actigraphy), validated measures, or self-reported questionnaires; and (6) the study design was a RCT. Exclusion criteria were as follows: (1) adjunctive interventions (e.g., MBIs plus other interventions) making it difficult to disentangle the effects of DMBIs from other interventions; (2) studies involving participants with severe physical or cognitive impairments, or those with prior meditation experiences; (3) studies lacking sufficient statistical data to be included in the meta-analysis; and (4) studies in which the control group received MBIs through in-person delivery. Both peer-reviewed articles and conference abstracts were included in the main analysis, although conference abstracts were required to report RCT findings explicitly.

After eliminating duplicate records, a two-phase screening procedure was implemented: initial screening of titles and abstracts, followed by a full-text review. Two independent reviewers (S.-A. Lee and J.-H. Park) assessed the eligibility of each record at both stages. To maintain consistency across the process, a random 10% subset of entries was jointly reviewed during each screening phase to calculate inter-rater agreement. Any disagreements were resolved through discussion, and if needed, a third senior reviewer acted as the final arbiter. Inter-rater reliability, quantified using Cohen’s Kappa, indicated near-perfect agreement at both the title/abstract screening stage (0.94) and full-text review (0.92). Reviewers remained blinded to each other’s assessments throughout to ensure objectivity. Although gray literature sources were screened, none fulfilled the eligibility criteria, and thus no gray literature was included in the final analysis.

Data management and extraction

From each included study, the following information was extracted when available: (1) bibliographical data (i.e., first authors(s), year), (2) sample characteristics (i.e., total sample size, mean age, population), (3) intervention characteristics (i.e., details of intervention and control conditions, type of mindfulness techniques, minimum recommended dose, mode of delivery, duration, outcome domains), (4) data required to calculate within-group or between-group effect sizes for RCTs. When essential data were missing, corresponding authors were contacted for clarification. If the authors did not respond or provide insufficient data for quantitative synthesis, the study was excluded from the meta-analysis.

Intervention and outcomes

Included interventions were further categorized into three groups based on an established classification framework from previous studies^22,39,40: MBSR, ACT, and MT. Interventions were classified as MT when they included mindfulness meditation or related practices but did not explicitly follow standardized branded protocols such as MBSR or ACT.

The primary outcomes included sleep-related indicators such as insomnia severity and sleep quality assessed using either objective methods, validated subjective instruments, or self-reported questionnaires. Secondary outcomes included indicators of mental health, such as levels of depression, anxiety, perceived stress, and psychological well-being.

Quality assessment

The GRADE framework was used to evaluate the certainty of evidence across five domains: risk of bias, imprecision, inconsistency, indirectness, and publication bias³⁵. Depending on the quality assessment within these domains, the certainty of evidence could be either downgraded or upgraded. The final classification for each outcome was assigned to one of four levels: “Very low”, “Low”, “Moderate”, or “High”⁴¹.

Risk of bias in individual studies was appraised independently by two independent reviewers using the Cochrane Collaboration’s Risk of Bias 2 tool⁴² guidelines in five domains: randomization process, deviations from intended intervention, missing outcome data, measurement of the outcome, and selection of the reported result. A study was categorized as “High Risk” if at least one domain was rated high risk. If there were no high-risk domains, but at least one domain raised concerns, the study was rated as “Some Concerns”. Only studies with low risk in all domains were designated as “Low Risk”.

Publication bias

The presence of publication bias was assessed using funnel plots and the Egger-multilevel meta-analysis (MLMA) regression methods, which account for the statistical dependencies introduced by multiple correlated effect sizes within studies and thereby provide more reliable Type I error control compared to traditional Egger’s regression⁴³. Funnel plots were used as an initial visual tool, with symmetrical distributions clustering suggesting the absence of substantial bias. In case where significant asymmetry was detected, the trim and fill method is applied to adjust⁴⁴.

Tests of heterogeneity

Overall residual heterogeneity was tested using the Q statistic. Following Higgins et al., I² values of 25%, 50%, and 75% were interpreted as low, moderate, and high heterogeneity, respectively⁴⁵. Variance components at Levels 2 and 3 were also estimated to identify the distribution of heterogeneity across within- and between-study sources.

Meta-analysis

Effect sizes were calculated as Hedges’ g, a small-sample bias–corrected standardized mean difference. Values of 0.20, 0.50, and 0.80 were interpreted as small, moderate, and large effects, respectively⁴⁶. Means, standard deviations, and sample sizes were extracted from each study to compute effect sizes; in cases where other statistics (e.g., correlation coefficients or F-values) were reported, these were converted into Hedges’ g using established procedures. To ensure comparability across trials and to avoid statistical dependence due to repeated assessments, only post-intervention values were included in the analyses. For measures where lower scores indicated better outcomes, values were adjusted so that positive effects consistently reflected beneficial intervention effects.

All computations were performed using CMA software (CMA, version 2.2) and R (version 4.5.1), employing the metafor package for three-level modeling.

Model selection

Because several studies reported multiple effect sizes (e.g., from different outcome domains, or multiple control groups within a single trial), the assumption of statistical independence across effect sizes would have been violated in a conventional two-level random-effects meta-analysis. To address this issue, we employed a three-level random-effects meta-analysis model⁴⁷, which partitions the total variance into three components. Specifically, Level 1 corresponds to sampling error variance, representing the variability attributable to the standard errors of the observed effect sizes; Level 2 represents within-study variance, reflecting heterogeneity among multiple effect sizes extracted from the same study; and Level 3 represents between-study variance, capturing heterogeneity among effect sizes originating from different studies. By explicitly modeling these variance components, the three-level approach accounts for the statistical dependencies among effect sizes, thereby providing more accurate variance estimation and improving both the precision and the power of the analyses.

Subgroup analyses and sensitivity analyses

Subgroup analyses were conducted for both primary and secondary outcomes based on (1) participant health status (healthy vs unhealthy conditions), (2) mode of intervention delivery (e.g., app, website, and video conferencing), and (3) type of control group (e.g., active, non-active, and waitlist).

Sensitivity analyses were performed using two approaches: first, by excluding studies at high risk of bias to evaluate the impact of study quality, and second, by employing Cook’s distance to assess the influence of potential outliers and ensure the robustness of the findings⁴⁸.

Meta-regression

To further explore potential sources of heterogeneity, we conducted meta-regression analyses. In particular, we examined whether intervention dose moderated the effects of standalone DMBIs. We operationalized dose as the daily practice duration (min per day) prescribed in each intervention. The three-level random-effects model framework was extended by including intervention dose as a continuous moderator. This approach allowed us to assess whether greater prescribed exposure was associated with stronger effects on sleep and mental health outcomes.

Author contributions

Conceptualization: S.-A.L. and J.-H.P. Methodology: S.-A.L. and J.-H.P. Investigation: S.-A.L. and J.-H.P. Data curation: S.-A.L. and J.-H.P. Writing—original draft preparation: S.-A.L. Writing—review and editing: S.-A.L. and J.-H.P. Supervision: J.-H.P. All authors have read and agreed to the published version of the manuscript.

Data availability

The datasets generated and/or analyzed during the current study, including the extracted study characteristics and outcome data used for meta-analyses, are available from the corresponding author upon reasonable request.

Code availability

No custom code was used to generate or analyze the results presented in this study.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at 10.1038/s41746-025-02120-0.