Forest therapy for psychological stress and emotional disorders: a systematic review and meta-analysis

Abstract

Background

Forest therapy has attracted growing research interest as a nature-based intervention for mental health. However, the true magnitude and consistency of its benefits remain uncertain, owing to heterogeneous findings across individual studies and notable methodological variability.

Objective

This systematic review and meta-analysis sought to quantitatively synthesize evidence on the effects of forest therapy on psychological stress and emotional disorders (operationally defined here as depressive and anxiety symptoms), and to explore potential moderating factors.

Methods

We searched PubMed, Web of Science, Cochrane Library, CNKI, and Wanfang databases through December 2024. Randomized controlled trials and quasi-experimental studies that examined forest therapy interventions using validated self-report psychological outcome measures were eligible. Random-effects models pooled standardized mean differences (SMDs). Heterogeneity was assessed with Q statistics and I² values; subgroup analyses and meta-regression explored candidate moderators. All analyses were conducted in R (version 4.3.2) using the metafor package (Viechtbauer W. J Stat Softw. 2010;36(3):1-48).

Results

Twenty-five studies involving 1,876 participants met the inclusion criteria. Forest therapy was associated with reductions in psychological stress (SMD = − 0.71, 95% CI: −0.89 to − 0.53), depressive symptoms (SMD = − 0.68, 95% CI: −0.86 to − 0.50), and anxiety symptoms (SMD = − 0.77, 95% CI: −0.97 to − 0.57). Substantial heterogeneity was present across all outcomes (I² > 74%), and the evidence quality was rated moderate for stress and low for emotional disorder outcomes under the GRADE framework. Among the stress outcomes, subgroup analyses indicated that extended intervention periods and clinical populations were linked to larger effect sizes. Publication bias assessment showed modest asymmetry for emotional disorder outcomes; trim-and-fill adjusted estimates remained statistically significant but were attenuated.

Conclusions

Forest therapy appears to be associated with medium-to-large short-term reductions in psychological stress, depressive symptoms, and anxiety symptoms, although high between-study heterogeneity and methodological limitations—particularly the reliance on self-report measures and the impossibility of participant blinding—call for cautious interpretation. The findings tentatively support incorporating forest-based interventions into mental health promotion strategies, especially through sustained, multi-session programs. Longer-term follow-up and more rigorous trial designs are needed before firm clinical recommendations can be made.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12889-026-27255-x.

Introduction

The escalating burden of mental health disorders in contemporary society has prompted researchers and clinicians to explore complementary therapeutic approaches beyond conventional pharmacological and psychological interventions. Forest therapy, often termed “Shinrin-yoku” in Japanese literature, represents one such nature-based intervention that has garnered substantial scientific attention over the past two decades [1]. This therapeutic modality involves structured exposure to forest environments, where participants engage in sensory experiences designed to promote physiological relaxation and psychological restoration. Unlike passive nature exposure, forest therapy incorporates guided activities that deliberately harness the restorative properties of woodland ecosystems.

Psychological stress and emotional disorders have emerged as critical public health concerns worldwide. Recent epidemiological surveys indicate that anxiety disorders affect approximately 301 million individuals globally, while depressive disorders impact an estimated 280 million people [2]. These conditions impose tremendous socioeconomic costs through reduced productivity, healthcare expenditure, and diminished quality of life. What makes this situation particularly alarming is the observed increase in prevalence following the COVID-19 pandemic, which exacerbated pre-existing mental health vulnerabilities across populations [3]. Urban residents face disproportionately elevated risks, partly attributable to chronic exposure to environmental stressors including noise pollution, crowding, and disconnection from natural settings.

Scientific inquiry into forest therapy has yielded promising yet heterogeneous findings. Japanese researchers pioneered experimental investigations demonstrating that forest bathing significantly reduces cortisol concentrations, blood pressure, and sympathetic nervous system activity compared to urban control conditions [4]. European studies subsequently expanded this evidence base, documenting improvements in mood states, self-reported stress levels, and psychological well-being among participants receiving forest-based interventions [5]. Meanwhile, investigations conducted in China and South Korea have examined the neurophysiological mechanisms underlying these benefits, revealing alterations in prefrontal cortex activity and autonomic nervous system regulation [6]. Such cross-cultural research collectively suggests that forest environments possess inherent qualities that facilitate stress recovery and emotional regulation.

Despite this growing body of evidence, several methodological challenges complicate the synthesis and interpretation of existing studies. Sample sizes in individual trials frequently remain modest, limiting statistical power to detect meaningful intervention effects. Considerable heterogeneity exists across studies regarding intervention protocols, outcome measures, participant characteristics, and control conditions [7]. Some investigations employ single-session forest exposures lasting merely two hours, whereas others implement week-long programs with multiple daily sessions. This variability renders direct comparisons problematic and obscures our understanding of optimal intervention parameters. Furthermore, effect size estimates vary substantially across studies, raising questions about the true magnitude of therapeutic benefits attributable to forest therapy [8].

The necessity of conducting a systematic meta-analysis to address these limitations becomes readily apparent. Meta-analytic methodology offers a rigorous framework for quantitatively synthesizing effect sizes across independent studies, thereby increasing statistical precision and enabling examination of potential moderating variables [9]. Through systematic pooling of available evidence, researchers can derive more reliable estimates of intervention efficacy while identifying sources of between-study heterogeneity. Such an approach proves particularly valuable when individual studies yield inconsistent or conflicting results, as observed in the forest therapy literature.

Understanding the therapeutic potential of forest environments requires grounding in established environmental psychology theories. Three complementary frameworks provide the conceptual scaffolding for interpreting how natural settings influence human psychological functioning: Attention Restoration Theory, Stress Recovery Theory, and the Biophilia Hypothesis.

Attention Restoration Theory, advanced by Kaplan and Kaplan, posits that natural environments possess qualities that facilitate recovery from directed attention fatigue [10]. Urban living demands sustained voluntary attention, which depletes cognitive resources and precipitates mental exhaustion. Forest settings, by contrast, engage involuntary attention through “soft fascination”—gentle stimuli that capture interest without requiring effortful concentration. This distinction proves crucial for understanding why brief forest exposures can restore depleted attentional capacity. Stress Recovery Theory, proposed by Ulrich, emphasizes the affective and physiological dimensions of nature exposure [11]. According to this perspective, evolutionarily ingrained responses to natural stimuli trigger rapid parasympathetic activation and positive emotional shifts, thereby counteracting stress-induced arousal.

Historically preceding both ART and SRT, the Biophilia Hypothesis—articulated by Wilson in 1984—proposed that humans harbor an innate affiliation with living systems, a predisposition forged through millennia of evolutionary history [12]. ART and SRT can be viewed as providing more specific mechanistic accounts that build upon this broader evolutionary premise. Together, the three frameworks suggest that forest environments satisfy fundamental psychological needs that built surroundings often leave unmet.

Beyond these overarching theories, specific environmental elements within forests exert measurable physiological effects. Phytoncides—volatile organic compounds released by trees—have demonstrated immunomodulatory and anxiolytic properties when inhaled [13]. Negative air ions, abundant in forested areas particularly near water features, appear to influence serotonin metabolism and mood regulation [14]. Natural soundscapes characterized by birdsong and rustling leaves activate parasympathetic pathways more effectively than urban acoustic environments [15]. Green visual stimuli reduce cortisol secretion and sympathetic nervous activity through mechanisms involving the hypothalamic-pituitary-adrenal axis [16].

Integrating these theoretical perspectives and empirical observations, we conceptualize forest therapy as operating through multiple convergent pathways. Sensory inputs from the forest environment simultaneously engage cognitive restoration processes, trigger phylogenetically ancient stress-dampening responses, and satisfy biophilic needs—collectively modulating neuroendocrine function to ameliorate psychological stress and emotional disturbance [17].

Before we outline the specific aims of the present work, it is important to situate it within the existing body of quantitative syntheses. Yeon et al. [18] conducted a systematic review and meta-analysis focused on depression and anxiety, drawing on 20 studies published through December 2020, and reported large pooled effects favoring forest therapy. Zhang et al. [19] carried out a systematic review concentrating on stress reduction through forest therapy programs, though without performing a quantitative meta-analytic synthesis for stress outcomes. Both of these contributions advanced the field, yet several gaps remain. First, neither synthesis covered the full range of outcomes we address here—perceived stress, depressive symptoms, and anxiety symptoms—within a single analytic framework. Second, neither incorporated literature published after 2020, a period during which the evidence base expanded considerably. Third, moderator analyses in prior reviews were limited, leaving open questions about what intervention and population characteristics shape the magnitude of effects. Our review extends this earlier work by searching through December 2024, spanning both English- and Chinese-language databases, and conducting structured subgroup analyses alongside meta-regression to systematically probe sources of heterogeneity.

Throughout this paper, we use the term “emotional disorders” as an umbrella label referring specifically to depressive symptoms and anxiety symptoms as measured by validated self-report instruments. This operational definition does not encompass the full clinical diagnostic spectrum; rather, it serves as shorthand for the two symptom domains most frequently assessed in the forest therapy literature. We chose to focus on self-reported psychological stress rather than physiological stress markers (e.g., cortisol, heart rate variability) because a dedicated systematic review of physiological indicators already exists [19], and because combining self-report and physiological measures in a single meta-analysis raises interpretive difficulties given their different metric properties.

This investigation aims to quantitatively integrate the available evidence on the effects of forest therapy on psychological stress and emotional disorders. We seek to estimate pooled effect sizes for perceived stress, anxiety symptoms, and depressive symptoms. Additionally, we explore potential moderator variables—including participant health status, intervention duration and frequency, and geographic region—that may influence intervention effectiveness.

Several features distinguish the present work from prior syntheses. We implemented broad literature inclusion criteria spanning multiple databases and languages to reduce the selection bias inherent in English-only searches. Our analytical framework incorporates both subgroup analyses and meta-regression, examining intervention characteristics and population subtypes that received insufficient attention in previous syntheses. Rigorous publication bias assessment employing complementary techniques—funnel plot inspection, Egger’s regression test, and trim-and-fill analysis—strengthens our ability to gauge the robustness of pooled estimates [20]. Through these approaches, we aim to provide updated and more precisely bounded quantitative evidence regarding forest therapy’s short-term effects while identifying critical gaps that warrant future investigation.

Research methods

Statistical approach

We adopted standard meta-analytic methods as described by Borenstein et al. [9]. Effect sizes were expressed as Hedges’ g, which corrects the standardized mean difference for small-sample bias [21]. Given the anticipated clinical and methodological diversity across included studies, we used the DerSimonian-Laird random-effects model to pool effect sizes, thereby incorporating both within-study sampling error and between-study variance (τ²) into the weighting scheme [22].

Heterogeneity was evaluated using Cochran’s Q statistic and the I² index, where values of 25%, 50%, and 75% are conventionally interpreted as low, moderate, and high heterogeneity, respectively [23, 24]. When substantial heterogeneity was detected, we performed subgroup analyses on categorical moderators and random-effects meta-regression on continuous moderators to explore potential sources of effect-size variability [25, 26]. The choice among alternative approaches for explaining heterogeneity followed recommendations by Thompson and Sharp [27], and the meta-regression specification drew on procedures described by Harbord and Higgins [28]. All analyses were conducted in R (version 4.3.2) with the metafor package [29]; detailed mathematical formulations for each statistical procedure are provided in Supplementary Material S1.

Publication bias detection

Publication bias—the preferential publication of statistically significant findings—poses a well-recognized threat to meta-analytic validity [30]. We employed several complementary methods to assess this risk. First, we visually inspected funnel plots for asymmetry, where a deficit of small studies with null findings in the lower portion would suggest selective reporting [31]. Second, Egger’s regression test quantified any systematic relationship between study precision and effect magnitude. Third, Begg’s rank correlation test provided a nonparametric check using Kendall’s tau [32]. When evidence of asymmetry emerged, we applied the Duval and Tweedie trim-and-fill algorithm, which estimates the number of hypothetically missing studies and recalculates the pooled effect after imputing their values [33]. Full mathematical specifications for each of these procedures appear in Supplementary Material S1.

Literature search and data extraction

Literature search strategy

A comprehensive and systematic literature search forms the cornerstone of any rigorous meta-analysis. We conducted searches across multiple electronic databases to capture the breadth of available evidence on forest therapy interventions for psychological outcomes. English-language databases included PubMed, Web of Science Core Collection, and the Cochrane Central Register of Controlled Trials. To ensure adequate representation of Asian research—where much forest therapy investigation has originated—we also searched Chinese-language databases including China National Knowledge Infrastructure (CNKI) and Wanfang Data [34].

The search strategy combined Medical Subject Headings (MeSH) terms with free-text keywords to maximize retrieval sensitivity. For the intervention domain, we employed terms such as “forest therapy,” “forest bathing,” “Shinrin-yoku,” “forest environment,” and “woodland exposure.” Outcome-related terms encompassed “psychological stress,” “perceived stress,” “anxiety,” “depression,” “mood disorder,” and “emotional disturbance.” Boolean operators connected these concept clusters appropriately. Table 1 summarizes the complete search strategy across databases.

Table 1.

Summary of Literature Search Strategy

Database	Language	Search Fields	Intervention Terms	Outcome Terms
PubMed	English	Title/Abstract/MeSH	Forest therapy OR forest bathing OR Shinrin-yoku	Stress OR anxiety OR depression OR mood
Web of Science	English	Topic	Forest therap OR forest bath OR forest bath OR woodland exposure	Psychological stress OR anxiety disorder OR depress*
Cochrane Library	English	Title/Abstract/Keywords	Forest environment OR nature-based intervention	Mental health OR emotional disorder OR stress
CNKI	Chinese	Subject/Keyword	Forest rehabilitation OR forest bathing	Psychological stress OR anxiety OR depression
Wanfang	Chinese	Title/Keyword/Abstract	Forest therapy OR forest康养	Stress response OR emotional disorder

The temporal scope extended from database inception through December 2024, imposing no restrictions on publication date to capture the full historical development of this research field [35].

Inclusion criteria specified: (a) randomized controlled trials or quasi-experimental designs with comparison groups; (b) interventions involving direct exposure to forest environments; (c) outcome measures including validated self-report instruments assessing psychological stress, anxiety symptoms, or depressive symptoms; and (d) sufficient statistical data to compute effect sizes. We excluded narrative reviews, case reports, conference abstracts lacking full methodology, studies with incomplete outcome data, and duplicate publications identified through cross-referencing [36]. When multiple reports derived from identical datasets, we retained only the most comprehensive publication to avoid dependency violations in pooled analyses.

We included both randomized and quasi-experimental designs because the number of randomized trials alone was insufficient to support robust subgroup analyses across the moderator categories we wished to examine. To verify that study design did not drive the observed effects, we conducted a sensitivity analysis restricting the pooled estimate to randomized trials only and compared the result with the full sample (see Sect.  4.3).

This review was not prospectively registered in PROSPERO or any other protocol repository—an oversight we acknowledge as a limitation. We note, however, that the primary outcomes (perceived stress, depressive symptoms, anxiety symptoms), the random-effects pooling model, and the three planned subgroup moderators for stress outcomes (intervention duration, participant health status, geographic region) were specified before data extraction commenced. Additional moderator and meta-regression analyses reported below were exploratory and are identified as such in the relevant results sections.

For stress outcomes, we adopted categorical subgroup analysis because the candidate moderators (intervention duration, participant status, geographic region) were naturally categorical and supported by sufficient study counts per subgroup. For emotional disorder outcomes, we used meta-regression with continuous predictors (intervention frequency, session duration, total intervention period) because the variability in these parameters was better captured on a continuous scale, and because the available data did not afford adequate subgroup cell sizes for categorical splits. We recognize that this difference in analytic approach limits direct comparability between the two outcome domains; this point is addressed further in the Discussion.

When a single study reported both depressive and anxiety outcomes, we computed a combined emotional disorder effect size by averaging the two standardized mean differences while adjusting for their correlation using the approach described by Borenstein et al. [9]. This procedure avoids counting the same participants twice while preserving the information from both outcome domains.

Literature screening process

The screening procedure adhered to Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines to ensure methodological transparency and reproducibility [37]. Two investigators independently conducted all screening stages, with disagreements resolved through discussion or consultation with a third reviewer when consensus proved elusive.

Initial database searches yielded 1,847 potentially relevant records. After removing 412 duplicates through reference management software, 1,435 unique citations remained for preliminary evaluation. During the initial screening phase, reviewers examined titles and abstracts to identify obviously irrelevant records—those addressing unrelated interventions, lacking human participants, or presenting non-empirical content. This process excluded 1,198 articles, leaving 237 publications for full-text assessment.

The subsequent phase involved retrieving and carefully evaluating complete manuscripts against predetermined eligibility criteria. Reasons for exclusion at this stage included: non-experimental study designs (n = 78), absence of forest-specific intervention components (n = 41), lack of relevant psychological outcome measures (n = 35), insufficient statistical information for effect size computation (n = 29), duplicate cohort reports (n = 18), and conference abstracts without full methodology (n = 11). As Fig. 1 illustrates, this rigorous process ultimately identified 25 studies meeting all inclusion criteria.

Fig. 1.

PRISMA Flow Diagram of Literature Screening Process

Inter-rater agreement during screening was substantial, with Cohen’s kappa coefficients of 0.84 for title/abstract review and 0.91 for full-text evaluation [38]. The 25 included studies, published between 2010 and 2024, encompassed 1,876 total participants across diverse geographic regions and population characteristics. Table 2 presents the fundamental characteristics of each included investigation, detailing authorship, publication year, country of origin, sample composition, intervention parameters, control conditions, outcome instruments, and methodological quality ratings.

Table 2.

Characteristics of Included Studies

First Author	Year	Country (a)	Sample Size	Population	Intervention Duration	Intervention Format	Control Condition	Control Type (b)	Outcome Measures
Morita	2011	Japan	71	Healthy adults	2 days	Group-guided	Urban walking	Active-urban	POMS, STAI
Park	2010	Japan	280	University students	15 min	Group-guided	City environment	Active-urban	POMS, cortisol
Lee	2011	Japan	48	Middle-aged males	3 days	Group-guided	Office setting	Passive-indoor	PSS, anxiety VAS
Song	2015	Japan	63	Young adults	15 min	Group-guided	Urban viewing	Active-urban	STAI, HRV
Li	2016	China	94	Elderly hypertensive	7 days	Group-guided	Indoor rest	Passive-indoor	SAS, SDS
Ochiai	2015	Japan	17	Middle-aged females	5 h	Group-guided	Urban environment	Active-urban	POMS, cortisol
Tsunetsugu	2013	Japan	46	Male workers	4 h	Group-guided	City walking	Active-urban	POMS, blood pressure
Shin	2012	South Korea	56	Depressed patients	4 weeks	Group-guided	Hospital setting	Treatment as usual	BDI, STAI
Kim	2015	South Korea	62	Office workers	2 days	Group-guided	Workplace	Passive-indoor	PSS, POMS
Yu	2017	China	60	College students	2 h	Group-guided	Classroom	Passive-indoor	SAS, SDS
Mao	2017	China	24	Chronic stroke patients	8 weeks	Group-guided	Conventional rehab	Treatment as usual	HADS, cortisol
Hassan	2018	Malaysia	54	Urban residents	3 h	Group-guided	Indoor rest	Passive-indoor	DASS-21
Bielinis	2019	Poland	32	Young adults	15 min	Group-guided	Urban walking	Active-urban	POMS, restorativeness
Kobayashi	2015	Japan	585	University students	15 min	Group-guided	City viewing	Active-urban	STAI, HRV
Takayama	2014	Japan	44	Middle-aged adults	4 h	Group-guided	Urban sitting	Active-urban	POMS, cortisol
Chen	2018	Taiwan	72	Healthcare workers	2 days	Group-guided	Normal routine	Treatment as usual	PSS, STAI
Guan	2019	China	80	Anxious patients	4 weeks	Group-guided	Medication only	Treatment as usual	HAMA, PSS
Dolling	2017	Sweden	46	Burnout patients	12 weeks	Group-guided	Indoor therapy	Active-non-nature	PSS, HADS
Chun	2017	South Korea	43	Elderly females	12 weeks	Group-guided	Social activities	Active-non-nature	GDS, STAI
Wang	2019	China	96	University students	1 week	Group-guided	Campus walking	Active-non-nature	SAS, SDS, PSS
Rajoo	2020	Malaysia	59	Depressed patients	6 weeks	Group-guided	Treatment as usual	Treatment as usual	BDI, BAI
Antonelli	2019	Italy	38	Healthy adults	2 h	Group-guided	Urban park	Active-non-nature	POMS, PSS
Zhang	2021	China	108	Nurses	3 days	Group-guided	Normal routine	Treatment as usual	PSS, SAS, SDS
Furuyashiki	2019	Japan	44	Stressed workers	2 h	Self-guided	Office break	Passive-indoor	PSS, POMS
Koselka	2019	USA	59	Veterans with PTSD	6 weeks	Group-guided	Waitlist	Waitlist	PCL-5, PHQ-9

(a) Country refers to the geographic location where participants were recruited, not the researchers’ institutional affiliation. (b) Control types were classified as: Active-urban (urban walking, city viewing, or urban environment exposure), Passive-indoor (indoor rest, office, classroom settings), Treatment as usual (normal routine, conventional treatment, medication only), Active-non-nature (indoor therapy, social activities, campus walking, urban park), and Waitlist (no active intervention). POMS Profile of Mood States, STAI State-Trait Anxiety Inventory, PSS Perceived Stress Scale, SAS Self-rating Anxiety Scale, SDS Self-rating Depression Scale, BDI Beck Depression Inventory, HADS Hospital Anxiety and Depression Scale, DASS-21 Depression Anxiety Stress Scales, VAS Visual Analog Scale, HRV Heart Rate Variability, HAMA Hamilton Anxiety Rating Scale, GDS Geriatric Depression Scale, BAI Beck Anxiety Inventory, PCL-5 PTSD Checklist, PHQ-9 Patient Health Questionnaire-9

Data extraction and quality assessment

Two reviewers independently extracted relevant information from each eligible study using a standardized data collection form. Extracted variables encompassed bibliographic details (first author, publication year, country), methodological characteristics (study design, randomization method), participant information (sample size, age range, sex distribution, health status), intervention parameters (forest type, exposure duration, activity components, session frequency), control conditions (urban environment, indoor setting, usual care), and outcome data (measurement instruments, pre-post means, standard deviations) [39]. When studies reported multiple time points, we extracted immediate post-intervention assessments to maintain comparability across investigations.

Methodological quality evaluation followed the Cochrane Collaboration’s Risk of Bias tool, which appraises six domains: random sequence generation, allocation concealment, blinding of participants and personnel, incomplete outcome data, selective outcome reporting, and other potential biases [40]. Each domain received a judgment of “low risk,” “unclear risk,” or “high risk” based on explicit criteria. Given the nature of forest therapy interventions, participant blinding proves inherently impossible—a limitation acknowledged across behavioral intervention research [41]. However, outcome assessor blinding and allocation concealment remain achievable and were evaluated accordingly.

Table 3 presents domain-specific quality ratings for all included studies. Based on the ratings shown in Tables 3, 15 studies were judged at low risk for random sequence generation, while 8 provided adequate allocation concealment (rated low risk). All 25 studies received a high-risk rating for blinding, which was expected given that participant blinding is inherently infeasible in forest therapy trials [41]. Twenty-two studies were rated low risk for incomplete outcome data, suggesting that attrition bias was generally well-controlled. Selective reporting was rated low risk in 23 studies, and other sources of bias were rated low risk in 23 studies [42].

Table 3.

Risk of Bias Assessment Results for Included Studies

First Author	Random Sequence	Allocation Concealment	Blinding	Incomplete Data	Selective Reporting	Other Bias
Morita	Low	Unclear	High	Low	Low	Low
Park	Low	Low	High	Low	Low	Low
Lee	Unclear	Unclear	High	Low	Low	Low
Song	Low	Low	High	Low	Low	Low
Li	Low	Unclear	High	Low	Unclear	Low
Ochiai	Unclear	Unclear	High	Low	Low	Low
Tsunetsugu	Low	Unclear	High	Low	Low	Low
Shin	Low	Low	High	Low	Low	Low
Kim	Unclear	Unclear	High	Unclear	Low	Low
Yu	Unclear	Unclear	High	Low	Low	Unclear
Mao	Low	Low	High	Low	Low	Low
Hassan	Unclear	Unclear	High	Low	Low	Low
Bielinis	Low	Unclear	High	Low	Low	Low
Kobayashi	Low	Low	High	Low	Low	Low
Takayama	Unclear	Unclear	High	Low	Low	Low
Chen	Low	Unclear	High	Low	Low	Low
Guan	Low	Low	High	Low	Low	Unclear
Dolling	Low	Low	High	Unclear	Low	Low
Chun	Unclear	Unclear	High	Low	Low	Low
Wang	Unclear	Unclear	High	Low	Low	Low
Rajoo	Low	Unclear	High	Low	Low	Low
Antonelli	Unclear	Unclear	High	Low	Low	Low
Zhang	Low	Unclear	High	Low	Low	Low
Furuyashiki	Unclear	Unclear	High	Low	Unclear	Low
Koselka	Low	Low	High	Unclear	Low	Low

The obtained kappa value of 0.87 indicated excellent inter-rater agreement across all quality domains [43]. Figure 2 depicts the overall risk of bias distribution, revealing that blinding limitations represent the most prevalent concern—an unavoidable constraint in behavioral intervention research.

Fig. 2.

Risk of Bias Summary Across Included Studies

Meta-analysis results

Intervention effects of forest therapy on psychological stress

Eighteen of the 25 included studies reported outcomes related to psychological stress, encompassing 1,342 participants across intervention and control conditions. Consistent with our stated focus on self-reported psychological outcomes, we extracted data from validated questionnaires—primarily the Perceived Stress Scale (PSS), the stress subscale of the Depression Anxiety Stress Scales (DASS-21), and comparable instruments [44]. When studies reported both self-report and physiological stress indicators (e.g., cortisol, heart rate variability), only the self-report data were included in the pooled analysis to maintain measurement consistency across the synthesis.

We pooled effect sizes using a random-effects model given anticipated heterogeneity arising from methodological and population differences. The standardized mean difference served as the common effect size metric, with negative values indicating stress reduction favoring the forest therapy condition. The pooled effect estimate revealed a statistically significant reduction in psychological stress following forest-based interventions (SMD = -0.71, 95% CI: -0.89 to -0.53, p < 0.001). According to conventional benchmarks proposed by Cohen, this magnitude represents a medium-to-large effect [45].

Figure 3 presents the forest plot displaying individual study effect sizes alongside their 95% confidence intervals. The visual representation confirms that most investigations yielded effects favoring forest therapy, though considerable variation exists in effect magnitude. Several Japanese studies demonstrated particularly robust effects, with standardized differences exceeding − 1.0 [46].

Fig. 3.

Forest Plot of Forest Therapy Effects on Psychological Stress

Heterogeneity assessment revealed substantial between-study variability. The Q statistic reached 67.42 (df = 17, p < 0.001), rejecting the homogeneity assumption. The I² value of 74.8% indicated that approximately three-quarters of observed variance reflected true heterogeneity rather than sampling error [47]. The resulting τ² estimate of 0.18 confirmed meaningful effect size dispersion warranting exploration through moderator analyses.

Subgroup analyses examined potential sources of heterogeneity across three pre-specified categorical variables: intervention duration, participant health status, and geographic region. Results indicated significant moderation by duration (Q_b = 9.84, df = 2, p = 0.007), with extended programs yielding larger effects (SMD = − 0.93) compared to brief exposures (SMD = − 0.54). Table 4 summarizes these subgroup findings comprehensively.

Table 4.

Subgroup Analysis Results for Psychological Stress Outcomes

Moderator Variable	Subgroup Category	k	SMD (95% CI)	(%)	(p-value)
Intervention Duration(a)	Brief (≤ 1 day)	7	-0.54 (-0.78, -0.30)	62.3	9.84 (0.007)
	Short-term (2–7 days)	6	-0.72 (-0.98, -0.46)	58.7
	Extended (> 1 week)	5	-0.93 (-1.24, -0.62)	71.4
Participant Status(a)	Healthy individuals	11	-0.58 (-0.77, -0.39)	65.2	6.71 (0.010)
	Clinical populations	7	-0.91 (-1.18, -0.64)	69.8
Geographic Region(a)	East Asia	14	-0.75 (-0.95, -0.55)	76.3	1.23 (0.267)
	Western countries	4	-0.59 (-0.92, -0.26)	68.4
Control Condition Type(b)	Active-urban	8	−0.62 (− 0.84, − 0.40)	60.1	4.53 (0.104)
	Passive-indoor / TAU	7	−0.81 (− 1.08, − 0.54)	72.6
	Active-non-nature / Waitlist	3	−0.78 (− 1.19, − 0.37)	69.3

(a) Pre-specified subgroup analyses. (b) Exploratory subgroup analysis. TAU treatment as usual. The active-non-nature / waitlist subgroup contains only 3 studies and should be interpreted with particular caution

As Table 4 shows, participant health status emerged as a significant moderator for stress outcomes (Q_b = 6.71, p = 0.010). Clinical populations—individuals with diagnosed stress-related conditions, burnout, or elevated baseline distress—showed larger reductions (SMD = − 0.91) than healthy participants (SMD = − 0.58) [48]. We note that this finding, while consistent with the plausible expectation that more distressed individuals have greater room for improvement, derives solely from the stress outcome domain and should not be extrapolated to emotional disorder outcomes without further testing.

Geographic region did not significantly moderate effects (Q_b = 1.23, p = 0.267), though the Western-countries subgroup contained only four studies, limiting the power of this comparison [49]. An exploratory subgroup analysis by control condition type (Table 4, bottom rows) did not reach statistical significance (Q_b = 4.53, p = 0.104), yet a numerical pattern was apparent: studies using active-urban controls yielded somewhat smaller effects (SMD = − 0.62) than those using passive-indoor or treatment-as-usual controls (SMD = − 0.81). This trend—if replicated—would suggest that part of the observed benefit may reflect the relative disadvantage of the comparator environment rather than a forest-specific therapeutic mechanism. However, the active-non-nature/waitlist subgroup included only three studies, and we flag this analysis as exploratory and underpowered.

Figure 4 depicts the subgroup analysis forest plot stratified by intervention duration, visually illustrating the dose-response relationship between exposure length and stress reduction magnitude.

Fig. 4.

Subgroup Analysis Forest Plot by Intervention Duration

An exploratory meta-regression incorporating mean participant age and baseline stress severity as continuous predictors accounted for roughly 23% of between-study variance (detailed formulas are presented in Supplementary Material S1). Baseline stress severity reached statistical significance (β = 0.034, p = 0.018), tentatively supporting the observation that participants with higher initial distress may derive greater short-term benefit [50]. That said, this model left the majority of heterogeneity unexplained, and the modest proportion of variance accounted for cautions against overinterpreting these results.

Publication bias assessment employed multiple complementary approaches. Visual inspection of the funnel plot, displayed in Fig. 5, revealed mild asymmetry with a slight deficit of small studies reporting negligible effects. Egger’s regression test approached but did not reach statistical significance (intercept = -1.42, p = 0.078) [51].

Fig. 5.

Funnel Plot for Psychological Stress Outcomes

Begg’s rank correlation test similarly yielded a non-significant result (Kendall’s τ = −0.21, p = 0.143). Nonetheless, we applied the trim-and-fill procedure as a sensitivity measure, which imputed three potentially missing studies. The adjusted pooled estimate (SMD = − 0.64, 95% CI: −0.82 to − 0.46) remained statistically significant and clinically meaningful, suggesting that publication bias, if present, does not fundamentally alter our conclusions.

Intervention effects of forest therapy on emotional disorders

Twenty-one studies assessed emotional disorder outcomes—that is, depressive symptoms, anxiety symptoms, or both—using instruments such as the SDS, BDI, HADS, SAS, STAI, and POMS negative affect subscales [52]. We first conducted separate meta-analyses for depression and anxiety. For the combined emotional disorder analysis shown in Fig. 6, when a single study reported both depressive and anxiety effect sizes, we computed a composite by averaging the two standardized mean differences and adjusting their variance for an assumed inter-outcome correlation of r = 0.50, following the approach outlined in Borenstein et al. [9]. We chose this correlated-outcomes method to avoid double-counting participants while retaining information from both symptom domains. A sensitivity analysis using r = 0.30 and r = 0.70 produced substantively identical pooled estimates (results available in Supplementary Material S2).

Fig. 6.

Forest Plot of Forest Therapy Effects on Emotional Disorders

For depressive symptoms, 16 studies involving 1,089 participants contributed data. The random-effects pooled estimate indicated a significant reduction favoring forest therapy interventions (SMD = − 0.68, 95% CI: −0.86 to − 0.50, p < 0.001). This effect magnitude falls within the medium range according to conventional interpretation guidelines [53]. Anxiety symptom outcomes, reported across 19 studies with 1,247 participants, yielded a somewhat larger pooled effect (SMD = − 0.77, 95% CI: −0.97 to − 0.57, p < 0.001). The difference between depression and anxiety effect sizes did not reach statistical significance when formally compared (Q_diff = 0.89, p = 0.346), though the numerical trend favoring anxiety reduction warrants further investigation [54].

For moderator analyses of emotional disorder outcomes, we used meta-regression rather than categorical subgroup analysis—the approach adopted for stress outcomes—for two reasons. First, the key intervention parameters of interest (frequency, session duration, total period) varied on a continuous scale and were more efficiently modeled as such. Second, the distribution of studies across categorical splits (e.g., by participant status or geographic region) was too uneven to yield reliable between-subgroup comparisons for emotional disorder outcomes specifically. We acknowledge that this difference in analytic strategy limits the direct comparability of moderator findings between the stress and emotional disorder domains.

Figure 6 displays the forest plot for combined emotional disorder outcomes.

Heterogeneity testing confirmed substantial variability across investigations. The omnibus Q statistic of 89.67 (df = 20, p < 0.001) definitively rejected homogeneity. The I² estimate of 77.7% indicated that over three-quarters of total variance stemmed from true between-study differences rather than chance fluctuation. The prediction interval of − 1.38 to − 0.02 remained entirely below zero, suggesting that beneficial effects would be expected even in studies with substantially different characteristics [55].

We conducted an exploratory meta-regression to examine whether continuous intervention characteristics—frequency (sessions per week), single session duration (hours), and total intervention period (days)—helped explain the observed heterogeneity in emotional disorder outcomes. All three predictors were entered simultaneously into a random-effects meta-regression model.

Table 5 presents the results. Total intervention period was the only predictor that reached significance (β = −0.024, p = 0.003), suggesting that each additional day was associated with a 0.024 standard deviation larger effect. Figure 7 illustrates this relationship, displaying the scatter plot of individual study effect sizes against total intervention period alongside the fitted meta-regression line.

Table 5.

Meta-Regression Analysis Results for Emotional Disorder Outcomes

Predictor Variable	Coefficient	Standard Error	z-value	p-value
Intercept	-0.312	0.187	-1.67	0.095
Intervention frequency (sessions/week)	-0.076	0.058	-1.31	0.189
Session duration (hours)	-0.043	0.031	-1.39	0.165
Total intervention period (days)	-0.024	0.008	-3.00	0.003
Model fit: = 14.72, df = 3, p = 0.002
Residual heterogeneity: = 68.4%
Variance explained: = 31.2%

Fig. 7.

Meta-Regression Scatter Plot: Effect Size by Total Intervention Period

However, it is worth noting that the overall model, while statistically significant (Q_M = 14.72, p = 0.002), accounted for approximately 31% of between-study variance, leaving residual heterogeneity at I²_res = 68.4%. This means that nearly two-thirds of the variability across studies remained unexplained, and unmeasured factors—such as participant baseline severity, specific forest features, accompanying therapeutic activities, and measurement sensitivity—likely contributed meaningfully [56]. The meta-regression findings should therefore be understood as partial and tentative explanations for the observed variability, not definitive conclusions about what drives intervention effectiveness.

Publication bias evaluation employed both visual and statistical approaches. The funnel plot depicted in Fig. 8 shows the relationship between study effect sizes and their standard errors. Inspection reveals modest asymmetry, with some absence of smaller studies reporting null or negative findings in the lower-right quadrant.

Fig. 8.

Funnel Plot for Emotional Disorder Outcomes

Egger’s linear regression test indicated a statistically significant intercept (α = −2.18, p = 0.024), pointing to potential publication bias for emotional disorder outcomes [57]. Begg’s rank correlation test corroborated this finding (Kendall’s τ = −0.29, p = 0.038). These results warrant interpretive caution: the observed funnel plot asymmetry might reflect either selective publication of positive findings or genuine methodological differences between smaller and larger studies. To gauge the impact on our conclusions, we applied the trim-and-fill procedure, which imputed five hypothetically missing studies. The bias-adjusted pooled estimate was SMD = − 0.58 (95% CI: −0.76 to − 0.40)—a 15% attenuation from the unadjusted figure, yet still statistically significant [58]. This sensitivity analysis suggests that even if publication bias has modestly inflated the observed effects, forest therapy interventions still appear to produce short-term benefits for emotional disorder symptoms.

Sensitivity analysis and evidence quality assessment

The robustness of pooled effect estimates warrants scrutiny through sensitivity analysis, particularly given the heterogeneity observed across included studies. We employed the leave-one-out approach, systematically removing each study and recalculating the pooled effect to determine whether any single investigation disproportionately influenced overall conclusions [59]. This iterative procedure generates a range of effect estimates, the stability of which reflects confidence in the synthesized findings.

For psychological stress outcomes, the leave-one-out analysis yielded pooled SMDs ranging from − 0.66 to − 0.76, with no individual study’s removal shifting the estimate beyond the original 95% confidence interval. Two Japanese studies contributed disproportionately to heterogeneity, yet their removal did not substantively alter pooled estimates—suggesting that outlying effects reflected genuine variation rather than methodological artifact.

Emotional disorder outcomes demonstrated comparable stability. The pooled SMD fluctuated between − 0.63 and − 0.81 across leave-one-out iterations. Notably, exclusion of the Shin (2012) study—which reported the largest effect in our sample—reduced the estimate to -0.63, still representing a medium effect size with statistical significance intact. This pattern confirms that conclusions do not hinge upon any single investigation.

Evidence quality assessment followed the Grading of Recommendations Assessment, Development, and Evaluation (GRADE) framework, which evaluates five domains: risk of bias, inconsistency, indirectness, imprecision, and publication bias [60]. Each domain receives a rating that may downgrade initial evidence quality from “high” for randomized trials.

Regarding risk of bias, the inherent impossibility of participant blinding in behavioral interventions necessitated a one-level downgrade. Inconsistency concerns arose from substantial heterogeneity (I² > 75%), warranting another downgrade despite partial explanation through moderator analyses. Indirectness posed minimal concern—included studies directly addressed our research question with relevant populations and outcomes. The cumulative sample size of 1,876 participants exceeded the calculated optimal information size threshold, indicating adequate precision [61]. Publication bias, evidenced by Egger’s test significance for emotional outcomes, prompted a one-level downgrade for that outcome domain.

As an additional robustness check, we restricted the pooled analysis to randomized controlled trials only, excluding quasi-experimental designs. For psychological stress, the RCT-only pooled estimate was SMD = − 0.66 (95% CI: −0.87 to − 0.45, k = 13), slightly attenuated from the full-sample estimate of − 0.71 but qualitatively consistent. For emotional disorders, the RCT-only estimate was SMD = − 0.62 (95% CI: −0.82 to − 0.42, k = 15), compared with − 0.68 in the full sample. These results suggest that the inclusion of quasi-experimental studies did not substantively inflate our pooled effects, though we note that the modest attenuation is consistent with the generally higher risk of bias in non-randomized designs.

Synthesizing across all GRADE domains, we rated the overall evidence quality as “moderate” for psychological stress outcomes (downgraded for blinding limitations) and “low” for emotional disorder outcomes (downgraded for blinding limitations, high inconsistency, and evidence of publication bias) [62]. These ratings carry important implications for how our conclusions should be read. For stress outcomes, there is moderate confidence that the true effect lies close to our pooled estimate, though it could differ meaningfully. For emotional disorder outcomes, the low certainty rating means that the true effect may be substantially different from the estimate reported here; future well-powered, blinding-optimized trials may considerably revise the magnitude. An additional consideration cutting across both outcome domains is the near-complete reliance on self-report instruments. Because forest therapy cannot be blinded—participants know they are in a forest—expectancy effects, demand characteristics, and social desirability may inflate self-reported benefits. This concern is not merely hypothetical; it has been documented in other behavioral intervention literatures [41]. Until more studies incorporate observer-rated or physiological outcome measures alongside self-report, the contribution of expectancy bias to the observed effects cannot be ruled out.

Discussion

Drawing on 25 studies and 1,876 participants, this meta-analysis found that forest therapy was associated with medium-to-large short-term reductions in psychological stress (SMD = − 0.71), depressive symptoms (SMD = − 0.68), and anxiety symptoms (SMD = − 0.77). While these pooled estimates reached statistical significance, they should be interpreted alongside the substantial heterogeneity observed across all outcomes (I² > 74%) and the GRADE certainty ratings of moderate (stress) and low (emotional disorders). The comparison with established psychological treatments therefore warrants caution: the observed effect sizes may partly reflect expectancy bias, favorable comparator conditions, or measurement artifacts, rather than genuine therapeutic superiority. We position forest therapy as a potentially useful complementary modality, but refrain from asserting clinically definitive effects on the basis of the current evidence.

How do these results compare with earlier quantitative syntheses? Yeon et al. [18] reported large pooled effects for depression (Hedges’ g = 1.133) and anxiety (Hedges’ g = 1.715) based on 20 studies through 2020; our estimates for the corresponding outcomes are notably smaller in magnitude. Several factors may explain this difference: we included studies published through 2024 with a broader geographic and language scope, we used a different composite procedure for studies reporting multiple outcomes, and our subgroup and meta-regression analyses adjusted for a wider set of moderators. Zhang et al. [19] reviewed stress reduction through forest therapy but did not perform a quantitative meta-analysis for stress outcomes, making direct numerical comparison impossible; our pooled SMD of − 0.71 for stress fills this gap. Across these comparisons, a common thread emerges: forest therapy appears beneficial, yet the precise magnitude of benefit varies appreciably depending on analytic choices, inclusion criteria, and the recency of the literature base.

The mechanisms underlying forest therapy’s efficacy likely operate through multiple convergent pathways. From a neurophysiological perspective, forest environments reduce sympathetic nervous system activation while enhancing parasympathetic tone. Phytoncides and negative air ions interact with olfactory and respiratory systems to modulate hypothalamic-pituitary-adrenal axis activity, dampening cortisol release. These physiological shifts create conditions conducive to psychological restoration—the subjective experience of mental refreshment and renewed attentional capacity that Attention Restoration Theory describes. Beyond individual-level processes, forest therapy programs often incorporate group activities that foster social connection and peer support. This relational dimension may amplify therapeutic benefits, particularly for individuals experiencing social isolation as a component of their distress.

Our moderator analyses offer partial—though far from complete—insight into what shapes intervention effects. For stress outcomes specifically, pre-specified subgroup analyses indicated that extended programs (> 1 week) yielded larger effects than brief single-session exposures, and that clinical populations showed greater reductions than healthy participants. We emphasize that these moderator findings apply to the stress outcome domain and should not be generalized to emotional disorders without corresponding evidence. An exploratory subgroup analysis by control condition type did not reach significance for stress outcomes, but the pattern was suggestive: effects were numerically smaller when forest therapy was compared against active-urban controls than against passive-indoor or treatment-as-usual comparators. If confirmed in future work, this would imply that at least part of the observed benefit reflects the relative disadvantage of the comparator rather than a forest-specific mechanism.

For emotional disorder outcomes, meta-regression identified total intervention period as a statistically significant predictor, with longer programs associated with larger effects. However, this model explained only about 31% of between-study variance, and residual heterogeneity remained high (I²_res = 68.4%). Intervention frequency and session duration showed expected directional trends without reaching significance—possibly a reflection of restricted variability across studies or insufficient statistical power. These results are best read as tentative hints about dose-response patterns rather than firm prescriptions. Practitioners considering program design might lean toward multi-session formats over isolated visits, but direct head-to-head trials comparing different dosing schedules are needed before this guidance can be offered with confidence.

Several limitations temper the conclusions we can draw, and we want to be transparent about their collective weight. This review was not prospectively registered, which means we cannot fully rule out the possibility that some analytical decisions were influenced by emerging results, even though our primary outcomes and main subgroup moderators were defined before data extraction. Our search, while spanning multiple databases and two languages, may have missed relevant grey literature—unpublished dissertations, government reports, and conference proceedings—as well as studies published in Japanese, Korean, or other languages where forest therapy research has flourished.

Perhaps the most consequential limitation is the near-total absence of follow-up data beyond the immediate post-intervention time point. Virtually all included studies measured outcomes right after the intervention ended; we therefore have no basis for claiming that benefits persist over weeks, months, or longer. Although isolated reports have suggested that certain physiological benefits—such as blood pressure reductions among office workers—may be sustained for a period after the program ends [63], no comparable evidence exists for the psychological outcomes examined here. Our conclusions should be understood as applying strictly to short-term, immediate effects. Without evidence of durability, recommending forest therapy as a treatment for chronic mental health conditions remains speculative.

The impossibility of participant blinding in forest-based interventions is another concern that cuts across the entire evidence base. Participants know they are in a forest, and the pleasant sensory context may amplify self-reported improvements through expectancy, novelty, or demand characteristics. This issue is compounded by the heavy reliance on self-report outcome measures: only a minority of included studies incorporated observer-rated or physiological endpoints. Future trials should pair self-report instruments with blinded outcome assessors and physiological measures to disentangle genuine therapeutic change from reporting bias [41].

The heterogeneity of control conditions across studies further complicates interpretation. Some comparators were active (urban walking), others passive (indoor rest), and still others represented usual care or waitlist conditions. Our exploratory subgroup analysis hints that effect sizes may be inflated when forest therapy is compared against less stimulating controls, but this analysis was underpowered. Future research should prioritize active comparators—particularly non-forest outdoor settings—to isolate forest-specific mechanisms from the general benefits of physical activity, social engagement, or being away from routine stressors.

Finally, substantial residual heterogeneity persisted even after our moderator analyses, indicating that important but unmeasured factors—such as specific forest characteristics, the skill of the therapy facilitator, or cultural attitudes toward nature—also shape outcomes. Identifying and systematically coding these factors should be a priority for the next generation of studies.

Future research should pursue several directions. First, rigorously designed dose-response studies manipulating exposure duration, frequency, and intensity would clarify optimal intervention parameters. Second, mechanistic investigations incorporating neuroimaging, biomarker panels, and ecological momentary assessment could elucidate how forest environments produce psychological change. Third, studies targeting special populations—children, older adults, individuals with specific psychiatric diagnoses—would establish population-specific efficacy and safety profiles. Finally, economic evaluations comparing forest therapy costs and outcomes against standard treatments would inform healthcare policy and resource allocation decisions.

Conclusion

This systematic review and meta-analysis synthesized evidence from 25 controlled studies involving 1,876 participants to examine the short-term effects of forest therapy on psychological stress and emotional disorders (depressive and anxiety symptoms).

The pooled estimates indicate that forest therapy is associated with medium-to-large immediate reductions in all three outcome domains (SMD = − 0.71 for stress, − 0.68 for depression, − 0.77 for anxiety). However, these estimates are accompanied by high between-study heterogeneity (I² > 74%), and the certainty of the evidence is moderate for stress and low for emotional disorder outcomes per the GRADE framework. We therefore describe these findings as promising rather than definitive.

Regarding moderating factors, our analyses indicate that longer intervention periods were linked to larger effects for both stress (subgroup analysis) and emotional disorders (meta-regression), suggesting a dose-response pattern. For stress outcomes specifically, clinical populations showed greater improvements than healthy participants—a pattern consistent with the notion that those with higher baseline distress stand to gain more. We caution that these moderator results apply to the specific outcome domains in which they were tested and should not be generalized beyond them. The exploratory nature of several analyses, small subgroup sizes, and the modest proportion of variance explained by meta-regression all constrain the strength of these conclusions.

From a practical standpoint, the available evidence tentatively supports the incorporation of forest therapy into broader mental health promotion strategies, particularly in the form of sustained, multi-session programs rather than isolated exposures. However, we stress that nearly all evidence pertains to immediate post-intervention assessments; whether these short-term benefits endure over weeks or months remains unknown. Additionally, the reliance on self-report measures in the context of an inherently unblinded intervention raises the possibility that expectancy effects contribute to the observed benefits.

Future research should prioritize adequately powered randomized trials with extended follow-up periods, blinded outcome assessment where feasible, standardized self-report instruments supplemented by physiological markers, and active comparator conditions that can help disentangle forest-specific mechanisms from generic effects of outdoor activity or structured leisure. Pre-registration of study protocols and analytic plans would also strengthen the credibility of future contributions to this field. Only through such methodological refinements can we build the evidence base needed to guide confident clinical and policy recommendations about forest therapy.

Abbreviations

SMD

Standardized Mean Difference

CI

Confidence Interval

PRISMA

Preferred Reporting Items for Systematic Reviews and Meta-Analyses

GRADE

Grading of Recommendations Assessment, Development, and Evaluation

MeSH

Medical Subject Headings

CNKI

China National Knowledge Infrastructure

POMS

Profile of Mood States

STAI

State-Trait Anxiety Inventory

PSS

Perceived Stress Scale

SAS

Self-rating Anxiety Scale

SDS

Self-rating Depression Scale

BDI

Beck Depression Inventory

HADS

Hospital Anxiety and Depression Scale

DASS-21

Depression Anxiety Stress Scales

VAS

Visual Analog Scale

HRV

Heart Rate Variability

HAMA

Hamilton Anxiety Rating Scale

GDS

Geriatric Depression Scale

BAI

Beck Anxiety Inventory

PCL-5

PTSD Checklist for DSM-5

PHQ-9

Patient Health Questionnaire-9

OIS

Optimal Information Size

Authors’ contributions

GQ and CP conceptualized and designed the study. GQ and JY developed the search strategy and conducted the literature search. GQ and FL independently screened titles, abstracts, and full texts for eligibility. JY and ZZ performed data extraction and quality assessment. GQ conducted the statistical analyses. GQ drafted the initial manuscript. CP supervised the entire study process. All authors critically reviewed and revised the manuscript for important intellectual content. All authors read and approved the final manuscript.

Funding

No funding was received for this research.

Data availability

All data generated and analyzed during the current study are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. This study is a systematic review and meta-analysis of previously published studies and did not involve the collection of new data from human participants.

Consent for publication

All authors have reviewed the manuscript and consent to its publication. No identifiable information regarding participants has been included.

Competing interests

The authors declare no competing interests.

Clinical Trial Number

Not applicable.