The protocol for this systematic review and meta-analysis was prospectively registered in the PROSPERO International Prospective Register of Systematic Reviews (CRD42024618260). Reporting follows the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) 2020 statement [14].

We included randomized controlled trials (RCTs) published in English between 2003 and 2023 that evaluated the efficacy of BTs, specifically Reiki, Therapeutic Touch (TT), Healing Touch (HT), and External Qigong Treatment (EQT), on pain or on symptoms associated with mental disorders (SAMD), encompassing anxiety, depression, stress, and aggressiveness. Observational studies, non-randomized designs, case series, reviews, and meta-analyses were excluded to minimize data duplication and bias.

Electronic searches were conducted in PubMed (January 2003 to December 2024), Scopus (January 2003 to December 2024), and CINAHL via EBSCOhost (January 2003 to December 2024) using Boolean combinations of intervention terms “Reiki”, “Therapeutic Touch”, “Healing Touch”, “External Qigong”, “Biofield therapy”, “Energy healing”, and “Energy therapy”, crossed with outcome-specific term blocks covering pain, anxiety, depression, stress, aggressiveness, and cancer. In PubMed, additional filters were applied for human participants, English language, and publication type (Randomized Controlled Trial; Clinical Trial); for Scopus and CINAHL, no database-level study design filter was available, and RCT status was verified manually during title/abstract and full-text screening. A complementary manual search in Google Scholar targeted grey literature and studies not indexed in the primary databases. Complete search strings for each database and outcome are provided in Supplementary Material S1. After deduplication, 312 unique records were screened by title and abstract. Thirty-four full-text articles were assessed for eligibility; 28 RCTs met the inclusion criteria and were retained for analysis. The study selection process is illustrated in Figure 1.

Six full-text reports were excluded after eligibility assessment for the following reasons: one study used a non-randomized quasi-experimental design without a formal allocation procedure meeting the RCT inclusion criterion; four studies did not report sufficient quantitative data to permit effect size calculation, as means, standard deviations, or both were absent or incomplete and could not be recovered from supplementary materials or by contacting the corresponding authors; and one study was excluded because participants were informal caregivers of patients with cancer rather than patients receiving biofield therapy directly, not meeting the population inclusion criterion. Detailed references and specific reasons for exclusion of each full-text report are provided in Supplementary Table S2.

Two independent reviewers screened records and extracted data using a prespecified spreadsheet that captured sample characteristics, intervention protocol details (modality, number and duration of sessions, frequency, total weeks), outcome measures, and summary statistics (means, standard deviations, and sample sizes per group). Disagreements were resolved by consensus or adjudication by a third reviewer. When standard deviations were not reported, they were estimated from confidence intervals using the formula:

where z_α/2 = 1.96 for a 95% confidence interval, and SD was subsequently derived as.

The 28 RCTs were allocated to two analytic groups: Pain (k = 14) and SAMD (k = 22). Six studies contributed data to both groups; two SAMD studies provided separate effect sizes for anxiety and depression and were counted twice within the SAMD group.

Two reviewers (RS and PN) independently assessed risk of bias for all 28 included trials using the Cochrane Risk of Bias tool (RoB 2). Disagreements were resolved by consensus or by a third reviewer (RQ).

The search identified 351 records. After deduplication, 312 unique records were screened; 34 full-text articles were retrieved, and 28 RCTs met eligibility criteria (Figure 1). The included trials evaluated four BT modalities: Reiki (n = 10), Healing Touch (n = 8), Therapeutic Touch (n = 7), and External Qigong (n = 3). The total sample comprised N = 2,013 participants, with n = 892 allocated to experimental groups.

Clinical populations encompassed cancer patients (7 trials; N = 447) [17,18,19,20,21,22,23], post-surgical or peri-operative patients (6 trials; N = 611) [24,25,26,27,28,29], patients with chronic pain conditions including fibromyalgia and osteoarthritis (4 trials; N = 281) [30,31,32,33], healthy adults (3 trials; N = 281) [34,35,36], and other specific populations including drug rehabilitation [37], chronic obstructive pulmonary disease [38], neurodegenerative conditions [39,40], hematopoietic stem cell transplantation [41], HIV [42], sickle cell disease [43], and the elderly [44]. Within the SAMD group (22 trials; N = 1,581), nine trials evaluated anxiety (n_exp = 349) [18,21,25,26,27,29,38,42,44] seven evaluated depression (n_exp = 176) [22,23,27,37,39,41,44], five evaluated stress (n_exp = 144) [20,34,35,36,43], and one evaluated aggressiveness (n_exp = 17) [40]. Key study characteristics are summarized in Table 1.

The majority of included trials, 64.3% (n = 18), were rated as having ‘some concerns’, 25.0% (n = 7) were classified as high risk, and 10.7% (n = 3) demonstrated low risk of overall bias (Figure 2). Domain 2 (deviations from intended interventions) was the most critical source of bias, with 79% of trials at high risk owing to the inherent inability to blind therapists and participants to BT allocation. Domain 4 (measurement of outcomes) was also problematic (61% high risk), reflecting the subjectivity of pain and psychological symptom assessment. Domain 1 (randomization process) was rated high risk in six of the seven overall high-risk trials. Domains 3 and 5 were predominantly low risk.

The primary random-effects model, with outliers (k = 14; REML estimator), yielded a pooled Hedges’ g = 0.72 (95% CI: 0.27 to 1.16; p = 0.002), indicating a statistically significant small-to-moderate effect of BTs on pain outcomes. Substantial heterogeneity was observed (τ² = 0.59; I² = 88.2%; Q = 94.7, p < 0.001). The 95% prediction interval (PI: -0.85 to 2.28) crossed zero, indicating that in future comparable clinical contexts the direction of effect cannot be assumed to be consistently positive (Figure 3a).

The near-complete reduction of heterogeneity after removal of Olson et al. [17] and McCormack [24] is methodologically instructive. Reiki was applied as an adjunct to opioid therapy in patients with advanced cancer (n = 24), producing an effect size with confidence intervals spanning approximately 166 standard deviation units, a result driven entirely by extreme intragroup variance given the very small sample size [17]. Conducted in elderly post-surgical patients with a non-contact, single-session targeted therapy design, the BT represented a markedly different clinical context from most other clinical trials on pain [24]. The contribution of these studies to the pooled estimate is better characterized as a sign of methodological heterogeneity than of genuine clinical effect, reinforcing the importance of influence diagnosis in systematic reviews.

The small positive effect of BTs on pain (g = 0.27 in the sensitivity model) is biologically plausible and consistent with evidence suggesting that non-pharmacological interventions can modulate pain perception through neurophysiological and psychosocial pathways. Specific mechanisms include: activation of the endogenous μ-opioid system via the descending pain inhibitory circuit [45]; facilitation of the relaxation response [46]; expectancy-driven analgesia mediated by prefrontal-limbic networks [47]; and, for touch-based modalities specifically, reduction of cortisol and modulation of oxytocin signaling at spinal and supraspinal levels [48,49].

A negative association between session duration and effect size magnitude was consistently observed across both the primary and sensitivity Pain models. While counterintuitive, this association may reflect study design artifacts: shorter sessions may have been tested in populations with more acute or higher-intensity pain where baseline scores allow greater absolute change, or in studies using more sensitive pain measurement tools. Alternatively, practitioner attention intensity during brief contacts may produce acute relaxation responses that attenuate pain transiently. These hypotheses require prospective investigation with standardized dosing protocols.

The descriptive finding that touch-based BTs (mean g = 0.83; CV = 0.94) produced more consistent outcomes than non-touch modalities (mean g = 0.68; CV = 1.33) for pain corroborates earlier work identifying physical contact as a moderator of BT efficacy [50]. The tactile component may facilitate muscle relaxation, reduce cortisol, and enhance the therapeutic alliance, mechanisms absent in distant or non-contact modalities. However, this comparison is exploratory; formal meta-regression with adequate sample sizes is required before clinical recommendations can be made.

The robust SAMD estimate (g = 0.42) is compatible with effect sizes reported for other non-pharmacological complementary interventions, such as mindfulness-based stress reduction and relaxation therapies, across anxiety and depression outcomes [51]. Symptom-specific subgroup analyses revealed an important internal gradient: anxiety and depression exhibited small, consistent effects (g = 0.32 and 0.39, respectively) with low heterogeneity (I² < 14%), whereas stress, despite showing the highest point estimate (g = 0.68), was characterized by substantial residual heterogeneity (I² = 73.1%) even after outlier exclusion.

This heterogeneity pattern is clinically meaningful: anxiety is well established as a condition characterized by autonomic dysregulation, specifically reduced vagal tone and sympathetic dominance [52], and reactive aggressiveness has similarly been theorized to involve impaired parasympathetic recovery from arousal states [53]. BTs, particularly touch-based modalities, appear to engage these autonomic pathways: Reiki has demonstrated significant anxiolytic effects (SMD = -0.82) across 824 participants in a recent meta-analysis [54], and mechanical stimulation of C-tactile afferents, the peripheral pathway activated by gentle touch, has produced measurable reductions in anxiety with corresponding EEG biomarker changes [55]. The low heterogeneity observed in the anxiety and depression subgroups of the present analysis supports the internal consistency of autonomic modulation as a plausible common therapeutic mechanism.

The influence analysis in the SAMD group identified Richeson et al. [44] and Tabatabaee et al. [22] as statistically decisive observations. Richeson et al., despite its small sample (n = 20) and waitlist control design, contributed multiple large effect sizes for pain, depression, and anxiety, a pattern consistent with high-expectancy, open-label contexts with minimal blinding. Tabatabaee et al. reported an implausibly large effect for depression (Hedges’ g ≈ 2.95) in cancer patients; such a magnitude could reflect scale directionality inconsistencies, ceiling-floor effects at baseline, or reporting errors. The selective sensitivity of the pooled SAMD estimate to these two studies (89.2% reduction in τ²) highlights how meta-analytical conclusions in this field can be dominated by a small number of underpowered or methodologically imprecise trials.

The multilevel model partitioned variance equally across levels (σ² ≈ 0.18 between studies; σ² ≈ 0.18 within-study across symptoms), confirming that anxiety, depression, stress, and aggressiveness are not interchangeable outcomes within a meta-analytical framework. Although an aggregate SAMD summary is statistically permissible for global synthesis, clinical interpretation requires symptom-specific reporting, as the mechanisms, instruments, and patient populations differ substantially across these outcomes.

A methodologically important observation is that publication bias, as assessed by Egger’s test and trim-and-fill, was not detectable in the primary models for either Pain (p = 0.542) or SAMD (p = 0.213), but became statistically evident in both groups after removal of the influential studies (Pain: p = 0.018; SAMD: p = 0.009). This paradoxical pattern suggests that influential outliers with large positive effect sizes may have obscured funnel plot asymmetry in the primary models, leading to a falsely symmetric appearance. After trim-and-fill adjustment in the Pain sensitivity model, the pooled estimate falls to g = 0.16, a value of limited clinical relevance. Although the trim-and-fill method has known limitations under high heterogeneity [15,56], this convergence of evidence warrants transparency in reporting. It should be incorporated in future updates as additional studies accumulate.

Several limitations should be acknowledged. First, the absolute number of trials per subgroup is small (k = 5-8), which limits the power of subgroup and meta-regression analyses. Second, 25% of included trials were rated as high overall risk of bias, and 64.3% had some concerns, primarily due to the inherent inability to blind therapists and recipients of BTs. Third, the stress subgroup retained substantial heterogeneity (I² = 73.1%) even after outlier removal, indicating the presence of additional sources of variability not captured by the available moderators. Fourth, the use of estimated SDs in some studies introduces measurement error into the calculation of effect sizes. Fifth, the trim-and-fill estimates should be interpreted with caution, given the residual heterogeneity and the limited number of studies in the sensitivity models. Finally, the analysis did not include unpublished trial data or grey literature beyond Google Scholar, which may introduce selection bias.

The BiFi REGs [57], developed through a Delphi consensus process specifically to standardize reporting of practitioner training, intervention fidelity, and protocol details in BT trials, provide a framework directly relevant to the limitations identified in this review. The null findings from meta-regression for protocol-related variables (intervention type, session frequency, duration) may partly reflect inconsistent and incomplete reporting of these variables in primary studies, precisely the reporting gaps that BiFi REGs were designed to address. We recommend that future systematic reviews in this area employ BiFi REGs as a complementary reporting quality assessment tool alongside RoB2 for methodological quality.