A systematic review and multivariate meta-analysis of the physical and mental health benefits of touch interventions

The study addresses how receiving touch influences physical and mental health across the lifespan and which factors moderate the efficacy of touch interventions. Motivated by the heightened awareness of touch reduction during COVID-19 and a heterogeneous literature spanning cohorts (newborns, children, adults), touch types (e.g., massage, kangaroo care), outcomes (e.g., cortisol, pain, mood), and who applies touch (familiar vs. professional; human vs. object/robot), the authors conduct a large-scale, preregistered systematic review and multilevel, multivariate meta-analysis. Their primary aims are to quantify overall health benefits of touch interventions and identify moderators related to dyadic dynamics (human vs. object/robot, familiarity, skin-to-skin, directionality), demographics (clinical status, age, sex), delivery (type of touch, body part), and procedure (duration, session number), analyzing newborns separately due to distinct outcomes.

Prior meta-analyses have shown benefits of touch but were limited by scope, focusing on specific touch types (e.g., massage), populations (e.g., preterms), or single outcomes (e.g., neck/shoulder pain, cancer pain). These studies often used separate random-effects models without accounting for multivariate reporting and multilevel data structures. Evidence also existed for benefits in animals, though sparse. Gaps included limited identification of meaningful moderators (e.g., toucher familiarity, skin-to-skin contact, human vs. object/robot), underpowered primary studies, and narrow intervention types. This review expands the evidence base, includes diverse touch interventions beyond massage/kangaroo care, and applies multivariate, multilevel methods to better capture dependence among multiple outcomes within studies.

Preregistration: PROSPERO CRD42022304281 with documented deviations reflecting the multivariate/meta-analytic addition, exclusion of non-randomized trials and dissertations, and inability to implement a Bayesian RoBMA due to multilevel/multivariate constraints. - Search: Google Scholar, PubMed, Web of Science from 2 Dec 2021 to 1 Oct 2022; languages: English, German, Dutch, French (yielded English-only articles). PRISMA 2020 guidelines followed. - Eligibility: Experimental studies with explicit physical touch interventions (human/animal/object/robot) contrasted against no-touch controls (or matched active controls), measuring health outcomes (physical or mental) in humans or animals. For meta-analysis: between-subjects design, randomized allocation (non-randomized excluded), sufficient data for effect size computation; immediate effects only (long-term without intervening touch excluded). Animal outcomes included in the systematic review but excluded from meta-analysis due to insufficient studies. - Data extraction: Two independent coders; discrepancies resolved by senior authors. Extracted descriptive/stats to compute Cohen’s d then Hedges’ g (using standard converters), preferentially delta change scores; handling of threshold p-values and zero frequencies per metafor defaults. For studies with multiple timepoints within a day, peak effect chosen; repeated sessions treated as separate immediate effects. Control group prioritized to most closely match touch condition (e.g., relaxation vs. standard care). Moderators coded per preregistration (e.g., mental vs. physical outcomes; dyad type; familiarity; skin-to-skin; body part; directionality; session number; duration; sex ratio; age; clinical status; study location). - Samples: Meta-analysis included adults/children (85 studies, 103 cohorts; touch n=2,841, control n=2,556) and newborns (52 studies, 63 cohorts; touch n=2,134, control n=2,086). Systematic review included an additional 75 studies and 19 animal studies (n=911 animals across species). - Statistical analysis: R (4.2.2) metafor rma.mv multilevel, multivariate models with random effects at study, cohort, effect levels; variance–covariance matrix assuming within-cohort effect correlation ρ=0.6 (sensitivity 0–0.8 showed robust conclusions). Robust variance estimation with cluster-robust inference at cohort level. Heterogeneity via Cochran’s Q and random-effect σ²; influential points via Cook’s D (threshold >0.5). Small-study bias assessed via funnel plots and modeling standard error as moderator. Power sensitivity analyses determined minimum k (adults k≥9; newborns k≥8) for subgroup analyses at 80% power using median sample sizes and τ²≈62.5%. Moderator significance via omnibus F-tests; post hoc comparisons via two-sided t-tests without multiple-comparisons correction. Firepower plots assessed primary study power. Risk of bias assessed for randomization, sequence, performance (blinding), attrition; performance bias inherently high for adult studies due to non-blindable conditions; moderate for newborns/animals.

- Overall effects: Adults/children Hedges’ g=0.52 (95% CI 0.42–0.63; t(102)=9.74, P<0.001; k=469 effects). Newborns g=0.56 (0.41–0.71; t(62)=7.53, P<0.001; k=174 effects). Mental vs. physical benefits did not differ significantly in adults (Δg=−0.05, P=0.432) or newborns (Δg=−0.19, P=0.284). - Specific outcomes (adults; k and g [95% CI]): Pain k=67, g=0.69 [0.48, 0.89]; State anxiety k=78, g=0.64 [0.44, 0.84]; Depression k=33, g=0.59 [0.40, 0.78]; Trait anxiety k=25, g=0.59 [0.40, 0.77]; Cortisol k=31, g=0.53 [0.33, 0.72]; Systolic BP k=21, g=0.47 [0.20, 0.74]; Mobility k=10, g=0.41 [0.16, 0.66]; Positive affect k=18, g=0.40 [0.18, 0.62]; Diastolic BP k=21, g=0.39 [0.11, 0.68]; Negative affect k=33, g=0.37 [0.18, 0.57]; Sleep k=14, g=0.31 [0.15, 0.47]; Heart rate k=31, g=0.27 [0.07, 0.46]; Respiration k=13, g=0.23 [−0.00, 0.46]. Post hoc: Pain, state and trait anxiety, depression > respiratory/sleep/heart rate; pain > positive affect; cortisol > heart rate. - Specific outcomes (newborns): Liver enzymes k=9, g=0.84 [0.37, 1.32]; Cortisol k=8, g=0.78 [0.24, 1.31]; Weight gain k=45, g=0.65 [0.37, 0.94]; Respiration k=10, g=0.61 [0.25, 0.97]; Temperature k=9, g=0.59 [0.17, 1.01]; Heart rate k=12, g=0.35 [−0.02, 0.72] (ns); Digestion k=10, g=0.32 [−0.17, 0.80] (ns). No significant differences across newborn outcomes with sufficient data. - Human vs. object/robot touch (adults): Overall non-significant difference (Δg=0.12, P=0.295). Physical health: human-object g=0.56 [0.24, 0.88]; human-human g=0.51 [0.38, 0.64] (similar). Mental health: human-object g=0.34 [0.19, 0.49] vs. human-human g=0.58 [0.43, 0.73]; Δg=0.24 (P=0.022) favoring human-human. Skin-to-skin among human-human suggested larger mental benefits with skin-to-skin (Δg=0.41, P=0.055; borderline, limited power). - Type of touch: No significant differences between massage and other touch types for adults overall (Δg=0.02, P=0.916), or split by outcome; newborns similarly showed no differences among massage, kangaroo care, and other forms. - Clinical status: Significant benefits in both clinical and healthy cohorts. Adults: mental health benefits larger in clinical vs. healthy (Δg=0.25, P=0.037). Newborns: no significant differences by clinical status overall or by outcome. - Familiarity: Adults showed no effect of familiar vs. unfamiliar toucher overall or by outcome. Newborns: parental (familiar) touch more beneficial than unfamiliar (Δg=0.30, P=0.041); differences not significant when split by outcome. - Frequency and duration: Adults—more sessions associated with larger benefits overall (slope g≈+0.02 per session; F(1,101)=23.99, P<0.001), and for physical (P=0.007) and mental (P<0.001) outcomes; specifically, trait anxiety (slope +0.03, P<0.001), depression (+0.03, P<0.001), pain (+0.03, P<0.001). Duration per session not associated with greater benefits; negative associations for cortisol (t(24)=2.71, g=−0.01, P=0.012) and heart rate parameters (t(21)=2.35, g=−0.01, P=0.029). Newborns—no significant association of number of sessions or duration with outcomes. - Sex and age: Adults—no overall associations with sex ratio or mean age; specific findings: higher female ratio associated with larger cortisol reductions (t(18)=2.31, g=+0.01, P=0.033); older mean age associated with improved positive affect (t(10)=2.54, g=+0.01, P=0.030) and systolic BP (t(11)=2.39, g=+0.02, P=0.036). Newborns—no associations with sex ratio. - Body part: Head touch conferred higher benefits vs. arm (Δg=+0.78, P=0.039) and torso (Δg=+0.84, P=0.031); greater physical benefit for head vs. torso (Δg=+0.96, P=0.043). Arm touch yielded lower mental vs. physical benefits (Δg=−0.35, P=0.028). - Directionality: Unidirectional touch yielded higher benefits than bidirectional in adults overall (Δg=+0.30, P=0.032), driven by mental health (Δg=+0.46, P=0.022). - Study location: Adults—stronger benefits in South America vs. North America (Δg=+0.37, P=0.046) and vs. Europe (Δg=+0.36, P=0.029). Newborns—weaker effects in North America vs. Asia (Δg=−0.37, P=0.026) and vs. Europe (Δg=−0.40, P=0.022). No interactions with outcome type. - Small-study bias: Significant small-study bias detected in adult (F(1,101)=21.24, P<0.001) and newborn (F(1,61)=5.25, P=0.025) meta-analyses; Cook’s D <0.06 indicated no overly influential individual effects. Many single studies underpowered per firepower plots. - Systematic review of animal studies: Across 19 studies (rats, mice, macaques, cats, lambs, fish), 71.4% of mental-like and 81.8% of physical outcomes showed benefits; most touch delivered by humans; interventions included stroking and tickling.

The findings demonstrate that touch interventions yield robust, medium-sized benefits across physical and mental health in both adults/children and newborns. The analyses clarify which factors modulate efficacy. In adults, strong effects on pain, anxiety, and depression pinpoint touch as particularly effective for mental health symptoms and pain management, while newborns benefit most in physiological domains such as weight gain and hormonal/metabolic regulation. Human-applied touch confers superior mental health benefits relative to object/robot touch, with skin-to-skin contact likely mediating this advantage. Frequency matters more than duration in adults, particularly for trait outcomes (depression, anxiety) and pain, indicating cumulative benefits of repeated sessions without requiring prolonged sessions. Clinical populations gain larger mental health benefits than healthy individuals, consistent with greater touch wanting and heightened symptomatology. Familiarity of toucher appears unimportant in adults but critical in newborns, where parental touch provides additional benefit, aligning with the importance of early skin-to-skin and maternal cues for adaptation. Specific procedural and contextual moderators—body region (head), directionality (unidirectional), and geography—further refine optimal application contexts. Limited sex- and age-related moderators suggest broad applicability across demographics, with some evidence of stronger stress-buffering in women and enhanced BP modulation with age. Collectively, the study addresses prior gaps by leveraging multivariate, multilevel methods, integrating diverse interventions, and systematically testing moderators, thereby providing actionable guidance on optimizing touch interventions and extending their applicability (including potential for object/robot-mediated touch in constrained contexts).

This large-scale, preregistered systematic review and multilevel, multivariate meta-analysis shows that touch interventions reliably improve both mental and physical health across ages and health statuses with medium effect sizes. The greatest benefits in adults/children occur for pain, anxiety, and depression, and in newborns for weight gain and physiological regulation. Efficacy is enhanced by more frequent sessions (adults), head-focused touch, unidirectional delivery, and human-applied, skin-to-skin contact for mental outcomes, while touch type (massage vs. others) is flexible without major differences. Clinical cohorts derive larger mental health improvements than healthy cohorts; parental touch is especially beneficial for newborns. Future research should: improve study power; assess long-term outcomes; report sex/gender-disaggregated effects; clarify mechanisms (e.g., oxytocin pathways); optimize robot/object touch (e.g., synthetic skin, social bonding); and examine cultural/contextual influences on pleasantness and acceptability to further tailor interventions.

- Small-study bias present in both meta-analyses, potentially inflating effects due to unpublished null results. - Impossibility of blinding participants and experimenters in touch vs. no-touch interventions increases performance/placebo risks. - Multiple moderator tests without correction raise false-positive risk (some near-threshold findings). - Language restrictions (primarily Western languages) may have missed non-English/non-Western studies despite substantial research in Asia. - Limited data for secondary moderators by specific outcomes constrained specificity (e.g., precise mental health subdomains for skin-to-skin or human vs. object). - Long-term outcomes largely not included; conclusions focus on immediate effects. - Sparse reporting of sex/gender-specific outcomes in primary studies limited deeper analyses. - Meta-analysis excluded animal outcomes due to insufficient data; generalizability to animals based on systematic review only. - Duration and time-span confounded with session number (high correlation), limiting independent assessment of some procedural variables. - Constraints prevented Bayesian RoBMA modeling for publication bias in multilevel/multivariate data structures.