The effect of self-compassion versus mindfulness interventions on autonomic responses to stress in generalized anxiety disorders

Generalized anxiety disorder (GAD) is characterized by chronic worry, tension, and somatic symptoms and remains challenging to treat effectively with medication or psychotherapy. Mindfulness- and self-compassion–based interventions have shown promise for reducing self-reported anxiety and improving well-being, but their effects on physiological arousal are less clear. Heart rate (HR) is a useful marker of sympathetic activation in anxiety disorders. Prior work suggests self-compassion may soothe sympathetic arousal and that mindfulness may enhance emotion regulation, yet findings on physiological outcomes—especially HR responses to stress—are mixed. Because mindfulness is theorized as a component of self-compassion but may engage distinct biological systems, direct comparisons are needed in clinical GAD populations. This study tested whether brief group-based self-compassion or mindfulness interventions (each adjunctive to usual pharmacotherapy) reduce HR reactivity to an induced stressor, and whether they improve state anxiety and affect. The authors hypothesized both interventions would reduce HR responses and decrease state anxiety and negative affect.

The paper reviews evidence that: (a) self-compassion is linked to better mental health and reduced anxiety/depression, with meta-analytic support for benefits in GAD; (b) self-compassion may downregulate sympathetic arousal (e.g., HR, HRV) and engage mammalian caregiving systems (oxytocin, affiliation-related neural circuits); (c) mindfulness interventions show efficacy for GAD but have mixed effects on physiological markers such as HR and blood pressure, with some studies showing reductions and others no change; (d) mindfulness is associated with prefrontal regulatory circuits and is a core component of self-compassion, yet the two may involve distinct mechanisms; (e) both self-compassion and mindfulness have been associated with increased HRV in prior research. Few studies directly compare these interventions in GAD, motivating the present work to clarify overlapping and unique physiological effects, particularly on sympathetic arousal indexed by HR during acute stress.

Design: Post-hoc analysis of a nonrandomized controlled clinical trial comparing self-compassion + treatment as usual (TAU), mindfulness + TAU, and TAU alone in adults with GAD. Pre- and post-intervention assessments included questionnaires and an induced stress task during ECG recording.

Participants: Recruited from Hangzhou Seventh People's Hospital. Inclusion: age 18–65, DSM-5 GAD (M.I.N.I.), HAMA ≥ 14, HAMD < 23. Exclusion: major psychiatric/medical comorbidities (e.g., bipolar disorder, suicidality, substance use disorder), severe physical disease, cognitive/hearing impairment, concurrent psychotherapy. Ethics approval obtained; informed consent provided. Of 75 enrolled in the parent study (25/group), 47 had complete pre/post HR data and were analyzed: Self-compassion (n=19), Mindfulness (n=17), TAU (n=11). Exclusions (n=28) were due to missing post data, technical issues, or muscle noise.

Interventions: Group-based programs delivered over two weeks (8 sessions) adjunctive to pharmacotherapy for active arms. Self-compassion content included affectionate breathing, compassionate body scan/movement, self-compassion break, and meditations to cultivate self-soothing during distress. Mindfulness emphasized body movement as anchor (mindful breathing, standing, stretching, awareness of sounds and thoughts) to cultivate nonjudgmental present-moment awareness. TAU received pharmacotherapy only.

Stress task: 13 difficult Raven’s Standard Progressive Matrices items (Chinese City Edition). Participants were told the test predicts intelligence and life outcomes. Each trial: 1 s fixation, 20 s item with response, followed by 6 s feedback; only two trials randomly marked “Correct,” remaining “Wrong,” and the task ended with “Unfortunately, you failed!” to induce evaluative stress.

Measures: State anxiety (STAI-S, 20 items, 4-point scale), Positive and Negative Affect Schedule (PANAS; 10-item PA, 10-item NA, 5-point scale), clinician-rated HAMA and HAMD at baseline. STAI-S and PANAS were administered pre- and post-intervention before the stress task.

ECG/Heart rate acquisition: BITalino (r)evolution Board, 3 chest electrodes (bilateral clavicles and left lower rib), 1000 Hz sampling via OpenSignals. Inter-beat intervals (IBIs) derived using the Pan–Tompkins algorithm; artifacts visually inspected/edited per guidelines. IBIs converted to beats per minute (BPM). Continuous data segmented time-locked to feedback onset (−1 to 6 s). Trials with ‘Wrong’ feedback were retained and baseline-corrected per trial using −1 to 0 s (0 = feedback onset) to control for baseline HR and capture event-related dynamics over a short window.

Rationale for HR vs HRV: HRV metrics typically require longer recordings; the design targeted a 6 s stress-evoked sympathetic window, so HR was analyzed.

Sliding window analysis: Window length 500 ms, step 50 ms across 0–6 s after feedback; paired t-tests compared pre vs post within each group to detect dynamic treatment effects.

Statistics: Baseline group differences assessed via t-tests/χ² (SPSS v23). Two-way ANOVAs (Group: Self-compassion, Mindfulness, TAU; Time: Pre, Post) for STAI-S and PANAS with Greenhouse–Geisser correction as needed; Bonferroni-adjusted post-hoc tests (α<0.05). For HR, due to a baseline difference across groups (p=0.008), baseline correction from pre to post was performed per subject; then one-way ANOVA on 6 s average HR change by group, with Bonferroni post-hoc tests. Pearson correlations examined associations between HR change and subjective measures.

• Sample characteristics: No significant baseline differences across groups in age, gender, education, employment, anxiety, or depression; marital status differed (p=0.046).
• State anxiety (STAI-S): Main effect of time (F1,44=26.61, p<0.001, η²p=0.38), indicating decreased state anxiety from pre (Mean=46.75) to post (Mean=37.41) across groups; effect mainly driven by the Self-compassion group (t=4.65, Pcorrected<0.001). No significant Group or Time×Group interaction.
• Positive affect (PANAS-PA): Main effect of time (F1,44=11.20, p=0.002, η²p=0.20), increased from pre (Mean=24.23) to post (Mean=27.28); effect mainly driven by the Mindfulness group (t=−3.07, p=0.007). No significant Group or interaction effects.
• Negative affect (PANAS-NA): Main effect of time (F1,44=67.36, p<0.001, η²p=0.61) and Group effect (F2,44=3.66, p=0.034, η²p=0.14), with post-hoc showing Mindfulness < TAU (Pcorrected=0.031) post-intervention.
• Heart rate (HR) change (6 s after negative feedback): Group effect (F2,46=5.35, p=0.008). Post-hoc: Self-compassion reduced HR response more than Mindfulness (Pcorrected=0.018) and TAU (Pcorrected=0.037); Mindfulness vs TAU not significant (Pcorrected=1.000).
• Correlation: Across all participants, greater decreases in HR from pre to post were associated with greater reductions in negative affect (r=0.30, p=0.04, n=47).

The study directly compared short-term self-compassion versus mindfulness interventions, adjunctive to usual pharmacotherapy, on sympathetic stress reactivity in GAD. Self-compassion uniquely reduced heart rate responses to an evaluative stressor, supporting the hypothesis that it downregulates sympathetic arousal and improves emotional regulation in GAD. Both interventions improved self-reported outcomes—reducing state anxiety and negative affect and increasing positive affect—indicating broad benefits on mood. The association between decreased HR reactivity and reduced negative affect further links physiological and affective improvements.

Mindfulness did not reduce HR reactivity relative to TAU, diverging from some prior findings. Potential explanations include reduced autonomic flexibility in chronic clinical populations, insufficient dose or duration (2 weeks) to modulate sympathetic responses under acute stress, and heterogeneity across mindfulness protocols. Distinct mechanisms may underlie the interventions: compassion-related processes may engage caregiving and affiliation systems (e.g., oxytocinergic pathways), whereas mindfulness emphasizes prefrontal regulatory networks. Prior literature showing both approaches can increase HRV suggests they may differentially influence sympathetic versus parasympathetic components. Clinically, self-compassion might be especially useful for dampening physiological reactivity and state anxiety, while mindfulness may confer advantages for enhancing positive affect and reducing negative affect.

This study provides novel physiological evidence that a brief, group-based self-compassion intervention can reduce sympathetic stress reactivity (heart rate responses) and improve state anxiety in patients with GAD. Mindfulness training improved affective outcomes but did not alter heart rate reactivity over the 2-week course. These findings suggest complementary, mechanism-specific benefits and support incorporating self-compassion into GAD treatment to buffer stress physiology. Future research should employ larger, randomized designs; extend intervention duration and dosage; include comprehensive autonomic indices (e.g., HRV); and utilize standardized stress paradigms to confirm and generalize these effects.

• Nonrandomized design limits causal inference and raises potential selection bias despite largely comparable baselines.
• Small and unequal group sizes (n=19, 17, 11) reduce power and precision; notable attrition and exclusions due to missing data and technical artifacts.
• Baseline differences in heart rate required correction; marital status differed across groups.
• Short intervention duration (2 weeks) may be insufficient to elicit physiological changes, particularly for mindfulness.
• Heart rate variability was not analyzed due to the brief (6 s) analysis window; reliance on HR alone limits autonomic system characterization.
• The stress task, while practical, is not a classical paradigm like the Trier Social Stress Test, which may affect generalizability.