Symptom-specific multilevel response architectures of virtual reality analgesia in zoster-associated pain

Technical Summary: Symptom-Specific Multilevel Response Architectures of Virtual Reality Analgesia in Zoster-Associated Pain

Problem Statement
Zoster-associated pain (ZAP), spanning acute herpes zoster neuralgia to postherpetic neuralgia (PHN), presents a complex sensory-affective burden often inadequately managed by pharmacological interventions due to side effects and incomplete relief. While immersive Virtual Reality (VR) is a promising non-pharmacological modality for pain and distress, current understanding often relies on a "distraction account" that treats VR as a uniform exposure. This approach obscures clinically critical nuances: different VR contents (e.g., psychoeducation, mindfulness, breathing training) may engage distinct therapeutic mechanisms, and symptom relief may vary based on individual differences, session-specific states, and their interactions. Furthermore, existing research often conflates scene-level effects with participant-level tendencies by relying solely on pre-post change scores or scene-average comparisons, failing to disentangle how specific physiological and subjective response features align with specific symptom domains (pain vs. anxiety).

Methodology
The study employed a counterbalanced within-participant design involving 27 adults with ZAP. Participants underwent three distinct VR interventions on separate inpatient days:

1. Psychoeducation
2. Mindfulness Meditation
3. Breathing Training

Data Acquisition:

• Subjective Measures: Patient-reported outcomes (Pain Intensity via VAS, State Anxiety via STAI, Depression via BDI, and Affect via PANAS) were collected pre- and post-session. Immersion was rated post-exposure.
• Physiological Measures: 5-minute resting-state EEG and ECG recordings were acquired immediately before and after each VR session.
  • EEG: 24-electrode system (Neurogin SR1), analyzed via MNE-Python. Spectral power was estimated for delta, theta, alpha, beta, low gamma, and high gamma bands.
  • ECG: Time-domain heart rate variability (HRV) metrics (RMSSD, SDNN, etc.) were derived from pre-to-post changes.

Analytical Framework:

• Statistical Modeling: Covariate-adjusted mixed-effects models were used to account for the repeated-measures design.
• Decomposition: Immersion, EEG, and ECG features were decomposed into within-participant (session-specific deviations from a participant's mean) and between-participant (participant-level average tendencies) components.
• Feature Selection: Whole-head cluster-based permutation tests identified significant EEG spectral clusters associated with Time and Scene interactions.
• Architecture Comparison: Candidate response architectures were compared using Akaike Information Criterion (AIC) and Akaike weights. Models tested whether symptom variation was best explained by individual-difference components, scene-state components, or scene-modulated associations. Serial mediation analyses were conducted as exploratory sensitivity checks.

Key Results

1. Broad Symptom Reduction: All three VR scenes resulted in significant short-term reductions in pain intensity, state anxiety, depressive symptoms, and negative affect. Positive affect increased most notably after breathing training. However, there were no clear average differences in symptom change magnitude across the three scenes.
2. Distinct Symptom Axes:
   • Pain (VAS): Pain variation was most strongly associated with immersion (both within- and between-participant components) and low gamma EEG (Cluster 3: C3, P3, O1). The low gamma association was dominated by a breathing-specific within-participant slope. RMSSD (HRV) showed scene-modulated between-participant associations with pain.
   • Anxiety (STAI): Anxiety variation involved a more distributed pattern, linking to immersion, theta-band EEG (Cluster 2: C3, Cz, C4), and SDNN (HRV). Associations were present in both within- and between-participant components, particularly in the psychoeducation condition.
3. Multilevel Response Architectures:
   • Pain Architecture: The best-supported model for pain retained multilevel terms plus scene-modulated between-participant ECG associations. This architecture was heavily driven by scene-state components (low gamma EEG and immersion), with a significant loss of model support when scene-state components were removed.
   • Anxiety Architecture: The best-supported model for anxiety included multilevel cortical, autonomic, and immersion terms. However, model support was more distributed; both individual-difference and scene-state components contributed significantly, and alternative architectures remained plausible.
4. Physiological Specificity: The study found no single physiological signature for VR analgesia. Instead, pain relief aligned with scene-state low gamma changes and immersion, while anxiety relief involved a broader mix of theta-band EEG, SDNN, and immersion components.

Key Contributions

• Differentiation of Response Architectures: The study demonstrates that pain and anxiety relief following VR are not merely different endpoints of a single process but are organized around distinct multimodal association patterns. Pain is more tightly coupled with scene-state cortical (low gamma) and immersion factors, whereas anxiety reflects a more balanced mixture of individual tendencies and scene-state factors.
• Methodological Advancement: By decomposing variables into within- and between-participant components and utilizing mixed-effects models, the study separates scene-specific effects from individual response tendencies, offering a more granular view of VR efficacy than average pre-post comparisons.
• Content-Agnostic Efficacy with Content-Specific Mechanisms: While all scenes reduced symptoms, the underlying physiological and subjective drivers of that relief varied by symptom domain and scene type (e.g., the specific link between breathing training and low gamma/pain).

Significance and Claims
The authors claim that these findings support a response-informed account of VR-associated symptom relief in clinical ZAP. The study argues that clinical translation of VR analgesia requires moving beyond the question of "does VR work?" to "how does relief vary across symptom domains and response features?"

The paper posits that understanding these symptom-specific multilevel response architectures is essential for developing personalized digital interventions. Specifically, the distinct alignment of pain with scene-state cortical/immersion factors versus anxiety with a broader mix of factors suggests that tailoring VR content and monitoring strategies to "pain-dominant" versus "anxiety-dominant" profiles may improve clinical outcomes.

The authors explicitly state that their models are descriptive syntheses of association patterns rather than causal mediation models, given the pre-post resting-state nature of the physiological data and the lack of a non-VR control condition. They emphasize that while the findings are hypothesis-generating, they provide a structural framework for future prospective trials to test whether matching VR content to specific response architectures enhances therapeutic efficacy.