This retrospective, nonrandomized, natural-group comparison used anonymized medical records from inpatients with depressive disorders who were treated in the Clinical Psychology Department of Lishui Second People’s Hospital, Lishui, China, from January 1 to December 31, 2022. The VR group was naturally defined by receipt of adjunctive VR-based relaxation training during hospitalization, whereas the control group received usual care only. Because group allocation was not randomized and was based on routine clinical care, the study can identify associations but cannot establish causal effects.

Patients were eligible if they met International Classification of Diseases, Tenth Revision (ICD-10) diagnostic criteria for depressive episodes or recurrent depressive disorders, were aged 16 to 60 years, and did not receive psychotherapy or modified electroconvulsive therapy during hospitalization. Patients were excluded if they had severe comorbidities; significant visual or auditory deficits; a history of psychotic features, psychotic disorders, or affective disorders; high suicide risk (Mini International Neuropsychiatric Interview suicidal tendency item score ≥10); severe personality disorders (eg, paranoid or antisocial personality disorder); or intellectual disability. Patients with other psychiatric comorbidities were not excluded.

A total of 665 inpatient records were screened. After applying inclusion and exclusion criteria, 133 (20%) patients were included in the analytic sample (including n=63, 47.4% in the VR group and n=70, 52.6% in the control group). Major demographic and clinical characteristics included age, sex, education, marital status, occupation, BMI, course of disease, length of hospitalization, family history, number of repetitive transcranial magnetic stimulation (rTMS) sessions, and number of biofeedback sessions.

All available inpatient medical records from the predefined study period were screened. The study therefore used a retrospective consecutive-record sampling approach rather than prospective recruitment. Data were collected from hospital records and clinician-rated admission and discharge assessments. The setting was a single psychiatric hospital department. No new participants were recruited for this retrospective analysis, and no compensation was provided.

The sample size was determined by the number of eligible complete medical records available during the study period. Because this was a retrospective analysis of routine clinical data, no a priori power calculation was conducted before data collection. To describe the precision of the estimates, we reported 95% CIs for mean differences and odds ratios (ORs) wherever applicable.

The primary symptom measures were the 17-item Hamilton Depression Rating Scale (HAMD) [19] and the Hamilton Anxiety Rating Scale (HAMA) [20], assessed at admission and discharge. Response was defined as a 50% or greater reduction from baseline in HAMD or HAMA total score, and remission was defined as a total score ≤7 [21]. Demographic and clinical covariates used for baseline comparability analyses included age, sex, education, marital status, occupation, BMI, course of disease, length of hospitalization, family history, rTMS sessions, and biofeedback sessions.

HAMD and HAMA ratings were conducted by trained attending physicians using standardized interviews. Assessors received standardized training on item interpretation, scoring criteria, interview procedures, and interrater reliability. Assessments were conducted in quiet, private rooms at admission and discharge.

All patients received usual care, including antidepressant medication and physiotherapy as clinically indicated. Physiotherapy included rTMS and biofeedback therapy. Patients in the VR group additionally received 25-minute VR-based relaxation training approximately 3 times per week. The VR program used immersive natural or indoor relaxation scenes with audio guidance based on progressive relaxation and/or autogenic techniques. All VR sessions were delivered by the same trained psychotherapist. Representative VR relaxation training scenes are shown in Figures 1-5, and the experimental setting is shown in Figure 6.

Baseline assessments were conducted at admission, and posttreatment assessments were conducted at discharge. Data were extracted from complete anonymized medical records. Because this was a retrospective naturalistic study conducted in a routine clinical setting, prospective masking of patients and clinicians to treatment exposure was not feasible. However, symptom assessments were conducted using standardized clinician-rated scales and predefined scoring criteria to reduce assessment variability.

Before analysis, all variables included in the analytic dataset were checked against the complete medical records. No missing values were identified for any variables included in the analyses; therefore, no data imputation was performed. Distributional characteristics were evaluated to determine the appropriate statistical tests for continuous variables.

SPSS (version 24; IBM Corp) was used for statistical analyses. Baseline demographic and clinical variables were compared between groups using independent samples t tests for normally distributed continuous variables, Mann-Whitney U tests for nonnormally distributed continuous variables, and chi-square tests or Fisher exact tests for categorical variables as appropriate. Between-group differences in HAMD and HAMA scores were evaluated using mean differences and 95% CIs. Response and remission rates were compared using ORs and 95% CIs. Exploratory analyses examined session-frequency subgroups within the VR group and hospitalization duration–matched control subgroups. All tests were 2-sided, with statistical significance set at P<.05.

This study involved a retrospective secondary analysis of human-subject medical record data and was reviewed and approved by the Ethics Review Committee of Lishui Second People’s Hospital (023/2023; approved November 17, 2023). All procedures in this study were performed in compliance with the Declaration of Helsinki. Because this study used existing clinical records, no additional intervention, prospective recruitment, or study-specific procedure was conducted for this analysis. Written informed consent for clinical data collection and future anonymized scientific use had been obtained during the original clinical data collection, and the ethics approval permitted the secondary analysis of anonymized data without additional consent.

All data used for analysis were deidentified before extraction and analysis. The analytic dataset did not contain names, contact information, medical record numbers, facial images, or other direct identifiers. Only coded and anonymized data were accessed by the research team, and the data were used only for the purposes approved by the ethics committee. No compensation was provided because no new participants were recruited and no additional study-specific procedures were performed.

All figures and supplementary materials were reviewed to ensure that no individual participant or user could be identified. For Figure 6, which shows the experimental scene, the image of the individual wearing the VR device was masked or deidentified to remove facial features and other potentially identifying information. Therefore, no individual participant or user can be identified from the revised figure, and additional consent specifically for publication of an identifiable image was not applicable.

The patient selection flowchart is shown (Figure 7). Baseline characteristics of the study participants are presented in Table 1. There were no significant differences between the VR group and the control group in any demographic or clinical variables at baseline. The mean BMI was 21.65 (SD 3.72) in the VR group and 21.31 (SD 3.47) in the control group (mean difference 0.35, 95% CI −0.89 to 1.58; t₁₃₁= 0.56; P=.58). The median age was 18 (IQR 15‐44) years in the VR group and 17 (IQR 15‐43) years in the control group (median difference 1.0, 95% CI −1.0 to 3.0; U=2010.0; P=.38). Similarly, no significant differences were observed in sex distribution (OR 2.19, 95% CI 0.99‐4.87; χ²₁=3.8; P=.05), education level (χ²₄=3.1; P=.54), marital status (χ²₂=0.6; P=.73), occupation (χ²₄=5.6; P=.23), or family history (OR 0.66, 95% CI 0.14‐3.07; χ²₁=0.3; P=.60).

Table 2 summarizes the comparisons of clinical outcomes between the 2 groups. At baseline, there were no significant differences in HAMD scores (mean difference −1.38, 95% CI −3.56 to 0.81; t₁₃₁=−1.25; P=.22) or HAMA scores (mean difference 0.29, 95% CI −0.82 to 1.40; t₁₃₁=0.52; P=.60) between groups.

After the intervention, the VR group showed significantly lower HAMD scores compared with the control group (mean difference −2.59, 95% CI −4.22 to −0.96; t₁₃₁=−3.09; P=.002). Similarly, HAMA scores were significantly lower in the VR group after treatment (mean difference −2.45, 95% CI −3.64 to −1.27; t₁₃₁=−4.10; P<.001). The 95% CIs for these differences did not cross zero, indicating a high level of precision in the estimated treatment effects.

For remission (total score ≤7), the VR group had a significantly higher HAMA remission rate (49/63, 77.8% vs 41/70, 58.6%; OR 2.48, 95% CI 1.16‐5.30; χ²₁=5.6; P=.02). Although the point estimate suggests a clinically meaningful benefit, the CI, while not crossing 1, is relatively wide (1.16‐5.30), indicating moderate precision. In contrast, the difference in HAMD remission rates did not reach statistical significance (39/63, 61.9% vs 35/70, 50%; OR 1.63, 95% CI 0.81‐3.24; χ²₁=1.9; P=.17). The CI for HAMD remission includes 1, confirming the lack of statistical significance, and its width (0.81‐3.24) reflects the uncertainty surrounding this estimate.

The HAMD response rate (≥50% reduction) was significantly higher in the VR group (58/62, 92.1% vs 53/70, 75.7%; OR 3.72, 95% CI 1.28‐10.79; χ²₁=6.4; P=.01). The HAMA response rate was also significantly higher in the VR group (57/63, 90.5% vs 46/70, 65.7%; OR 4.96, 95% CI 1.87‐13.14; χ²₁=11.6; P<.001). While both ORs demonstrate a substantial benefit for the VR group, the relatively wide CIs (eg, 1.28‐10.79 for HAMD response) suggest some imprecision in the exact magnitude of the effect.

To further explore whether symptom outcomes varied by session frequency of VR relaxation training, the VR group was divided according to the number of treatment sessions: 2 to 4 sessions (VR1; n=19), 5 to 7 sessions (VR2; n=22), and 8 sessions or more (VR3; n=22). The VR group was also matched with control groups according to the length of hospitalization, with less than 9 days for control group 1 (CG1; n=23), 10 to 16 days for control group 2 (CG2; n=22), and 17 days or more for control group 3 (CG3; n=25).

There were no differences between the 3 treatment groups and their controls, in terms of age, course of disease, length of hospitalization, BMI, number of physiotherapy sessions (ie, biofeedback and rTMS), marital status, level of education, occupation, sex, and family history (Table S1 in Multimedia Appendix 1).

At baseline, there were no significant differences in HAMD and HAMA scores between the VR groups and their controls. There were no significant differences in HAMD (F_5,127=2.002; P=.08) and HAMA (F_5,127=1.458; P=.21) scores among the 6 groups. At discharge, VR1’s HAMD score was not significantly different from its control group and its HAMA score was significantly lower than that of CG1 (t(40)=−2.338; P=.02). VR2’s HAMD and HAMA scores were significantly lower than those of its control group (HAMD: t(42)=−2.340; P=.02; HAMA: t(42)=−2.177; P=.04). VR3 also had significantly lower HAMD and HAMA scores than its controls (HAMD: t(45)=−2.645; P=.01; HAMA: t(45)=−2.969; P=.005; Table S2 in Multimedia Appendix 1 and Figures 8 and 9).

There was no significant difference between VR1 and CG1 in terms of response and remission rates for depressive symptoms. VR1 had a significantly higher response rate than CG1 (χ²₁=6.3; P=.01), whereas the remission rate did not differ significantly between groups (χ²₁=3.1; P=.08). VR2 and CG2 did not have a significant difference in terms of response and remission rates, and the remission rate of anxiety symptoms was not significantly different; however, the response rate of anxiety symptoms was significantly higher in VR2 than in CG2 (χ²₁=8.3; P=.004). There was no significant difference between VR3 and CG3 in the response and remission rates of both depressive and anxiety symptoms (Table S2 in Multimedia Appendix 1).

This retrospective study examined whether adjunctive VR-based relaxation training was associated with acute changes in depressive and anxiety symptoms among hospitalized patients with depressive disorders and explored whether symptom improvement differed by session frequency. Consistent with the study objective, patients who received adjunctive VR-based relaxation training showed greater improvement in depressive and anxiety symptoms than patients who received usual care alone. The pattern of findings was more consistent for anxiety-related outcomes than for depression remission. Exploratory session-frequency analyses suggested that anxiety improvement may be observable across relatively short VR exposure, whereas depressive symptom outcomes did not show a stable or clearly interpretable session-frequency pattern.

These findings are broadly consistent with previous research suggesting that VR-supported relaxation and immersive restorative environments may be associated with reductions in anxiety, stress, and depressive symptoms [13-16,22-24]. Immersive VR relaxation may support symptom improvement through several complementary processes. Slow-paced breathing with immersive VR nature scenery has been used in heart rate variability biofeedback to facilitate relaxation-related physiological regulation [13]. Restorative VR environments have also been associated with emotional and cognitive recovery in individuals with mild-to-moderate anxiety and depression [22]. In addition, a randomized controlled trial in older women found greater improvements in depression- and anxiety-related outcomes after an immersive VR program incorporating relaxation and psychotherapeutic elements than after the comparison intervention [16]. A pilot randomized controlled trial also found that personalized VR scenarios combined with progressive muscle relaxation training were associated with greater engagement and reductions in anxiety than guided imagery [23]. These studies provide a plausible rationale for using immersive relaxation environments as an adjunctive component of care. However, the psychological and physiological mechanisms were not directly tested in this study.

The more consistent pattern observed for anxiety-related outcomes may be clinically relevant. Anxiety symptoms are common among patients with depressive disorders and are associated with greater symptom burden, functional impairment, suicide-related risk, and poorer outcomes [4-7]. One possible interpretation is that the relaxation-oriented content of the VR sessions may be particularly relevant when anxiety relief is an important clinical goal. However, this interpretation remains tentative because this study was retrospective and nonrandomized and did not isolate the effects of the immersive environment, relaxation content, or concurrent treatments.

The exploratory session-frequency analyses suggested that symptom outcomes, particularly anxiety-related outcomes, may vary across different levels of VR exposure. However, the pattern was not strictly linear. Because session frequency was not randomly assigned, it may have been related to hospitalization duration, clinical scheduling, patient availability, adherence, exposure to concurrent treatments, or other unmeasured factors. Accordingly, these findings should be regarded as hypothesis-generating associations rather than evidence of a causal dose-response relationship [25,26].

Multiple sessions may provide repeated opportunities to practice relaxation and engage with calming virtual environments. Previous studies have described and evaluated VR-supported progressive muscle relaxation protocols [23]. Nevertheless, the current design cannot determine whether increasing the number of sessions independently improves symptom outcomes. Future prospective studies with randomized or protocol-defined session schedules, standardized background treatments, adherence measures, and follow-up assessments are needed to clarify clinically useful session-frequency ranges and determine whether repeated VR relaxation sessions provide incremental benefits beyond usual inpatient care.

From a clinical perspective, adjunctive VR-based relaxation training may be considered a structured supplement to usual inpatient care rather than a replacement for established treatments. Previous studies have examined VR-supported relaxation and restorative VR environments across mental health and symptom-management contexts [13-16,22,23]. In this study, VR sessions were brief, structured, and delivered alongside routine treatment, providing preliminary real-world information about implementation in one psychiatric inpatient setting.

These findings may be relevant to hospitals interested in evaluating immersive relaxation as part of comprehensive care. However, this study did not formally assess feasibility, acceptability, staffing requirements, costs, or cost-effectiveness. These implementation outcomes, together with follow-up symptom outcomes and standardized background treatments, should be evaluated in prospective studies before broader clinical recommendations are made.

Several limitations should be considered. First, the retrospective, nonrandomized design prevents causal inference, and unmeasured confounding cannot be excluded, which is an important interpretive limitation of observational clinical research [25,26]. Second, the VR and control groups were naturally formed in routine care; therefore, patient preference, clinician judgment, symptom severity not captured in the records, and ward-level scheduling may have influenced group membership. Third, all patients received multiple concurrent treatments, including pharmacotherapy, rTMS, and biofeedback, and the medication type, dosage, adherence, and intensity of concomitant therapies could not be fully standardized. Fourth, the study was conducted in a single hospital and focused on acute inpatient outcomes at discharge, limiting generalizability and preventing conclusions about long-term durability. Fifth, physiological measures such as heart rate variability or cortisol were not available, so potential relaxation mechanisms could not be directly tested. Finally, the exploratory session-frequency analysis was not randomized and should be used only to generate hypotheses for future trials rather than to establish an optimal treatment frequency [25,26].

This retrospective study found that adjunctive VR-based relaxation training was associated with improvements in depressive and anxiety symptoms among inpatients with depressive disorders, with a more evident pattern observed for anxiety-related outcomes. Exploratory session-frequency analyses suggested potentially relevant patterns across different levels of VR exposure, but these findings should not be interpreted as evidence of a causal dose-response effect. The study is innovative in examining structured VR-based relaxation training as an adjunct to routine care in a real-world Chinese psychiatric inpatient setting, which differs from many previous VR studies conducted in noninpatient or experimental samples [14-16,18,22,23]. The findings contribute preliminary clinical evidence regarding the integration of immersive relaxation into comprehensive inpatient care, particularly when additional anxiety relief is desired. Because the study was retrospective and nonrandomized, prospective randomized controlled trials with standardized protocols, implementation measures, and longer follow-up periods are needed to confirm these associations and clarify clinically useful session-frequency ranges.