This study was a nationwide, population-based, and cross-sectional survey conducted in China between June and September 2024 as part of the Psychology and Behavior Investigation of Chinese Residents program.

The study was conducted across 33 provincial regions in China as part of a national psychological and behavioral health program. Data collection took place between June 23 and September 29, 2024, at the selected survey sites.

To improve national coverage of Chinese adults, we used a 5-stage geographic sampling framework across 33 provincial-level administrative regions. The sampling process combined multistage area selection with quota-based recruitment at the final stage. Specifically, 150 cities, 202 districts or counties, 390 townships or streets, and 800 communities or villages were selected through multiple geographic stages, after which participant recruitment within selected communities or villages was guided by demographic quota control. Poststratification raking weights were then applied to align the sample with the national age- and sex-specific distribution.

Of 38,793 questionnaires distributed, 38,424 (99.0%) were returned; after exclusions, 36,240 valid cases remained. A final 2-investigator audit yielded 35,861 records for analysis. Given this large national sample, the study was expected to provide stable estimates and reasonably narrow CIs for the primary descriptive and regression analyses. Only 4 of 139 predictors had missing data (range 1.54%‐15.30%); the outcome and all other predictors were complete. We therefore used complete-case analysis and did not perform multiple imputation.

Between June 23 and September 29, 2024, trained surveyors visited selected sites, confirmed eligibility, obtained e-consent, and administered an electronic questionnaire via Wenjuanxing. Responses were uploaded to a secure server and subjected to weekly supervisor review and automated quality control, checking for logic errors, duplicates, >20% missing data, or completion times <300 seconds.

Analyses comprised three stages: (1) descriptive summaries using poststratification raking weights, (2) multivariable modeling via 7-block hierarchical regression and elastic net validation, and (3) development of a classification and regression tree. Analyses used R (version 4.1.1; R Core Team), with 2-sided P<.05.

First, poststratification raking weights aligned the sample to 2020 census age-sex margins. We produced unweighted and weighted summaries (continuous: mean, SD; categorical: n [%]) and plotted weighted acceptance with 95% CIs by province, age-sex strata, single-year age (locally estimated scatterplot smoothing—smoothed by sex), and chronic condition. Province-level estimates were presented as descriptive summaries of geographic variation and were not intended to model province-level contextual effects.

Second, we fit survey-weighted hierarchical linear regression in 7 blocks—sociodemographics, life adversity and stress, personality, literacy and empowerment, lifestyle, physical health, and mental health—ordering blocks by presumed causal order to quantify independent effects while avoiding overadjustment. This hierarchical approach was used to examine whether each predictor domain explained additional variance in XR acceptance beyond domains entered earlier. For block k, we fitted a weighted linear model of acceptance on retained predictors from blocks 1 to k, applying Benjamini-Hochberg false discovery rate (FDR) correction within each block to address multiple testing, and carrying forward only predictors that remained significant after FDR correction. Continuous variables entered in original units; categorical variables were dummy-coded. Variance inflation factor >5 flagged multicollinearity. All candidate predictors were then weighted, z-standardized, and entered into a 10-fold cross-validated elastic net (α=.50) with optimal penalty λ_min selected by minimum mean squared error, which was used as a complementary data-driven approach because the study included a large number of potentially correlated predictors; this allowed us to assess the robustness of findings from the hierarchical regression and identify a more parsimonious set of predictors. We quantified agreement between elastic net and hierarchical selection via overlap statistics. For reporting, we prespecified the smallest effect size of interest for standardized β coefficients (β_std). Predictors with an absolute value |β_std| ≥0.03 were reported by name in the main text. Other FDR-significant predictors are detailed in main text tables and figures; nonsignificant predictors are tabulated in the supplement.

Third, to build a fast classification tool, we dichotomized acceptance at ≥ 80 (“acceptors”) versus <80 (“nonacceptors”) and randomly split the weighted sample 70/30 (outcome-stratified) into training and test sets. We constructed a Gini-based classification and regression tree on the training data, selecting the complexity parameter by the elbow method. For classification performance, acceptor status (score ≥80) was defined as the positive class; therefore, sensitivity represents the proportion of acceptors correctly classified and specificity represents the proportion of nonacceptors correctly classified. In the test set, we assessed discrimination using the survey-weighted receiver operating characteristic curve and its area under the curve with 1000 bootstrap CIs, and reported accuracy, sensitivity, specificity, positive predictive value, negative predictive value, and balanced accuracy at a 0.50 cutoff.

This study was approved by the Health-Economics and Policy Research (approval number NHC-HEPR202401) and the Shanghai Jiao Tong University Ethics Committee (approval number H20240237I). Written informed consent was obtained from all participants prior to their participation. To protect privacy and confidentiality, all study data were fully anonymized prior to analysis. Participants were not compensated for their participation. Finally, no individual participants or users can be identified in any of the images or figures included in this manuscript or its supplementary materials.

Among the 35,861 weighted respondents (Figure 1), mean age was 47.5 (SD 6.7) years; 50.4% (18,084/35,861) were male, 93.1% (33,381/35,861) were of Han ethnicity, 82.0% (29,409/35,861) had at least secondary education, and 72.5% (26,005/35,861) lived in urban areas. Complete sociodemographic, economic, residential, and health characteristics are shown in Table 1.

Overall mean weighted acceptance was 63.11 (95% CI 62.75‐63.46). Provincial means varied widely from 55.96 (95% CI 53.07-58.85) in Guizhou province to 72.34 (95% CI 66.28-78.39) in Jilin province. First-tier cities scored 63.01 (95% CI 60.92-65.10) in Beijing, 66.15 (95% CI 63.95-68.35) in Guangdong province, and 67.38 (95% CI 65.43-69.33) in Shanghai (Figure 2; Table S2 in Multimedia Appendix 1).

Acceptance varied by age, peaking at 18‐24 years for women (mean 68.51, 95% CI 68.05‐68.96) and 30‐34 years for men (mean 66.44, 95% CI 65.09‐67.79), with a general decline thereafter. Locally estimated scatterplot smoothing curves confirmed minimal sex differences (<3 points) (Figure 3A and B; Table S3 in Multimedia Appendix 1).

Individuals with any chronic condition had lower acceptance (mean 61.15, 95% CI 60.63‐61.67) than the overall sample (mean 63.11, 95% CI 62.75‐63.46). Scores ranged from rare diseases (mean 54.24, 95% CI 45.40‐63.07) to hyperlipidemia (mean 62.26, 95% CI 60.19‐64.33). Other common conditions showed relatively similar acceptance: obesity (mean 61.90, 95% CI 61.10‐62.70), hypertension (mean 60.31, 95% CI 59.23‐61.39), diabetes (mean 59.59, 95% CI 57.70‐61.48), anxiety (mean 59.32, 95% CI 58.40‐60.23), and depression (mean 59.37, 95% CI 58.58‐60.16) (Figure 3C; Table S4 in Multimedia Appendix 1).

A complete-case subsample (N=29,333) was used. The decision tree, trained on 70% (n=20,995) of the sample (Figure 5), identified 3 pathways for “acceptor” (score ≥80) and 4 pathways for “nonacceptor” (score <80). On the test set (n=8338), the tree achieved a modest weighted area under the curve of 0.61 (95% CI 0.60‐0.62), indicating modest discrimination (Figure S1 in Multimedia Appendix 1). Key performance metrics were accuracy=0.681, sensitivity=0.242 for acceptors, and specificity=0.903 for nonacceptors. Thus, the tree had limited ability to identify acceptors but performed substantially better in identifying likely nonacceptors.

The 3 “acceptor” profiles all required past digital health use and high personal existence (≥5). The largest (1922/20,955, 9%) also had high social support (Perceived Social Support Scale score ≥18) and low maladaptive daydreaming (5-item Maladaptive Daydreaming Short Form [MD-SF5] <25). The second (567/20,955, 3%) had low social support (<18) and hepatitis B vaccination. The smallest (191/20,955, 1%) had high social support, high MDS5 (≥25), and high youth socioeconomic status (≥5.5).

Conversely, 4 “nonacceptor” profiles were identified. The largest (13,010/20,955, 62% of sample) was defined simply by no past digital health use. The other 3 profiles all had past digital use but were defined by low personal existence (<5) (2446/20,955, 12%); or high personal existence, low social support (<18), and no hepatitis B vaccination (2164/20,955, 10%); or high personal existence, high social support, high MDS5 (≥25), and low youth socioeconomic status (<5.5) (655/20,955, 3%).

This national survey addressed three aims: (1) estimating population-level acceptance of immersive XR in health care, (2) examining multidomain correlates of acceptance, and (3) developing a parsimonious tool for identifying likely nonacceptors. In relation to these aims, our findings suggest that public acceptance of immersive XR in China is moderate but uneven, with variation across geographic, demographic, and health-related subgroups. Consistent with our hypothesis, acceptance was associated not only with demographic and disease-related characteristics but also with structural, psychosocial, digital, lifestyle, and health-related factors. To facilitate interpretation, we synthesize the seven prespecified analytic blocks into four broader dimensions shaping acceptance: (1) structural-environmental factors, encompassing sociodemographic conditions and life adversity; (2) trait-resilience factors, encompassing personality, self-efficacy, and mental health; (3) social-digital capital, encompassing health literacy, empowerment, social support, and prior digital health experience; and (4) behavioral health profiles, encompassing lifestyle and physical health factors. Finally, the classification tree further suggested that a small set of variables may help identify groups with lower acceptance, although its performance indicates that it should be interpreted as a pragmatic triage aid rather than an individual-level prediction model.

Overall acceptance was modest, and our findings may provide a useful benchmark for immersive XR pilot programs before wider-scale implementation. We found that acceptance was higher in affluent coastal cities and lower with increasing age. However, even among older adults, acceptance is still influenced by prior exposure, onboarding needs, sensory and cognitive demands, and the availability of caregiver or staff support [26]. This may underscore the need to co-design age-friendly XR systems and highlights the value of future subgroup analyses within older adult populations [27]. For older adults, simplified interfaces, structured orientation, self-paced interaction, and caregiver or staff support may be especially important [26,28]. We also observed heterogeneity by health status, suggesting that acceptance is not uniform across clinical groups. This may imply that immersive XR implementation should be condition-sensitive rather than one-size-fits-all [10].

Structural and environmental factors were important correlates of acceptance. Our findings indicate that higher socioeconomic status, education, perceived social standing, and better insurance coverage were associated with higher acceptance, whereas older age, lower youth socioeconomic conditions, and some family structure indicators were associated with lower acceptance. These patterns are consistent with the possibility that digital access, cumulative social advantage, and resource constraints may shape receptivity to new health care technologies [29-35]. Notably, the positive association of psychological adverse childhood experiences (abuse and neglect) and recent stressors with higher acceptance might reflect unmeasured confounding rather than a true effect of adversity; these findings should therefore be interpreted cautiously. In contrast, sexual or physical abuse and collective adversity were linked to lower acceptance, plausibly by undermining trust and self-efficacy [36]. From a structural-environmental perspective, equitable implementation of immersive XR requires attention to digital infrastructure, affordability, and shared access models, particularly in underserved settings [27].

Acceptance was also correlated with individual traits. We found that self-efficacy was a strong positive predictor, consistent with the possibility that perceived confidence and capability facilitate receptivity to immersive health tools [25]. Newly identified correlates included conscientiousness, narcissistic admiration, burnout, social media addiction, cyberchondria, attention-deficit/hyperactivity disorder traits, and depressive symptoms. The positive associations of narcissistic admiration, burnout, social media addiction, and cyberchondria are less straightforward and should be interpreted cautiously; they may reflect greater digital immersion or health information seeking rather than inherently greater readiness for sustained XR use [37,38]. Further research is needed to determine whether these factors predict actual XR uptake and sustained engagement, or whether they are markers of broader digital engagement and health information seeking. From an implementation perspective, the strong association with self-efficacy suggests that acceptance may depend partly on users’ confidence in trying, navigating, and troubleshooting unfamiliar digital tools. Users with lower confidence or fewer social resources may therefore require more intensive guidance and follow-up [28], such as confidence-building orientation, supervised first use, and simple step-by-step instructions.

Social and digital capital also showed strong associations. Prior digital health use and higher eHealth literacy were among the strongest predictors, consistent with established adoption models, such as the Unified Theory of Acceptance and Use of Technology [39,40]. Stronger family or neighbor ties and perceived social support likely act as motivational and practical resources (eg, encouragement and troubleshooting) [40]. This may be especially true in China’s collectivist context, where endorsement from close others functions as a powerful social norm [41]. Finally, dietary self-regulation may proxy a domain-general self-regulatory capacity, which is known to predict adherence to digital health tools [42]. From an implementation perspective, the associations with eHealth literacy, prior digital health use, and social support suggest that some users may need more than access to XR equipment. Implementation may require structured onboarding, hands-on demonstrations, plain-language instructions, troubleshooting support, and optional family or caregiver involvement [9]. For example, health systems could incorporate a brief digital health literacy or digital health readiness screener into intake workflows to identify low-readiness patients for staff-assisted XR orientation, troubleshooting, and follow-up support [43,44].

To our knowledge, few prior XR acceptance studies have examined lifestyle-related factors. Our findings therefore extend previous acceptance research by suggesting that behavioral health profiles may provide additional insight into population-level acceptance of immersive XR. Specifically, we found that more stable and sufficient sleep, arts exposure, evening chronotype, and COVID-19 vaccination were associated with higher acceptance, whereas smoking and salty diet were associated with lower acceptance. These patterns may reflect broader differences in health orientation, novelty receptivity, and familiarity with digital tools [45,46]. For instance, vaccination uptake is often associated with greater institutional trust and health engagement [47]. However, these associations should be interpreted as hypothesis-generating rather than causal, and future studies should examine whether lifestyle-related profiles predict actual XR uptake, adherence, and sustained use. In terms of implementation, strategies should account for users’ behavioral health profiles and focus on tailoring interventions to these specific lifestyle factors [48], such as incorporating more motivational framing into VR design for those who smoke.

The classification tree should be interpreted as a pragmatic implementation triage aid, rather than as an individual-level predictive classifier. According to standard interpretation of classification performance metrics [49], its low sensitivity indicates limited ability to identify individuals likely to accept immersive XR. However, its high specificity suggests better performance in identifying participants unlikely to accept immersive XR. Therefore, the tree may be more useful for resource allocation and targeted outreach. In this study, the tree used 6 implementation-relevant variables: past digital health intervention use, perceived meaning in life, perceived social support, hepatitis vaccination status, childhood socioeconomic status, and MDS5. These variables may help implementation teams identify subgroups less likely to accept immersive XR in health care and proactively allocate educational resources and intensive onboarding support to those who require the most assistance [50]. XR onboarding may include prebriefing, written and verbal controller instructions, a supervised first exposure, seated low-intensity familiarization, and real-time language or technical support [51].

This study has several limitations. First, its cross-sectional design captures acceptance at a single time point, precluding causal inference or trend analysis. Second, all measures were self-reported and subject to recall and social desirability biases. Third, as noted, the classification-tree model performed poorly on the test set, with very low sensitivity for identifying acceptors. Fourth, the vast regional disparities observed suggest that these individual-level predictors are moderated by powerful cultural, political, or infrastructural factors that were not fully captured in this model. Therefore, our China-based sample may limit generalizability to other cultural or health system contexts. Fifth, by focusing solely on user-side determinants, this study did not examine interactions with technology-side factors, which are likely to affect uptake and warrant further investigation. Sixth, although the survey used a nationwide multistage framework, participant recruitment at the final stage was guided by quota sampling rather than a fully probability-based design. In addition, poststratification weights were calibrated only to the age-sex distribution, so residual imbalance in other characteristics, such as province, urbanicity, or education, may remain. Moreover, clustering and stratification were not incorporated into variance estimation because the necessary design information was unavailable. Seventh, the geographic analysis was strictly descriptive and did not examine the ecological drivers underlying regional variation. Future research should build on these findings by using multilevel modeling to incorporate province-level covariates to distinguish individual-level determinants from broader contextual effects.

To our knowledge, this is among the first large-scale national studies to examine public acceptance of immersive health care technology using a multidomain framework in China. Unlike prior smaller or population-specific studies, it provides population-level evidence and suggests that receptivity is associated not only with demographic factors but also with structural, psychological, social, digital, behavioral, and health-related characteristics. This study contributes an equity-oriented and implementation-relevant perspective on which population groups may be more or less likely to accept immersive XR in health care. The findings suggest that equitable adoption may require more than technological availability alone; it may also depend on age-friendly design, condition-sensitive implementation, and targeted support for users with lower self-efficacy, lower digital readiness, or fewer social resources. In addition, the 6-item classification tree may offer a pragmatic starting point for identifying likely nonacceptors for outreach and support prioritization, but it requires external validation before practical use. Overall, our findings suggest that successful and equitable implementation of immersive XR is not only a technical issue but also a social, behavioral, and implementation challenge.