This study was approved by the Hungarian United Ethical Review Committee for Research in Psychology (reference number 2023-56). Participants provided informed consent at the beginning of the survey. Data were collected anonymously, and no incentives were offered.

A self-administered, anonymous online survey was developed, covering demographics, digital competencies, and attitudes toward e-learning, distance education, and VR/AR among orthopedic and trauma trainees, using validated scales (Attitude Toward E-Learning Scale [37], Distance Education Perception Scale [38], and VR/AR attitude items [39]) on a 5-point Likert scale (1=strongly disagree and 5=strongly agree). We did not alter the individual items of the validated scales. Instead, we selected relevant question groups from the original lengthy questionnaire without modifying their wording or content. This approach preserved the integrity and construct validity of the instrument. Expert review, cognitive pretesting, and pilot testing with 10 trainees (qualitative feedback confirmed clarity and comprehensibility) ensured validity and clarity [40]. All items of the survey were presented in English to ensure consistency across countries and to avoid potential discrepancies introduced by translation.

The e-learning and VR/AR-based learning questionnaires demonstrated reliability in our cohort with Cronbach α values of 0.837 and 0.931, respectively. The Distance Education Perception Scale questionnaire consists of 4 subscales: student’s perception, equipment facility, facility and support of the institution, and time management. The first 3 subscales demonstrated high degrees of reliability, ranging from 0.813 to 0.891; however, the time management subscale showed a low level of reliability (α=0.344) and therefore was excluded from the calculations.

Demographics included age, gender, training location (country and city), workplace type (university, regional, city, and rural hospitals), and previous experience with distance education, e-learning, medical simulation, and VR/AR simulation. Digital competence questions covered basic and advanced tools (eg, expertise ranging from basic competencies such as installing a program to advanced skills such as programming or network setup) and VR/AR familiarity (Multimedia Appendix 1).

The structure of the attitudinal scales and analytical framework was informed by the TAM and UTAUT frameworks. In this context, perceived usefulness and ease of use (TAM constructs) were operationalized through items assessing engagement and perceived effectiveness of e-learning, distance education, and VR/AR tools. UTAUT dimensions were represented by social influence and facilitating conditions (eg, equipment availability and workplace resources). These theoretical constructs guided variable grouping, factor structure validation, and subsequent analyses exploring how contextual and personal factors (including GDP, workplace type, and digital competence) influence acceptance and attitudes toward educational technologies.

Due to the lack of comprehensive registries enumerating orthopedic residents by country, a precise estimation of the eligible population and a predefined sample size were not feasible. Therefore, this study was conducted as a time-bounded, voluntary cross-sectional survey. Participants were recruited via Federation of Orthopaedic and Trauma Trainees in Europe newsletters and mailing lists (estimated reach: approximately 1000-1200 trainees across 27 countries), plus targeted emails and social media posts through national orthopedic resident societies. No incentives offered, reducing response bias from economic and regional disparities. A 2-week data collection period with a 1-week reminder yielded 235 responses (217 analyzed after exclusions). In the absence of centralized registries, an exact response rate was unavailable; however, diverse demographics (29 countries and balanced GDP and workplace types) support generalizability.

The survey was hosted on Google Forms, ensuring ease of access and data security. GDP per capita data were obtained from the International Monetary Fund (2022) and categorized into low (<US $29,000), medium (US $29,000-US $40,000), and high (>US $40,000) GDP groups (tertiles of 29 respondent countries), aligning with World Bank thresholds (the European Union average is US $41,806).

To evaluate trainees’ attitudes toward e-learning and distance education, mean scores for survey items were compared with the neutral midpoint (3.0) using Wilcoxon signed-rank tests due to nonnormal distributions indicated by Shapiro-Wilk tests (all P<.01). Effect sizes were reported as rank-biserial correlations (r_bs), with values of 0.10, 0.30, and 0.50 interpreted as small, medium, and large effects, respectively. To examine whether attitudes differed depending on the type of workplace and the economic context of the country, separate 2-way ANOVAs were run with workplace type (university vs nonuniversity clinics) and GDP per capita (low, medium, or high) as between-subject factors, and attitude variables (ie, scores on subscales of the attitude questionnaires) as outcomes. Significant interactions were followed up by simple effect analyses. All post hoc analyses were adjusted for multiple testing using the Bonferroni correction.

A priori power analyses using G*Power (version 3.1.9.7 [41]) indicated that the minimum sample size required for ANOVA was 155 (α=5%; power=80%), and thus the final sample of 217 (see below) had adequate statistical power to detect main effects as well as the interactions examined. Differences in experience levels and digital competencies across domains were tested using the Friedman test. Significant results were followed up by Wilcoxon tests with Bonferroni correction. Spearman partial correlations were used to assess relationships among the measures while controlling for digital competence and experience with digital learning. When an attitude variable was correlated with either competence or experience, the other served as the control variable. No control variables were applied for the correlation between competence and experience. Spearman correlations were chosen because Shapiro-Wilk tests indicated nonnormal distributions for all variables (all P<.05), except for the student’s perception subscale.

We used k-means cluster analysis to identify distinct subgroups of participants with similar attitudinal profiles toward digital learning. The optimal cluster count was determined using the elbow and silhouette methods (Multimedia Appendix 2). Cluster-wise stability was assessed using the mean Jaccard similarity value obtained via bootstrapping (n=1000). Mean Jaccard similarity values >0.75 and >0.85 were interpreted as indicators of stable or very stable clusters, respectively [42].

After clustering, the dataset was aggregated to calculate mean values for each cluster, providing insight into the dominant attitudes and experiences regarding digital learning across groups. Multinomial logistic regression was conducted to examine how background variables (eg, digital competence, experience, and demographics) predicted cluster membership and to better understand differences between subgroups.

To identify the underlying factor structure and consolidate conceptually overlapping constructs into a clearer and more parsimonious model, exploratory factor analysis (EFA) was conducted on the complete set of items from all 3 attitude questionnaires. Because items were ordinal, a polychoric correlation matrix was computed, and robust maximum-likelihood extraction with oblique (Promax) rotation was used. Factor retention was guided by (1) parallel analysis and scree plot inspection and (2) interpretability; primary loadings ≥0.40, cross-loadings <0.30, and communalities ≥0.30 were used as criteria for item retention. The full EFA diagnostics, including the parallel-analysis scree plot, communalities, and item loadings, are provided in Multimedia Appendix 3.

To confirm the factor structure identified by EFA, confirmatory factor analysis was conducted using robust maximum likelihood estimation (estimator=“MLR”) to account for nonnormality. Model adequacy was evaluated using standard fit indices and recommended thresholds: comparative fit index (CFI) and Tucker-Lewis index (TLI) ≥0.90 (acceptable) or ≥0.95 (excellent), root mean square error of approximation (RMSEA) ≤0.08 (acceptable) or ≤0.06 (good), and standardized root mean square residual (SRMR) ≤0.08. Standardized factor loadings ≥0.50 were interpreted as adequate. Model refinement involved reviewing modification indices (MI>10) and applying theoretically justified correlations between residuals where appropriate.

Statistical significance was set at P<.05 (2-tailed) for all analyses unless Bonferroni-corrected otherwise. All analyses were conducted using SPSS (version 28; IBM Corp) and R (version 4.4.0; R Foundation for Statistical Computing). R packages used are listed in Multimedia Appendix 4.

In total, 235 responses from 29 countries (geographical distribution shown in Multimedia Appendix 5) were received, of which 217 were included in the analysis after excluding responses with beginner-level English proficiency (n=17) or incomplete data (n=1) to minimize potential bias arising from language-related comprehension issues. Excluded respondents did not significantly differ from included respondents in terms of basic demographics (age: U=1482.5; P=.17; r_bs=0.2 and gender: χ²₁=0.02; P=.88); however, trainees working at university clinics were underrepresented among the excluded participants (χ²₁=5.3, P=.02). A detailed description is provided in Table 1. The dataset is available elsewhere [43].

Participants reported their experience across 4 domains: distance education, e-learning, medical simulation, and VR/AR. The Friedman test indicated significant differences in self-reported experience across domains (χ²₃=285.0; P<.001). Post hoc Wilcoxon signed-rank tests indicated that participants’ experience with e-learning was significantly higher than their experience with distance education (Wilcoxon Z=3.68; P=.001; r_bs=0.44), medical simulation (Z=9.30; P<.001; r_bs=0.88), and VR/AR (Z=11.12; P<.001; r_bs=0.97). Experience with distance education was also significantly higher than experience with medical simulation (Z=6.62; P<.001; r_bs=0.61) and VR/AR (Z=9.88; P<.001; r_bs=0.86). Finally, experience with medical simulation was significantly higher than that with VR/AR (Z=8.34; P<.001; r_bs=0.88).

The Friedman test showed significant differences in self-reported digital competence (χ²₄=542.9; P<.001). Post hoc analyses revealed that competence in basic tools and online platforms did not differ significantly (Z=2.76; P=.06; r_bs=0.31), but both were rated significantly higher than each of the other domains of competence (all z>10.51, all P<.001, and all effect sizes large per Cohen r_bs > 0.97). Nearly 80% of participants reported “advanced” or “expert” competence in both domains. Competence in digital problem-solving was rated significantly higher than both competence in advanced technologies (Z=5.55; P<.001; r_bs=0.55) and VR/AR (Z=9.60; P<.001; r_bs=0.88). Finally, self-reported competence in advanced technologies was significantly higher than that of VR/AR (Z=7.20; P<.001; r_bs=0.70). For digital problem-solving, advanced technologies, and VR/AR, 37.79% (82/217), 24.88% (54/217), and 11.06% (24/217) reported “advanced” or “expert” level of competence, respectively. Multimedia Appendix 6 provides a more detailed description of within-domain distributions.

Attitudes toward e-Learning and distance education varied among participants. Distance education recorded a mean score of 3.27 (SD 0.63) and a median of 3.18 (IQR 2.95-3.55), indicating moderately positive attitudes, with consistent responses (SD 0.63; skewness 0.37). Statistical analysis revealed significant positivity in the attitudes toward distance education (Wilcoxon Z=6.10; P<.001; r_bs=0.49; mean 3.27 (SD 0.63); 95% CI 3.19-3.36). e-Learning followed with a mean of 3.43 (SD 0.69) and a median of 3.44 (IQR 3.11-3.89), exhibiting slightly more favorable attitudes with moderate variability (SD 0.69). The Wilcoxon test confirmed the positivity of these attitudes (Z=8.03; P<.001; r_bs=0.64; mean 3.43 (SD 0.69); 95% CI 3.33-3.52). VR/AR received the highest mean score of 4.07 (SD 0.88) and a mode of 5, reflecting the most positive attitudes, albeit with greater variability (SD 0.88, skewness –1.17), and the analysis significantly supported these positive perceptions (Z=10.94; P<.001; r_bs=0.89; mean 4.07 (SD 0.88); 95% CI 3.96-4.19). Negative skewness in VR/AR attitudes suggests a ceiling effect, potentially reflecting aspirational responding. Overall, VR/AR elicited the strongest positive responses, although its variability suggested mixed familiarity or application, whereas distance education and e-learning garnered more tempered but consistent favorability (Figure 1).

K-means clustering based on responses to the attitude scales, optimized for internal coherence and separation, suggested 3 clusters that provided an optimal balance of granularity and distinctiveness. The average silhouette score was 0.27 for the 3-cluster solution, whereas it was 0.26 and 0.21 for the 2- and 4-cluster solutions, respectively (Multimedia Appendix 2 provides a detailed description of cluster number optimization). The clusters were tentatively named “enthusiastic” (47/217, 21.66%), “supportive” (129/217, 59.45%), and “hesitant” (41/217, 18.89%) (Figure 2). All 3 clusters showed high levels of stability according to mean Jaccard similarities (0.85, 0.86, and 0.89), confirming the robustness of the 3-cluster solution (mean Jaccard similarities for the 2-cluster solution: 0.87 and 0.88; for the 4-cluster solution: 0.57, 0.30, 0.57, and 0.74). The attitudes were measured using scales adapted from validated sources, including the Attitude Toward E-Learning Scale, the Distance Education Perception Scale, and items related to perceptions of VR/AR applications in learning. Enthusiastic respondents showed the highest endorsements across all attitude scales. Supportive individuals expressed favorable attitudes but with moderate enthusiasm compared with the enthusiastic cluster. Hesitant respondents exhibited the lowest agreement with positive statements regarding all educational technology aspects (Figure 2). Significant differences among clusters were confirmed by 1-way ANOVAs across all attitude dimensions, each showing large effect sizes (η_p²>0.14): distance education student’s perception, F_2,214=179.85; P<.001; η_p²=0.63; distance education equipment facility, F_2,214=132.94; P<.001; η_p²=0.55; distance education facility and support, F_2,214=52.27; P<.001; η_p²=0.33; e-learning engagement, F_2,214=105.04; P<.001; η_p²=0.50; and VR/AR engagement, F_2,214=50.77; P<.001; η_p²=0.32.

The results of multinomial logistic regression analysis are summarized in Table 2. Global model fit indices indicated that the multinomial regression model provided a significantly better fit than the null model (χ²₁₄=45.27; P<.001) with a McFadden pseudo-R² of 0.11, suggesting a modest yet meaningful amount of variance in cluster membership was accounted for by the predictors. The findings highlighted digital competence as a significant predictor in differentiating between the enthusiastic and hesitant clusters (estimate=0.22; SE=0.07; Z=3.22; P=.001; and odds ratio [OR] 1.25, 95% CI 1.09-1.43) and between the supportive and hesitant clusters (estimate=0.18; SE=0.06; Z=3.07; P=.002; OR 1.19, 95% CI 1.07-1.33). This indicates that higher levels of digital competence are associated with increased odds of belonging to the enthusiastic or supportive clusters rather than the hesitant cluster, suggesting that perceived capability with digital technologies plays a key role in shaping trainees’ attitudinal profiles. Economic conditions, indicated by GDP, also played a significant role in distinguishing between the enthusiastic and hesitant groups (estimate=–0.03; SE=0.01; Z=–2.44; P=.02; OR 0.97, 95% CI 0.95-0.99) Similarly, GDP was a significant predictor of cluster membership when the enthusiastic and supportive groups were compared (estimate=–0.03; SE=0.01; Z=–3.02; P=.003; OR 0.97, 95% CI 0.95-0.99), suggesting that trainees from countries with lower GDP were more likely to belong to the enthusiastic cluster. Gender, age, training year, workplace type (university vs nonuniversity clinic), and prior digital experience were not significant predictors (all P>.05). Given that demographic and training characteristics, as well as overall experience with various forms of digital learning, did not strongly differentiate clusters, attitudinal patterns may be largely independent of these background factors.

The results of the ANOVAs investigating the effects of GDP and workplace on attitudes are depicted in Figure 3 and Multimedia Appendix 7. The main effect of workplace reached significance for the student’s perception subscale (F_1,211=6.95; P=.009; η_p²=0.03). Trainees from nonuniversity clinics had significantly more positive attitudes toward distance learning than trainees from university clinics.

The analysis of the equipment facility subscale yielded a significant GDP main effect (F_2,211=5.02; P=.007; η_p²=0.05). Corrected post hoc tests showed that the scores of trainees from low-GDP countries were significantly higher than those of trainees from high-GDP countries (mean difference –2.07, 95% CI –3.64 to –0.49; t₂₁₁=–3.10; 2-tailed P=.007); Cohen d=–0.52). Similarly, there was a significant GDP main effect for facility and support of the institution (F_2,211=7.84; P<.001; η_p²=0.07), as well as a significant Workplace × GDP interaction (F_2,211=5.16; P=.007; η_p²=0.05). Simple effects analyses revealed that the GDP main effect was only significant for trainees from nonuniversity clinics (F_2,211=13.24; P<.001; η_p²=0.11). Within this group, scores of trainees from low-GDP countries were significantly higher than those of trainees from medium-GDP (mean difference 2.17, 95% CI 0.33 to 4.01; t₂₁₁=3.39; 2-tailed P=.01); Cohen d=–1.11) and high-GDP countries (mean difference –2.73, 95% CI –4.35 to –1.11; t₂₁₁=–4.85; 2-tailed P<.001; Cohen d=–1.11). The analyses also showed a significant workplace effect within the group of trainees from high-GDP countries (F_1,211=6.03; P=.02; η_p²=0.03): scores were higher for trainees from university clinics than for other trainees.

The analysis of attitudes toward e-learning also yielded a significant GDP main effect (F_2,211=4.59; P=.01; η_p²=0.04): trainees from low-GDP countries showed significantly more positive attitude toward e-learning than trainees from high-GDP countries (mean difference –3.62, 95% CI –6.45 to –0.80; t₂₁₁=–3.03; 2-tailed P=.008; Cohen d=–0.51).

Finally, a significant GDP main effect was found for experience with technologies (F_2,211=3.65; P=.03; η_p²=0.03): trainees from high-GDP countries had significantly more experience than trainees from low-GDP countries (mean difference 1.48, 95% CI 0.14 to 2.82; t₂₁₁=2.60; 2-tailed P=.03; Cohen d=0.44). No further significant main effects or interactions were found.

Correlation analyses (Figure 4) showed digital competence was strongly associated with favorable attitudes across all learning modalities, especially in low- and medium-GDP countries (Multimedia Appendix 8). Prior experience also correlated with more positive attitudes, highlighting the role of familiarity. Positive views of one modality often extended to others, reflecting broad openness to digital education, consistent with theoretical frameworks emphasizing usability and prior exposure.

The Mardia test showed significant multivariate skewness (b_1p=307.11; P<.001) and kurtosis (b_2p=1379.52; P<.001), indicating violation of multivariate normality. Consequently, a polychoric correlation matrix was used in all subsequent analyses. Factorability was supported by the Bartlett test of sphericity (χ²₅₉₅=7216.854; P<.001), the Kaiser-Meyer-Olkin index =0.78, and the determinant of the correlation matrix (3.97×10^–16). Interitem correlations were generally moderate (median rₛ=0.368, range 0-0.925). Parallel analysis indicated 6 factors; however, conceptual overlap among related dimensions motivated a theory-driven consolidation (eg, distance education engagement and conditions → distance education perception), yielding a more parsimonious 4-factor solution. In this reduced model, after rotation, the 4 factors explained 20.8%, 14.2%, 11.8%, and 10.7% of the variance (total=57.5%). No Heywood cases were identified. More information is available in Multimedia Appendix 3.

For model refinement, items with strong primary loadings and adequate communalities were retained, while theoretical interpretability was also considered. Further, 8 items (e-learning items 6 and 10; distance education items 7, 10, 12, and 20-22) were removed because of low communalities. Distance education item 8 was excluded because of a low loading (0.38) and cross-loading. e-Learning items 7-9 and 11 were excluded because they loaded on a theoretically implausible factor (distance education perception rather than e-learning).

The resulting model comprised 4 latent constructs: distance education perception, equipment facility perception, attitude toward VR/AR, and attitude toward e-learning. This structure was evaluated using confirmatory factor analysis. The initial model showed acceptable fit (χ²₂₀₃=370.57; CFI=0.938; TLI=0.93; RMSEA=0.062; SRMR=0.052), although modification indices indicated potential improvement. Accordingly, a residual covariance was specified between distance education items 13 and 14 (MI=33.49), resulting in improved fit (ΔCFI=0.01; ΔTLI=0.011; ΔRMSEA=–0.005; ΔSRMR=0.002). Subsequent indices suggested an additional residual covariance between e-learning items 1 and 2 (MI=24.52). In both cases, the items were semantically similar and reflected closely related constructs, justifying correlated residuals (Multimedia Appendix 3).

The final model (Figure 5) demonstrated good fit (χ²₂₀₁=323.203; CFI=0.955; TLI=0.948; RMSEA=0.053; SRMR=0.053). Further modifications were not pursued despite several MIs>10 to avoid overfitting. The model showed moderate to high correlations among the constructs, reflecting their conceptual interdependencies within the domain of e-learning environments. In more detail, interfactor correlations demonstrated meaningful relationships, with distance education perception positively associated with equipment facility perception, attitude toward VR/AR, and attitude toward e-learning, and equipment facility perception significantly correlated with both attitude toward VR/AR and attitude toward e-learning, reflecting theoretical interdependencies across educational dimensions. Findings from this more parsimonious model were consistent with the partial correlation patterns reported earlier, supporting the stability of the factor structure.

Digital tools have advanced medical education, particularly in procedural fields such as orthopedics, where tactile skills challenge remote learning. VR/AR simulations enabling safe rehearsal (eg, arthroscopy knee simulators show operating room transferability, and hip arthroscopy VR matches benchtop training) [44-46]. This study highlights the importance of digital competence and familiarity in shaping attitudes, aligning with findings from other disciplines such as nursing and parallel surgical evidence from veterinary medicine [47-50].

Our finding that VR/AR elicited the most positive attitudes (mean 4.07) despite limited exposure aligns with recent umbrella reviews showing VR’s growing dominance in surgical training and procedural skill acquisition [34]. The findings of Sommer et al [51] regarding VR’s reduction of errors and improvement of completion times in surgical simulations (60 students; 2-arm randomized trial) are consistent with the trainee optimism reported in this study. This is corroborated by the findings from Pettinelli et al [13], whose systematic review with a total sample size of 148 residents or students affirmed that VR is effective for learning specific arthroplasty techniques (no difference between VR-trained and control groups; P=.56; I²=85%), further supporting the notion that residents are open to integrating this technology into their training repertoire. However, similar to the narrative review by Ivanov et al [52], we note that statistical significance in performance outcomes remains inconsistent, underscoring the need for larger longitudinal studies.

Structural equation modeling (SEM) revealed factors that influence clinical cognition, health behavior, or the acceptance of digital learning modalities [53-55]. Our analysis demonstrates that engagement with distance education shows a strong association with acceptance, suggesting a bidirectional influence between perceptions and intent in shaping residents’ comfort with e-learning platforms and VR/AR tools. This finding underscores the importance of fostering engagement through distance education to build familiarity with advanced digital tools. SEM associations (eg, distance education acceptance) are correlational, precluding causality; bidirectionality is likely (eg, comfort shapes perceptions and vice versa). Longitudinal studies are needed to test directions and moderators such as competence.

Our SEM findings, which link digital confidence to VR/AR acceptance, support the “blended digital comfort” strategies described by Cabero-Almenara et al [56] (online questionnaire involving 17,301 Latin American students). For example, pairing VR modules with hands-on workshops could reduce effort expectancy (a UTAUT construct), as observed in proof-of-concept mixed-reality ward rounds by Bala et al [57] involving 11 students, in which 82% of trainees reported improved interaction. Tool proficiency is key for acceptance and integration.

The interplay between e-learning dimensions, such as blended digital comfort’s influence on e-learning attitudes, emphasizes the need to address broader digital literacy and confidence. Consistent with prior research (multinational survey involving 223 students), positive attitudes were more pronounced in settings with lower access to conventional resources, such as noncentral hospitals and low-GDP countries, where distance and e-learning provide critical educational support [58]. This is particularly important for designing interventions that integrate multiple technologies effectively.

Beyond economic resources, sociocultural factors also shape regional differences in technology adoption; national cultural dimensions (uncertainty avoidance, individualism/collectivism, and power distance) moderate links among ease of use, perceived usefulness, and intention to adopt [59]. Trainees from lower-GDP countries showed more positive attitudes, possibly reflecting adaptive innovation cultures rather than mere scarcity, whereas high-GDP regions may have more rigid educational norms and accreditation structures that slow adoption (Multimedia Appendix 9 presents the Hofstede dimensions of countries included in this study) [60]. Effective VR-based orthopedic education therefore requires strategies tailored to both economic capacity and cultural context: in lower-income settings, low-cost standalone VR, local VR “champions,” and team-based modules can support uptake, whereas in higher-income settings, integration into accredited curricula, assessment frameworks, and faculty development programs helps overcome inertia. Across contexts, culturally responsive leadership, clear standards, and evaluation using frameworks such as the Consolidated Framework for Implementation Research and reach, effectiveness, adoption, implementation, and maintenance (RE-AIM) are needed to sustain engagement, equity, and long-term integration [61,62].

TAM and UTAUT provide a useful interpretive lens for our attitudinal patterns, linking them to concepts such as perceived usefulness, ease of use, and contextual enablers. In our model, attitude toward e-learning and distance education perception map onto perceived usefulness and performance expectancy, with digital competence and experience reflecting effort expectancy, and workplace type and GDP representing facilitating conditions and social influence. Digital competence showed positive associations with attitudes toward e-learning, distance education, and VR/AR—patterns consistent with TAM and UTAUT concepts, while institutional effects align with UTAUT’s emphasis on environmental support and broader trends in which novel tools elicit optimism despite limited familiarity, although we did not estimate a full structural model (eg, with R² for acceptance) [63,64]. Perceived ease of use, digital confidence, and facilitating conditions—access to hardware, connectivity, and institutional backing—are therefore critical and can be strengthened through user-friendly tools, targeted training, institutional support, subsidized technology, and partnerships with providers to promote equitable access [65].

The findings suggest a tiered approach to educational design. At the core (immediate) level, digital competence training should be embedded within curricula through targeted workshops and organized testing campaigns to enhance familiarity and effort expectancy. At the supportive (medium-term) level, access to digital tools should be subsidized to address disparities associated with lower GDP and strengthen facilitating conditions. Collaborative programs and partnerships between institutions in lower- and higher-GDP countries should also be promoted to support resident training, faculty development, and the sharing of educational resources and programs. At the advanced (long-term) level, VR/AR modules and e-learning platforms can be integrated into hybrid educational models, with content and implementation strategies tailored to local cultural and contextual needs.

Limitations include the unknown size of the trainee population, uneven geographic distribution, inability to characterize nonrespondents, and reliance on self-reported data. Due to the absence of standardized, publicly available databases enumerating orthopedic resident characteristics across European countries, it is not possible to ascertain whether the proportions observed in our sample reflect the true population. Uneven geographic representation limits the ability to extrapolate findings uniformly across all European contexts. Exclusion bias might have affected the results because of the higher exclusion rate among trainees working in nonuniversity clinics (10 of 18 exclusions were due to beginner-level English proficiency). The reliance on self-reported attitude data may have introduced social desirability bias, potentially inflating positive responses toward digital tools, as evidenced by the VR ceiling. No social desirability scale was used. Nonetheless, the findings are likely generalizable because diverse demographics (29 countries and balanced GDP and workplace types) support generalizability, as do global similarities in medical training and digital education challenges. Future multimethod validation (eg, objective logs) is needed; longitudinal studies are needed to test causality.

Further studies should assess the effects of sustained digital platform exposure on attitudes and outcomes over time, using mixed methods designs. Qualitative components should examine VR/AR as innovative tools for orthopedic and trauma trainee education. Phenomenological studies can explore trainees’ lived experiences with VR/AR tools, while institutional case studies can reveal local implementation mechanisms. Quantitative designs (exploratory, experimental, and longitudinal) should assess skills acquisition via the objective structured assessment of technical skills and the objective structured clinical examination across residency. Translational research should examine the impact of digital skills on patient outcomes and their correlations with attitudes, motivation, and retention. All studies should be grounded in self-directed learning, cognitive load, and deliberate practice theories for design and interpretation.

Based on the findings, future research should focus on 3 main directions:

To operationalize these research directions and secure funding, we propose a scalable 4-step approach (Textbox 1) aligned with European Union schemes (Erasmus+ and Horizon Europe) and frameworks (Consolidated Framework for Implementation Research for determinants; RE-AIM, and dynamic sustainability framework for sustainability). Each includes milestones, risks, and mitigations (more detailed suggestions are provided in Multimedia Appendix 10).

Analysis of this international multicountry survey highlights that foundational engagement and confidence are pivotal for promoting advanced adoption of VR and AR technologies and optimizing the use of digital tools. Programs should prioritize an interactive, well-supported, and blended approach to accommodate varying levels of digital familiarity and preferences.

The application of TAM and UTAUT in this study supports their utility in understanding technology adoption in medical education. By addressing perceived usefulness, ease of use, social influence, and facilitating conditions, institutions can design strategies to enhance digital tool integration. In this context, tailored implementation strategies should be developed to optimize digital education adoption. These theoretical advancements inform future research and guide pragmatic policy design aimed at integrating digital tools into postgraduate orthopedic training.

Our findings indicate that digital competence strongly influences acceptance of technology-enhanced education among orthopedic trainees. Therefore, implementation strategies should prioritize competence-building initiatives, such as structured digital literacy programs, as foundational steps. Following competency development, scalable interventions such as VR module integration can be implemented more effectively.

This study advances the understanding of factors influencing digital learning adoption in education, providing a foundation for designing effective, equitable, and innovative learning interventions aligned with trainees’ needs and preferences.

Orthopedic training must advance beyond pandemic-driven digital adoption. While e-learning remains a foundational element, the untapped potential of VR and AR technologies necessitates targeted investments in infrastructure, competency-based curricula, and cross-institutional collaborations. Consistent with the TAM and UTAUT, addressing socioeconomic inequities through policies (such as establishing European Union–funded VR hubs in low-GDP regions) can translate trainee enthusiasm into measurable clinical proficiency.

Beyond educational equity, improving access to high-quality digital and simulation-based training is expected to have downstream clinical relevance, as randomized-trial evidence suggests that simulation training can improve patient-related outcomes in surgery. While our findings are European, the proposed competence-first and equity-oriented implementation pathway could be adapted through partnerships with institutions in the Global South, aligned with World Health Organization priorities for equitable digital health capacity building and the broader global agenda for strengthening surgical systems and workforce development [66,67].