This study was conducted in 2023 at Stanford Hospital and Lucile Packard Children’s Hospital Stanford (Stanford University). Oculus Go (Meta, Inc) VR headsets displayed the application Pebbles the Penguin (Stanford Chariot Program). This third-person perspective application displays a head-controlled cartoon penguin sliding down a mountain collecting colorful objects for points. The application is intuitive, continues in perpetuity, and is successfully used to reduce perioperative and procedural anxiety [10,29,30].

The Stanford GME department includes over 1000 residents and fellows and over 100 training programs. The inclusion criteria were postgraduate medical trainees in any year of a Stanford residency or fellowship program. Exclusion criteria were individuals with nausea, motion sickness, seizure disorders, and active illness. With a significance level of α=.05, it was determined that a minimum sample size of 174 participants would be needed to test the study hypotheses.

The Stanford University Internal Review Board approved a waiver of consent. Trainees provided their informed consent prior to participation. All participants were informed of their ability to opt out of the study without consequence. No financial payments were provided for participation. No personal identifying information was collected as part of this study.

Trained research assistants (RAs), including coauthors BSKL, BK, MK, MR, and AG, conducted convenience sampling by recruiting volunteer participants in resident and fellow break rooms, outside cafeterias, and in patient care areas of the hospitals. Upon enrollment, RAs provided participants with descriptions of the clinical use of the VR experience as well as gameplay instructions. Participants completed a demographic survey before playing Pebbles the Penguin for 2 minutes. The application was initiated by the RAs to create a standardized experience for all participants. Following the VR experience, participants completed a survey adapted for health care professionals derived from previous VR TAM or UTAUT models (Multimedia Appendix 1) [21-23].

This TAM-UTAUT applied a validated hypothesis model exploring the intention to use VR [21,22]. Perceived ease of use, perceived usefulness, and perceived enjoyment were established in the earliest versions of the TAM and are widely applicable to technology acceptance, as Manis and Choi [21] revalidated in their VR hardware acceptance model in 2019. Perceptions regarding usefulness, ease of use, and enjoyment positively influence attitudes and behaviors toward purchasing and use [21]. These hypotheses stem from the idea that positive perceptions regarding new technologies promote decreased resistance to change [8,21]. Age is hypothesized to negatively influence perceptions surrounding technologies because age is often negatively associated with perceived future opportunities, limiting motivation for adoption [31]. Other demographic variables such as past use and price willing to pay are hypothesized to positively influence perceptions and intent to use. Curiosity, established as an inherent drive to seek novelty, is a positive indicator for perceived ease of use. Individuals with greater innate curiosity may engage with new technologies more out of interest rather than deprivation [21]. From the UTAUT model, social influence and facilitating conditions are both expected to have positive influential effects on perceptions, attitudes, and behavioral intentions (Figure 1) [23]. Social influence can result in a positive impact when adopting popular technologies that favor herd behavior and conformity over limited personal experiences [5,24]. Facilitating conditions are proposed to influence adoption through an organization’s infrastructure support of the proposed technology [5,24].

The primary outcome was to determine the predictors and validity of the TAM-UTAUT for the adoption of VR as an anxiolytic tool for patient care among a heterogeneous group of GME trainees. The secondary outcome explored the reliability of the model in the health care setting.

Demographic data collected prior to the intervention included age, sex, race, ethnicity, specialty, years of trainee experience, and prior VR use. The TAM-UTAUT surveyed aspects of perceived usefulness, perceived ease of use, perceived enjoyment, intention to use, intention to purchase, curiosity, attitude toward using, attitude toward purchasing, and price willing to pay. The TAM-UTAUT also included 2 socioenvironmental variables—social influence and facilitating conditions—to address previous limitations of conventional TAMs (Multimedia Appendix 2).

Price willing to pay was measured on a continuous scale from US $0 to US $1500. Attitudes toward using and purchasing were each measured using 5 questions graded on a 5-point sliding scale. Perceived usefulness and perceived ease of use were each measured using 5 questions graded on a 1‐5 Likert scale, ranging from 1=strongly disagree, 2=mostly disagree, 3=neither agree nor disagree, 4=mostly agree, and 5=strongly agree. The remaining variables were each measured using 4 questions graded on the same Likert scale (Multimedia Appendix 1). Study data were collected, encrypted, and managed securely using REDCap (Research Electronic Data Capture) tools hosted at Stanford University [32,33].

All elements were analyzed as continuous variables. To assess the predictive validity of the TAM-UTAUT, means and SDs were evaluated for each scale. Confirmatory factor analysis (CFA) was conducted using Mplus (version 8.6; Muthén & Muthén, 1998‐2022) to test the construct validity of different scales. Items over 0.7 were considered satisfactory [34,35]. Each item was tested for a normal distribution. For scales that were normally distributed, the maximum likelihood estimation method was used. For nonnormal scales, a robust maximum likelihood method was adopted to estimate the CFA model. The comparative fit index (CFI), root mean square error of approximation (RMSEA), Tucker-Lewis index (TLI), and standardized root mean square residual (SRMR) evaluated model fit. The CFI and TLI indices were accepted above 0.9 [36,37]. The RMSEA and SRMR below 0.08 indicated an acceptable fit [38,39]. The results were interpreted as the difference between each nonreference group.

To assess the secondary outcome of this TAM-UTAUT’s internal consistency and reliability, each scale was evaluated using Cronbach α and composite reliability. Composite reliability is similar to Cronbach α except that it does not assume each item to be equally weighted and accounts for actual factor loadings. Values were considered acceptable if greater than .6 [40].

In total, 202 of 1540 GME trainees participated (13.1% participation rate). After excluding responses with missing values, 198 participants (12.9% participation rate) were included in the final data analysis (Table 1). The ratio of female to male respondents was 1:1. The average age of all participants was 31.3 (SD 3.5) years, ranging from 23 to 45 years. Among the participants, 72 (36.4%) had no prior exposure to VR, 107 (54%) had 1‐5 experiences with VR, and 19 (9.6%) had 6 or more experiences with VR. The mean number of previous VR exposures was 2.7 (SD 4.5). In total, 44 unique specialties and fellowship subspecialties were represented among the participants. The most frequently represented specialties included pediatrics (n=32, 16.2%), internal medicine (n=27, 13.6%), and pathology (n=20, 10.1%; Multimedia Appendix 3).

For each composite domain, mean (SD) of perceived usefulness were 3.498 (0.883), perceived ease of use 3.911 (0.803), perceived enjoyment 4.349 (0.721), intention to use 3.460 (1.064), intention to purchase 3.446 (0.880), curiosity 3.404 (0.908), attitude toward using 3.979 (0.751), attitude toward purchasing 3.810 (0.801), social influence 2.910 (0.729), and facilitating conditions 3.684 (0.707). The mean price willing to pay was US $781.36 (SD US $375.50; Multimedia Appendix 4). All responses demonstrated normality, and the maximum likelihood method estimated the models.

Standardized factor loadings were greater than 0.7 for all items within the perceived usefulness, perceived ease of use, perceived enjoyment, intention to use, attitude toward using, and attitude toward purchasing domains. A total of 2 survey items within intention to purchase, 1 item within curiosity, 2 items within social influence, and 1 item with facilitating conditions were below 0.7 (Multimedia Appendix 5). All CFA models, which take into account all survey items within each scale, demonstrated good fit (CFI 0.984‐1; TLI 0.968‐1; RMSEA 0-0.137; SRMR 0-0.021; Multimedia Appendix 6).

Detailed estimates of the path coefficients were calculated (Figures 2 and 3 and Table 2). Most of the relationships between the perception, attitude, and intention domains were significant, except for the influence of perceived enjoyment on attitude toward using and perceived ease of use on intention to use. Perceived ease of use influenced perceived usefulness (P<.001), perceived enjoyment (P<.001), attitude toward using (P=.04), and attitude toward purchasing (P=.03). Perceived enjoyment influenced perceived usefulness (P<.001), attitude toward purchasing (P=.04), intention to purchase (P=.001), and intention to use (P<.001). Perceived usefulness influenced attitude toward using (P<.001), attitude toward purchasing (P<.001), and intention to purchase (P<.001; Figure 2 and Table 2).

Curiosity influenced social influence (P<.001) and facilitating conditions (P<.001) but not perceived ease of use (P=.44), as originally hypothesized (Figure 2 and Table 2). Price willing to pay influenced perceived usefulness (P=.005) and facilitating conditions (P=.002). There was no relationship between price willing to pay and perceived ease of use, perceived enjoyment, or social influence (Figure 2 and Table 2). Age and past use did not predict any outcomes, with the exception of a relationship between past use and facilitating conditions (P=.04; Figure 2 and Table 2). Social influence was a predictor of perceived usefulness (P<.001), attitude toward using (P=.001), intention to purchase (P=.002), and intention to use (P=.001). There was no relationship between social influence and perceived ease of use, perceived enjoyment, or attitude toward purchase. Facilitating conditions influenced perceived ease of use (P<.001), perceived enjoyment (P<.001), attitude toward purchasing (P<.001), and intention to use (P=.02) but did not influence perceived usefulness, attitude toward using, or intention to purchase (Figure 2 and Table 2).

All scales demonstrated acceptable reliability, as determined by Cronbach α and composite reliability values. Cronbach α results ranged from .753 to .962, and composite reliabilities ranged from .756 to .962 (Table 3).

The TAM-UTAUT predicted behavioral intentions associated with clinical VR use within this group of GME trainees. Predictors of intention to purchase include attitude toward purchasing, perceived enjoyment, perceived usefulness, and social influence, while predictors of intention to use include attitude toward using, perceived enjoyment, social influence, and facilitating conditions. Contrary to our hypothesis, perceived ease of use was not a predictor of intention to use. However, perceived ease of use predicted perceived enjoyment and attitude toward using; these latter 2 elements are effective mediators between perceived ease of use and intention to use.

The TAM-UTAUT also demonstrated that perceived enjoyment directly predicted future behavioral intentions to purchase and use VR technologies among GME trainees, but attitude toward using may not be a necessary mediator between perceived enjoyment and intention to use. Perceived enjoyment was a negative predictor of attitude toward purchase but a positive predictor of intention to purchase. It is possible that perceived usefulness and perceived ease of use constitute stronger justification to purchase a novel technology such as VR, whereas perceived enjoyment may not represent a compelling rationale to purchase a technology, even if behavioral intentions are predicted.

The VR-TAM that informed the tested model predicted that curiosity, age, past use, and price willing to pay all influenced perceived ease of use in the consumer market [21]. In contrast, this population of early career physicians indicated that curiosity, age, past use, and price willing to pay were not predictors of perceived ease of use. In addition, age and past use did not predict perceived usefulness. These results suggest that the previously hypothesized barriers to new technology adoption, such as older age or lack of prior exposure, do not affect GME trainees’ perceptions of ease of use or usefulness. The majority of trainees were 40 years or younger of age and had previous VR experience compared to the older mean age of faculty physicians. Given that GME trainees are more likely to be on the adoptive side of the “digital divide,” personal characteristics are less influential factors for predicting technology use [41].

This TAM-UTAUT demonstrated that price willing to pay was a predictor of perceived usefulness and facilitating conditions. These domains acted as mediators for predicting attitudes toward purchasing and using and intention to use, with implications for future adoption. VR is relatively affordable compared to other anxiolytics, especially as commercial equipment costs continue to decrease and health care VR applications expand [9,42-44]. As the difference between cost and price willing to pay continues to downtrend, there will be an increasing financial justification to apply VR in clinical settings from the perspective of GME trainees.

The UTAUT model elements, social influence and facilitating conditions, were hypothesized to predict behavioral intention to use [23]. The tested model indicated that social influence predicted perceived usefulness, attitude toward using, intention to purchase, and intention to use, while facilitating conditions predicted perceived ease of use, perceived enjoyment, attitude toward purchasing, and intention to use. These results further support these elements as key predictors for intention to use a novel technology [23]. However, in contrast to earlier studies where facilitating conditions was a stronger predictor of adoption than social influence, this TAM-UTAUT demonstrated social influence to be the stronger predictor [23,24]. In addition, this model indicates that social influence predicted intentions to both use and purchase, whereas facilitating conditions only predicted intention to use. One explanation for this incongruity may be that social influence and facilitating conditions are more context-dependent; different settings may engender different relationships between these extrinsic factors and technology adoption. Given that social influence and facilitating conditions are modifiable environmental factors, the positive directionality of their effects on behavioral intentions has strategic implications. Program directors and faculty responsible for educating GME trainees should foster learning environments that provide exposure to new technologies, relevant training and technical assistance, and social encouragement to drive VR adoption.

This TAM-UTAUT continues to expand the application of theoretical frameworks of technology adoption in health care and builds upon previous research by evaluating age, price willing to pay, and past use as continuous variables rather than categorical variables [24]. Additionally, with respect to the secondary aim, all TAM-UTAUT scales demonstrated good construct validity and reliability. This contributes to a growing body of evidence that TAMs and UTAUTs are appropriate modeling techniques to characterize technology adoption within clinical settings. This model also demonstrates the benefits of a hybrid model approach, as social influence and facilitating conditions from the UTAUT model were both predictors of intention to use. Future models can include other variables from the UTAUT model, such as performance expectancy and effort expectancy, as appropriate to the context and technology being studied.

There were several limitations to this investigation. First, trainees interested in VR may have been more likely to volunteer, leading to a selection bias. Second, participants were not involved in application setup or equipment troubleshooting, possibly leading to overestimated perceived ease of use. Third, factors such as perceived ease of use for patients and perceived usefulness for patients may affect attitudes to VR adoption. Future studies should explore patient perceptions as additional elements. Fourth, GME trainees were analyzed as a single group. A larger sample would have allowed for specialty subgroup analysis. Fifth, given the proximity to several notable technology company headquarters, participants may have stronger subjective norms toward VR compared to trainees in other regions. Similarly, the care settings in which this study was conducted already have established VR programs integrated into various aspects of patient care. While UTAUT scales like facilitating conditions attempt to characterize this context, this technology-driven culture may also have inflated other measures, such as perceived ease of use and perceived usefulness. To improve generalizability, additional studies should enroll trainees in other geographic settings or less technology-saturated regions.

This TAM-UTAUT demonstrated validity and reliability for predicting the behavioral intentions of GME trainees to use VR as a therapeutic anxiolytic tool in clinical practice. Perceptions of usefulness, ease of use, and enjoyment predicted intention to use VR. Age, past use, price willing to pay, and curiosity were less strong predictors of intention to use. Social influence and facilitating conditions also strongly predicted behavioral intentions, representing opportunities to advance VR adoption.