This study used a randomized controlled trial (RCT) with a posttest-only design.

Registered nurses working in the emergency department of Samsung Medical Center in Seoul, Republic of Korea, were recruited between July 17 and July 20, 2023. The required sample size of 42 participants (21 per group) was determined a priori using G*Power 3.1, based on an effect size (Cohen d) of 0.9, a significance level (α) of .05, and a statistical power (1−β) of .80.

Eligibility criteria included (1) voluntary participation with informed consent and (2) no prior experience operating the Level-1 device. Participants were randomly assigned to the experimental or control group through concealed allocation: each participant drew a slip of paper from an opaque box indicating group assignment in a 1:1 ratio. Both groups completed training and evaluation at the same location with 1:1 supervision.

Data were analyzed using SPSS (version 18.0; IBM Corp). Normality was assessed using the Shapiro-Wilk test together with skewness values. Most continuous outcomes were nonnormally distributed, and the timed and count variables were markedly right-skewed (skewness ranging from +0.50 to +6.09). Accordingly, between-group differences in clinical competency, self-efficacy, and educational satisfaction were compared using the Mann-Whitney U test, and the corresponding data are summarized as medians with IQRs. Baseline homogeneity was assessed using the independent-samples t test. Convergent validity between the 2 accuracy instruments was evaluated using the Spearman rank correlation. All tests were 2-tailed, and statistical significance was set at P<.05.

This study was approved by the Samsung Medical Center Institutional Review Board (2023-06-084). All participants provided written informed consent prior to enrollment. Participation was voluntary, and participants could withdraw at any time without consequence. To ensure ethical fairness, participants in the control group were offered the opportunity to receive HoloLens-based training after the study was completed, if they wished. The clinical trial was registered at ClinicalTrials.gov (NCT06506851).

A total of 42 participants (n=21, 50% per group) were enrolled and completed the study; there were no dropouts (Figure 2). The mean age was 28.00 (SD 3.93) years in the experimental group and 26.48 (SD 2.04) years in the control group. Clinical experience averaged 3.82 (SD 3.98) years and 2.32 (SD 2.17) years, respectively, with emergency department experience of 1.68 (SD 1.23) years and 1.75 (SD 1.09) years, respectively. No significant differences were found between groups in age, clinical experience, emergency department experience, or sex distribution, confirming baseline homogeneity (Table 1). One control-group participant did not complete the posttraining self-report questionnaire; therefore, the self-efficacy and educational satisfaction analyses include 20 control participants (Table 2), whereas all clinical competency analyses include the full sample (N=42).

Table 3 presents the clinical competency outcomes. Learning time was significantly longer in the experimental group (median 18.20, IQR 15.48‐21.67 min vs median 8.98, IQR 5.85‐11.68 min; Mann-Whitney U=377.0; P<.001), reflecting the more structured, step-by-step nature of the AR-guided program. However, this additional learning time translated into substantially faster performance: device setup time was markedly shorter in the experimental group (median 3.67, IQR 2.90‐4.63 min vs median 9.85, IQR 8.03‐11.35 min; Mann-Whitney U=3.5; P<.001), and total performance time was likewise shorter (median 5.83, IQR 4.80‐6.68 min vs median 13.63, IQR 10.33‐15.17 min; Mann-Whitney U=26.0; P<.001). Of note, the between-group difference in total performance time was not detected by a parametric test (independent-samples t test: P=.06) but was highly significant under the Mann-Whitney U test, underscoring the importance of using distribution-appropriate methods for these skewed data.

The experimental group also demonstrated superior accuracy, with a median device operation competency score of 90.00 (IQR 80.00‐100.00) compared with a median of 70.00 (IQR 50.00‐75.00) in the control group (Mann-Whitney U=407.0; P<.001). On the 22-step checklist, the experimental group passed a median of 20.00 (IQR 20.00‐22.00) steps compared with a median of 16.00 (IQR 13.00‐18.00) steps in the control group (Mann-Whitney U=414.0; P<.001; Multimedia Appendix 3). The 2 accuracy instruments were strongly correlated (Spearman ρ=0.86; P<.001), supporting their convergent validity. Notably, the experimental group required significantly fewer assistance requests (median 0.00, IQR 0.00‐1.00 vs median 2.00, IQR 2.00‐3.00; Mann-Whitney U=14.0; P<.001), indicating that AR training fostered greater independence during device operation. The distributions of the principal competency outcomes are shown in Figure 3.

The experimental group reported significantly higher total self-efficacy scores (median 20.00, IQR 18.00‐25.00) than the control group (median 16.50, IQR 13.00‐20.00; Mann-Whitney U=322.5; P=.003). The largest between-group differences were observed in confidence in advising others about Level-1 operation (Mann-Whitney U=345.0; P<.001) and confidence in troubleshooting device problems (Mann-Whitney U=329.0; P=.001). The reverse-scored item measuring performance pressure when observed by others showed no significant difference between groups (P=.80; Table 2).

Educational satisfaction was significantly higher in the experimental group across all items (Table 4). The total score was higher in the experimental group (median 18.00, IQR 16.00‐18.00) than in the control group (median 12.00, IQR 9.75‐15.00; Mann-Whitney U=385.0; P<.001). Participants in the experimental group reported significantly greater confidence in performing rapid Level-1 setups independently (median 4.00, IQR 3.00‐4.00 vs median 2.00, IQR 1.75‐3.00; Mann-Whitney U=369.5; P<.001) and higher perceived benefit from the training (median 5.00, IQR 5.00‐5.00 vs median 4.00, IQR 3.00‐4.00; Mann-Whitney U=354.0; P<.001).

This study provides the first randomized controlled evidence that AR-based training significantly improves emergency nurses’ clinical competency, self-efficacy, and educational satisfaction in operating the Level-1 rapid infusion system compared with traditional guideline-based self-directed learning. The experimental group showed a 62% reduction in median device setup time (3.67, IQR 2.90-4.63 min vs 9.85, IQR 8.03-11.35 min), a 20-point higher median device operation competency score (90 vs 70 out of 100), and a reduction in the median number of assistance requests from 2 to 0. These results suggest that AR technology can effectively address a critical gap in nursing education: the challenge of training for infrequently used but life-critical medical devices.

Our findings align with and extend the broader evidence base supporting AR in health professional education. A systematic mapping review of 156 studies (2010‐2025) confirmed that AR effectively supports hands-on skill acquisition and clinical competency development across health sciences disciplines [8], and an umbrella review of virtual reality (VR) and AR in medical education similarly concluded that these technologies enhance specific professional competencies compared with traditional methods [17]. In the nursing domain specifically, a 2025 scoping review of 41 AR studies in nursing found that head-mounted displays were the most commonly used AR device type, with more than 90% of studies published since 2020 and more than half using RCT designs [18,19]. Recent randomized trials have demonstrated that AR-based training improves procedural performance in specific clinical tasks: Sun et al [5] showed that an AR app significantly improved nurses’ advanced cardiac life support knowledge and crash cart skills, and Othman et al [13] reported that combining AR with gamification significantly enhanced critical care nursing students’ self-efficacy and motivation. A 2025 RCT of VR and AR in emergency health care training further confirmed that immersive technologies improve emergency skill acquisition compared with traditional instruction [20]. Our study adds to this literature by demonstrating AR’s efficacy for a specialized emergency medical device, the Level-1 rapid infuser, which poses unique training challenges owing to its infrequent use and high-stakes application context.

The effect sizes observed in this study were notably large. Expressed as Cohen d, the between-group differences were approximately 3.1 for device setup time, 2.0 for both the 22-step checklist and the number of assistance requests, 1.8 for the competency score, and 1.2 for learning time, substantially larger than the moderate-to-large effects typically reported in AR or immersive nursing education trials, which generally fall in the range of Cohen d≈0.5 to 1.0 (eg, the improvements in knowledge, skills, self-efficacy, and motivation reported by Sun et al [5] and Othman et al [13]). Several design factors most plausibly account for this magnitude. First, the comparator was self-directed learning from a printed manual without any hands-on guidance, a deliberately conservative, real-world control that maximizes the contrast with real-time, in situ AR guidance. Second, all participants were novices with no prior Level-1 experience, leaving substantial room for improvement and limiting ceiling effects in the AR group. Third, the AR program delivered step-by-step holographic instruction anchored to the actual device (finger-pointing cues, material checklists, and embedded video), tightly coupling instruction to the physical task and reducing cognitive load during a procedurally complex, low-frequency procedure. Fourth, the single-device focus permitted concentrated, uninterrupted practice. These instructional design features, rather than the technology alone, are the most likely drivers of the large effects, and the magnitude should therefore be interpreted in the context of this conservative comparator. Cohen d is reported here for comparability with the exemplar studies; the corresponding nonparametric effect sizes (rank-biserial correlation) were of similar magnitude (r=0.71 to 0.98).

The finding that AR training required significantly more learning time yet produced faster performance is consistent with prior work on immersive training modalities [21]. This pattern suggests that AR’s step-by-step, interactive format promotes deeper encoding of procedural knowledge during the learning phase, which subsequently translates into more efficient and accurate performance. A similar trade-off was observed in a controlled trial of AR-assisted intubation training, in which the AR group took longer to complete the training but demonstrated greater adherence to evidence-based procedures [22]. A systematic review and meta-analysis of immersive technology–based nursing education using the GRADE (Grading of Recommendations, Assessment, Development, and Evaluation) approach further supports that immersive technologies improve practical skill outcomes, although with varying effect sizes depending on the specific clinical domain [3].

The significant improvement in self-efficacy, particularly in confidence-related domains such as advising colleagues and troubleshooting equipment, is consistent with findings from studies showing that immersive learning environments foster greater confidence through realistic practice opportunities [21,23]. Notably, there was no significant between-group difference in the reverse-scored item measuring performance anxiety, suggesting that AR training enhanced positive confidence without increasing stress. A recent review of the HoloLens platform for health care simulation confirmed that mixed reality training supports competency development across diverse medical and procedural domains, with learners consistently reporting increased confidence [24]. The substantial improvement in educational satisfaction echoes findings from our prior mixed methods study evaluating AR adoption for nurses in the intensive care unit [4], reinforcing the acceptability and user experience advantages of AR-based educational approaches.

These findings carry practical implications for nursing management. The Level-1 rapid infuser represents a category of medical devices that are critical during emergencies but used too infrequently for conventional training programs to maintain competency effectively. AR-based training offers a scalable solution: once developed, the digital content can be deployed across multiple sites with consistent quality, reducing dependence on expert instructors and physical training resources [11,25,26]. The significant reduction in assistance requests observed in the experimental group suggests that AR training may improve team efficiency during actual resuscitation events by reducing the need for on-the-job guidance from senior staff. For nursing managers seeking to implement digital transformation initiatives [13,27,28], AR-based device training represents a concrete, evidence-supported application with measurable outcomes.

This study has several limitations that should be considered when interpreting the results. First, outcomes were assessed immediately after the intervention, and the long-term retention of skills and knowledge was not evaluated. Future studies should incorporate follow-up assessments at defined intervals (eg, 1, 3, and 6 months) to determine the durability of training effects and the need for refresher training. Second, this was a single-center study conducted within one emergency department, which limits the generalizability of the findings. Multicenter trials involving nurses from diverse clinical settings, including intensive care units and surgical departments, would provide a broader understanding of AR training’s applicability. Third, the study assessed individual performance metrics only; team-based outcomes such as communication, coordination, and collective decision-making during simulated resuscitation scenarios were not evaluated. Fourth, blinding was not feasible owing to the nature of the intervention, introducing the possibility of performance bias. Finally, cost-effectiveness was not assessed; future research should analyze the financial feasibility of AR training programs, including equipment acquisition, content development, and maintenance costs relative to improved clinical outcomes.

This RCT demonstrates that AR-based training using Microsoft HoloLens 2 significantly enhances emergency nurses’ clinical competency, self-efficacy, and educational satisfaction in operating the Level-1 rapid infusion system. Despite requiring additional initial learning time, AR training produced faster, more accurate, and more independent device operation compared with traditional self-directed learning. These findings position AR technology as a viable and effective educational tool for medical devices that are infrequently used but critically important in emergency care. Future research should evaluate long-term skill retention, cost-effectiveness, and the applicability of AR-based training across a broader range of medical devices and clinical settings.