The effectiveness of augmented reality/mixed reality in medical education: a meta-analysis

Abstract

Background

Augmented reality (AR) and mixed reality (MR) are increasingly being applied in the field of medical education. However, the effectiveness and acceptance of AR/MR in different teaching scenarios remain unclear. This meta-analysis aimed to examine the effectiveness of AR/MR in improving medical students' knowledge and skills training.

Methods

PubMed, Embase, Cochrane Library, PsycINFO, Education Source, Education Resources Information Center, CINAHL, Web of Science Core Collection, China National Knowledge Infrastructure, WanFang, WeiPu, and Chinese BioMedical Literature databases were searched. The search encompassed literature from the database establishment to December 2023. Studies examining the use of AR/MR in medical education and reporting outcomes related to knowledge learning or skill training were included. The data were analyzed using Stata version 15.1 software.

Results

Twenty-nine publications were included in this meta-analysis. Compared to traditional teaching methods, AR/MR-assisted teaching showed greater effectiveness in medical skills training, as indicated by the higher skill scores (WMD = 12.31, 95% CI: 4.12 to 20.50), reduced failure rate (RD = -0.08, 95% CI: -0.12 to -0.04), and shortened performance time (SMD = -0.20, 95% CI: -0.36 to -0.04). However, AR/MR did not significantly improve knowledge acquisition (WMD = 2.92, 95% CI: -1.73 to 7.57). The questionnaire survey revealed the advantages of AR/MR in terms of perceived usefulness (PU) (WMD = 0.27, 95% CI: 0.00 to 0.53), perceived ease of use (PEOU) (WMD = 0.35, 95% CI: 0.13 to 0.57), and enjoyment (WMD = 0.67, 95% CI: 0.20 to 1.13).

Conclusion

This meta-analysis highlights the effectiveness and user acceptance of AR/MR in medical education, particularly in skill training, although no significant improvement was observed in knowledge learning. The findings provide a foundational evidence base for the further expansion of AR/MR applications in medical education. However, it should be noted that these estimates may be influenced by heterogeneity and publication bias.

Supplementary Information

The online version contains supplementary material available at 10.1186/s12909-025-08166-8.

Introduction

Augmented reality (AR), along with mixed reality (MR), is increasingly used in medical education and clinical practice. AR and MR technologies have unique abilities to overlay digital information onto real-world environments or blend virtual and real worlds, respectively [1]. These technologies offer unique advantages, such as reducing cognitive load, enhancing spatial understanding, and enabling interactive learning through dynamic visualization and feedback [2]. Since AR/MR-based medical devices have already been approved for medical practice [3], this must be included in the curriculum not only for educational purposes but also for practical reasons.

In medical education, AR/MR applications fall into two broad categories: knowledge learning (e.g., anatomy, pathology) and skill training (e.g., surgery, physical examinations). For knowledge acquisition, AR/MR improves spatial reasoning and retention of complex structures [4], while operational training, such as surgical skills [5], physical examinations [6], interventional diagnosis and treatment, and emergency skills training [7], benefits from highly realistic, repeatable simulations with real-time feedback [8]. AR/MR greatly enriches teaching methods, enhances teaching effectiveness, and promotes the development of clinical thinking and skills in medical students [9]. Compared to traditional methods, AR/MR offers superior safety, engagement, and collaborative learning opportunities [10]. However, the existing published studies have contradictory results, with some suggesting a positive effect on learning outcomes [7] and others indicating no significant impact [11].

To our knowledge, apart from a study by Baashar et al. [12], no meta-analysis has been conducted to evaluate AR/MR in medical education. Moreover, Yahia's research included 13 studies as of April 2021, mainly analyzing the impact of AR on knowledge learning performance, skill operation completion time, and satisfaction, and lacked a multidimensional assessment of AR effects and user acceptance. This study advances beyond previous research by employing a Technology Acceptance Model (TAM)-based framework to comprehensively assess multidimensional indicators, including perceived usefulness (PU), perceived ease of use (PEOU) and user enjoyment, while systematically analyzing different performance metrics across both major AR/MR application scenarios. The evidence base has been substantially enhanced by extending the literature review to include studies up to December 2023, capturing the latest advancements in this rapidly evolving field. Significantly, this marks the first meta-analysis to quantitatively assess key training parameters, including failure rates and performance time. As a result, it provides more comprehensive and up-to-date evidence to inform the adoption of AR/MR educational technologies.

The results of this work have many implications for teachers, medical students, education professionals, and the users of AR/MR technology to promote learning and training in the medical field. This meta-analysis aims to address the following research questions (RQs):

• RQ1: Does the use of AR/MR have an impact on medical education when compared to other instructional methods?

• RQ2: What medical education application scenarios are AR/MR suitable?

• RQ3: How is the user experience of AR/MR?

Methods

Definitions

For the purposes of this meta-analysis, we defined AR as a “computer-generated holographic image overlaid into the real clinical environment, permitting the user to interact with the hologram and objects in the real environment in an integrated fashion” [13], and MR as a “blending virtual objects and information onto a physical environment, interacting with it, and responding to it [9]”.

Inclusion criteria

Based on the PICOS (Population, Intervention, Comparison, Outcomes, Study design) framework, the inclusion criteria were as follows:

• Population(P): Individuals enrolled in undergraduate or graduate medical programs.

• Intervention(I): The intervention group received education involving AR or MR (combining AR/MR with traditional teaching, or purely AR/MR-based teaching).

• Comparison(C): traditional teaching methods, which may include classroom-based learning, no intervention, or other forms of digital and blended education.

• Outcome(O) (at least one of these): (a) Skill assessment, incorporating: the skill score (SS), evaluated via standardized procedures and converted to a percentage format; failure rate (FR), documenting instances where tasks were not successfully completed; and performance time (PT), measuring the duration taken to finish a single task. (b) Knowledge assessment, incorporating: knowledge score (KS), assessed via a written examination designed to test the comprehension of pertinent theoretical knowledge. (c) Participants' experience was evaluated by a standardized self-report questionnaire survey (QS), which included items on PU, PEOU and enjoyment. Additional Notes: The KS and SS results were presented as scores out of 100 and for QS, the response options ranged from “very dissatisfied” (1 point) to “very satisfied” (5 points) on a five-point Likert scale.

• Study design(S): Randomized controlled trials (RCTs).

Exclusion criteria

The exclusion criteria were as follows: (a) studies with unclear outcome measures or insufficient data description; (b) duplicate publications; (c) studies without access to the full text; and (d) studies with a high risk of bias.

Search strategies

We searched seven English-language online databases, including MEDLINE, Embase (Elsevier), Cochrane Central Register of Controlled Trials, PsycINFO, Education Resources Information Center (Ovid), CINAHL (EBSCO), Web of Science Core Collection (Thomson Reuters), and four Chinese databases, including China National Knowledge Infrastructure, WanFang, WeiPu, and Chinese BioMedical Literature. The search encompassed literature from the database establishment to December 2023, and the languages were limited to Chinese and English. The keywords used in the search strategy corresponded to PubMed Subject Headings (MeSH) and included, but were not limited to: “Augmented Reality/Mixed Reality/Education/Learn*/Teach*/Train*/Health/Medicine/Randomized Controlled Trial”. Our study protocol was registered with PROSPERO (Code: CRD42024506938). Table S1 shows the literature search strategy for each database.

Data extraction

Two researchers conducted literature searches based on the inclusion and exclusion criteria. Noteexpress software was used for literature screening and data extraction. Disputed articles were discussed, and the final decision on inclusion was made by the corresponding author. The included literature was then cross-checked by two researchers to extract the following information: author, publication date, location, specialty, application scenarios, equipment, sample size, outcome measures, and intervention duration.

For studies that reported outcomes measured by multiple scales, we prioritized the primary outcome defined by the authors. If no primary outcome was specified, we calculated the average of standardized mean differences across all scales. For studies comparing multiple intervention groups, each group was treated as an independent comparison.

Quality assessment

Ruoxuan Zhang and Xiaoyan Jin independently evaluated the methodological quality of the included studies. Any discrepancies were resolved through discussion or, if necessary, by consultation with a third author, Ming Liu. The methodological quality of the RCTs was assessed using the Cochrane collaboration’s Risk Interventions Risk of Bias Tool (ROB 2.0) [14] reporting each of its five domains (Randomization process, Deviations from the intended interventions, Missing outcome data, Measurement of the outcome, and Selection of the reported result) into a low risk of bias, high risk of bias, or some concerns. Additionally, study domains were classified as low, some concerns, and high risk of bias, leading to an overall study-level risk of bias.

Statistical analysis

Heterogeneity among the included studies was evaluated using the chi-square test, with a significance level of P < 0.1 and I² > 50% indicating significant heterogeneity. A random effects model was employed for the meta-analysis. As part of data standardization, the original assessment scores, such as OSATS or Likert scales, were transformed into percentage values using linear normalization relative to their theoretical maximums. The pooled effects are presented as weighted or standardized mean difference (WMD/SMD) with 95% confidence intervals (CIs), with performance time specifically analyzed using SMD. For failure rates, the effects are presented as risk difference (RD) along with 95% CIs. Subgroup analysis was performed based on the teaching scenarios. Sensitivity analysis was conducted by sequentially excluding each study to assess its influence on the pooled effects. The publication bias was evaluated by a funnel plot and Egger’s test. The data were analyzed using Stata version 15.1 software. A significance level of P < 0.05 indicated statistical significance.

Results

Search results

A total of 386 potentially relevant articles were identified, of which 128 duplicates were removed. At the screening stage, 202 articles were excluded after reading the titles and abstracts, among which 136 were irrelevant to the topic, and 66 did not match the research type. According to the inclusion and exclusion criteria, 56 full-text articles were assessed for eligibility. Among these, 27 studies did not meet the inclusion criteria and had incomplete outcome indicators. Thus, a total of 29 articles were included in this meta-analysis. A summary of the process is outlined in Fig. 1.

Fig. 1.

Flow chart of literature screening

Study characteristics

Most of the 29 included studies were published in 2020 or later (n = 23, 79.31%) and the majority of the studies were conducted in Europe and North America (n = 19, 65.51%). The included studies covered 9 specialties, including Anatomy, Surgery, Imaging department, and Emergency Medicine. Among them, studies on theoretical knowledge learning accounted for 53.33% (n = 16), and those on skill training accounted for 46.67% (n = 14). One of the studies included both knowledge learning and skill training (Table 1).

Table 1.

Basic characteristics of the included studies (n = 29)

Article	Year	Location	Specialty	Application	Equipment(I/C)	Sample size(I/C)	Outcome measures	Intervention duration
Bogomolova et al. [15]	2023	Netherlands	Anatomy	KL	3D holography/monoscopic 3D	32/34	KS,PU,PEOU,EN	45 min
Cizmic et al. [16]	2023	Germany	Laparoscope	ST	iSurgeon/verbal guidance	20/20	PT,FR	/
D'Aiello et al. [17]	2023	Italy	Anatomy	KL	Microsoft HoloLens/regular slideware	19/20	KS	30 min
Farshad et al. [18]	2023	Switzerland	Ultrasound	ST	Microsoft HoloLens/standard US	22/22	PT	/
Felinska et al. [6]	2023	Germany	Laparoscope	ST	iSurgeon/no AR telestration	20/20	FR, PT,PU	/
Guha et al. [19]	2023	UK	Surgery	ST	Microsoft HoloLens/video	18/18	SS,PEOU,EN	20 min
Hayasaka et al. [20]	2023	Japan	Anesthesia	ST	Microsoft HoloLens/traditional	10/10	FR,PU,PEOU,EN	10 min
Malik et al. [21]	2023	UK	Otolaryngology	KL	Microsoft HoloLens/traditional	35/21	KS,PU,PEOU,EN	45 min
An et al. [22]	2022	Korea	Self-Regulated learning	KL	AR app/textbook	31/31	KS	4 weeks
Hou et al. [23]	2022	China	Cardiopulmonary Resuscitation	ST	Microsoft HoloLens/instructor- assisted	14/13	FR	10 min
Little et al. [24]	2022	USA	Anatomy	KL	IVALA®/textbook	38/36	KS,PU,PEOU	60 min
Nagayo et al. [25]	2022	Japan	Surgery	ST	Microsoft HoloLens/video	19/19	SS,PU,PEOU	10 min
Pickering et al. [26]	2022	UK,Greece	Anatomy	KL	MR/screencast	62/84	KS	90 min
Veer et al. [11]	2022	Australia	Asthma	KL	Microsoft HoloLens/textbook	33/34	KS	6 min
Yohannan et al. [27]	2022	India	Anatomy	KL	Air Anatomy/lecture	49/45	KS	3 h
Chen et al. [28]	2021		Health assessment and practice course	ST,KL	3D holography/lecture	40/39	SS,KS	18 weeks
Kamonwon et al. [29]	2021	Thailand	Ultrasound	KL	AR/no AR	23/26	KS	4 weeks
Liu [30]	2021	China	Pulmonary lesions	ST	3D holography/2D CT	10/10	SS	/
Weeks et al. [31]	2021	USA	Anatomy	KL	Microsoft HoloLens/2D screens	15/15	KS	60 min
Bogomolova et al. [32]	2020	Netherlands	Anatomy	KL	stereoscopic 3D AR/monoscopic 3D	20/22	KS,PU,PEOU,EN	45 min
Schoeb et al. [33]	2020	Germany	Surgery	ST	Microsoft HoloLens/instructor	59/105	SS,PU,PEOU,EN	30 min
Stojanovska et al. [34]	2020	USA	Anatomy	KL	Microsoft HoloLens/Dissection	31/33	KS	4 h
Wang et al. [35]	2020	New Zealand et al	Anatomy	KL	Microsoft HoloLens/3DM laptop program	19/15	KS,PU,PEOU,EN	20 min
Noll et al. [36]	2017	Germany	Dermatology	KL	mARble/traditional	22/22	KS,PU,PEOU,EN	45 min
Leitritz et al. [37]	2014	Germany	Ophthalmoscopy	ST	AR ophthalmoscopy/traditional	19/18	SS, FR,PU	15 min
Vera et al. [38]	2014	USA	Laparoscope	ST	ART platform/traditional	9/9	FR, PT	1 h
Albrecht et al. [39]	2013	Germany	Forensic medicine	KL	mARble/textbook	4/4	KS,PU,PEOU,EN	30 min
Wilson et al. [40]	2013	USA	Surgery	ST	AR goggles/lecture	13/21	SS, FR	/
Yeo et al. [41]	2011	Canada	Surgery	ST	Perk Station training suite/traditional	20/20	DS	8 time

ST Skill training, KL Knowledge learning, KS Knowledge scores, SS Skill scores, FR Failure rate, PT Performance time, PU Perceived Usefulness, PEOU Perceived Ease of Use, EN Enjoyment

Study quality

According to the ROB 2.0 tool, 23 of the RCTs were of low risk (79.3%) and 6 of the RCTs were of some concerns (20.7%) (Table 2).

Table 2.

Quality appraisal of the included studies (n = 29) [6, 11, 15–41]

Sensitivity analysis and publication bias

Due to high heterogeneity, a sensitivity analysis was conducted to assess the reliability of the results. In the analysis of performance time, when the study by Yeo et al. [41], Felinska et al. [6] or Cizmic et al. [16] was excluded, the overall effect was not different between the AR/MR and control groups, which was different from the primary pooled effect. In addition to the abovementioned studies, no single study was found to significantly influence the overall pooled effects, indicating the stability of our results (Figure S1). Aside from the failure rate analysis (P = 0.017), no other analyses revealed significant publication bias based on the Egger's test and funnel plots (P < 0.05) (Table S2, Figure S2).

Effects of AR/MR on skill training

A total of 8 publications involving 454 participants (AR/MR group = 205 and control group = 249) reported skill scores. Because high heterogeneity was observed across these studies (I² = 95.8%, P = 0.000), a random-effects model was used. The pooled effects showed a significant difference in skill scores (WMD = 12.31, 95% CI: 4.12 to 20.50, P = 0.003) in favor of the AR/MR group compared to the control group (Fig. 2). Subgroup analysis showed reduced heterogeneity in the invasive procedure subgroup (I² = 5.9%, P = 0.373), with effects consistent with the pooled analysis results (Figure S3). The results showed that the type of technology (AR vs. MR) was not meaningfully associated with the skill scores (the coefficient = −8.661, SE = 9.043, T = − 0.96, P = 0.375).

Fig. 2.

Forest plot for the effects of AR/MR on skill scores

A total of 9 studies reported failure rates, with 3,690 failures in the AR/MR group and 4,487 in the control group. Due to high heterogeneity in the meta-analysis (I² = 67.4%, P = 0.002), a random-effects model was used. Egger’s test indicated significant publication bias in the failure rate analysis (P = 0.028). Within the scope of currently published studies, AR/MR demonstrates a trend toward reducing failure rates (RD = −0.08, 95% CI: −0.12 to −0.04, P = 0.000). However, this estimate may be influenced by publication bias (Fig. 3). Subgroup analysis showed reduced heterogeneity in the invasive procedure subgroup (I² = 18.1%, P = 0.296), with effects consistent with the pooled analysis results (Figure S4).

Fig. 3.

Forest plot for the effects of AR/MR on the failure rate

A total of 5 publications involving 582 participants were identified (AR/MR group = 291 and control group = 291). Heterogeneity analysis revealed an I² of 0.0% and a P value of 0.438, indicating no significant heterogeneity. A random-effects model was used for the meta-analysis to ensure the reliability of the results. The results showed a significant reduction in performance time in the AR/MR group compared to the control group (SMD = −0.20, 95% CI: −0.36 to −0.04, P = 0.017) (Fig. 4).

Fig. 4.

Forest plot for the effects of AR/MR on performance time

Effects of AR/MR on knowledge learning

A total of 14 publications involving 826 participants (AR/MR group = 406 and control group = 420) reported the use of ARs/MRs for knowledge learning. Due to high heterogeneity in the meta-analysis (I² = 87.7%, P = 0.000), a random-effects model was used. The results showed no significant difference in knowledge test scores between the AR/MR group and the control group (WMD = 2.92, 95% CI: −1.73 to 7.57, P = 0.218) (Fig. 5). The meta-regression results showed that the type of technology (AR vs. MR) was not significantly associated with the knowledge scores (the coefficient = −1.342, SE = 4.640, T = − 0.29, P = 0.777). Meanwhile, the live teacher guidance was meaningfully associated with the knowledge scores (the coefficient = 9.380, SE = 3.771, T = 2.49, P = 0.029) (Figure S5). Subgroup analysis by live teacher guidance revealed a significant knowledge score improvement with AR/MR group (WMD = 7.15, 95% CI: 0.80 to 13.50, P = 0.027) (Fig. 6).

Fig. 5.

Forest plot for the effects of AR/MR on knowledge scores

Fig. 6.

Forest plot for the subgroup effects of AR/MR on knowledge scores

Effects of AR/MR on student experience indicators

A total of 12 publications reported 605 student evaluation questionnaires on the PU (AR/MR group = 281 and control group = 324). Due to high heterogeneity in the meta-analysis (I² = 77.4%, P = 0.000), a random-effects model was used. The pooled effects showed that the score of the AR/MR group was significantly greater than that of the control group (WMD = 0.27, 95% CI: 0.00–0.53, P = 0.048) (Figure S6). To understand the PU of AR/MR in knowledge learning and skill training activities, a subgroup analysis was conducted. In the skills training subgroup, which included 5 studies, the AR/MR scores were significantly greater than those in the control group (WMD = 0.30, 95% CI: 0.11–0.49, P = 0.002). Conversely, in the knowledge learning group, which included 7 studies, no significant difference was observed between the AR/MR score and those of the control group (Fig. 7).

Fig. 7.

Effects of AR/MR on student experience indicators

A total of 11 publications reported 568 student evaluation questionnaires on the PEOU (AR/MR group = 262 and control group = 306). Heterogeneity analysis revealed high heterogeneity (I² = 66.5%, P = 0.001), and a random-effects model was used for the meta-analysis. The pooled effects showed that the AR/MR score was significantly greater than that of the control group (WMD = 0.35, 95% CI: 0.13–0.57, P = 0.002) (Figure S7). Subgroup analysis was conducted for the knowledge learning group (7 studies) and the skill training group (4 studies). The analysis using a random effects model revealed that in both the knowledge learning (WMD = 0.40, 95% CI: 0.07–0.73, P = 0.018) and skills training (WMD = 0.32, 95% CI: 0.09–0.55, P = 0.007) subgroups, the PEOU scores of the AR/MR group were significantly greater than those of the control group (Fig. 7).

A total of 9 publications reported 456 student evaluations of enjoyment (AR/MR group = 205 and control group = 251). Heterogeneity analysis revealed high heterogeneity (I² = 90.2%, P = 0.000), and a random-effects model was used. The pooled effects showed that the score of the AR/MR group was significantly greater than that of the control group (WMD = 0.67, 95% CI: 0.20–1.13, P = 0.005) (Figure S8). The subgroup analysis included 6 studies on knowledge learning and 3 on skill training. The results indicated that in both the knowledge learning (WMD = 0.77, 95% CI: 0.10–1.44, P = 0.024) and skills training (WMD = 0.52, 95% CI: 0.23–0.81, P = 0.000) subgroups, the scores for the enjoyment aspect of AR/MR were significantly greater than those of the control group (Fig. 7).

Discussion

The current meta-analysis encompasses 29 studies, 23 of which were conducted between 2020 and 2023, highlighting the rapid expansion of AR/MR applications in medical education with technological advancements. This meta-analysis indicated that employing AR/MR in skill training significantly reduces both failure rates and performance time, while concurrently yielding higher skill training scores. Notably, no significant difference was detected in the knowledge learning scores. Participants find AR/MR more useful, easier to use, and more enjoyable than traditional training methods, especially for skill training.

Information processing theory divides knowledge into declarative and procedural knowledge [42]. AR/MR aids declarative (e.g., anatomy, pathology) via visualizations [43] and enhances procedural skills through immersive simulations, offering real-time operational practice [44].

AR/MR enhanced skill training but limited impact on knowledge learning

While AR/MR excels in skill training within medical education, as evidenced by reduced failure rates and performance times, its impact on theoretical knowledge learning appears less substantial. This meta-analysis revealed no significant advantage for AR/MR over traditional methods in enhancing knowledge comprehension, which is congruent with the findings of Moro et al. [45] and contrasts with the conclusion of Paloma et al. [8].

AR/MR's potential to reduce cognitive load during information processing [46] suggests benefits for learners [27], yet, its reliance on visualization and interaction may inadequately address deeper cognitive processes such as understanding and memory, which are crucial for knowledge acquisition. AR/MR content design must be highly aligned with learning strategies. Simply converting traditional textbooks into 3D models without fully leveraging their interactive and immersive features yields learning outcomes no better than conventional methods. Real-time guidance, explanations, and feedback from instructors can compensate for current technological limitations in content integration and cognitive depth support, thereby facilitating the mastery of declarative knowledge. This finding is consistent with our analysis which identified "teacher assistance" as a significant moderating factor. Therefore, future AR/MR educational interventions should emphasize human–machine collaboration, strengthen the role of teachers in knowledge structuring, and carefully manage cognitive load to prevent information overload and ensure user focus on learning objectives. Of the 14 studies included in this meta-analysis focusing on knowledge learning, all assessed learning outcomes using traditional evaluation methods emphasizing memorization and recall. Although AR/MR technology offers unique advantages in cultivating students' spatial thinking, problem-solving, and collaborative abilities, these competencies may not be adequately quantified through conventional examinations or tests [47]. There is a need for more holistic evaluation methods, such as 3D spatial tests (e.g., molecular structures) and performance-based tasks (e.g., virtual experiments), to assess deep understanding and real-world application. Combining these methods offers a more authentic evaluation of technology-enhanced learning, enhancing the validity of educational assessments.

In contrast, AR/MR technology significantly improved skill training outcomes and reduced failure rates and performance times, consistent with past meta-analyses [12, 13]. It enhances skill acquisition by offering standardized demos and key explanations, facilitating mental model formation and skill mimicry [6]. The real-time simulation of AR/MR in medical training provides authentic intrinsic feedback, which is augmented by extrinsic feedback mechanisms, accelerating skill development and refinement [16, 23, 48]. AR/MR simplifies operational orientation and task execution, reduces cognitive load, and thus accelerates skill formation [30, 49].

The impact of AR/MR in medical education in four stages was evaluated based on the Kirkpatrick model [50]: increased student satisfaction (reaction), enhanced skill performance (learning), reduced performance time (behavior), and reduced failure rate (result). In knowledge learning, although students react favorably (reaction), there is no marked improvement in learning outcomes (learning), and behavior and results assessments are lacking.

This emphasizes the necessity of refined, multilevel evaluation approaches, including spatial abilities and practical skills. These innovative biometric sensors are expected to achieve accurate quantification [6]. Future research should adopt standardized methodologies to address evaluation gaps and optimize AR/MR's implementation strategy for both knowledge and skill development, particularly in light of its notable potential for skill training.

The acceptance for AR/MR in skill training is better than that in knowledge learning

The meta-analysis shows that medical students have varying levels of acceptance of AR/MR in skill training and knowledge learning, which is related to the effectiveness demonstrated by the technology. User acceptance is the key to technological implementation, and Fred D. Davis's (1986) Technology Acceptance Model (TAM) is an important evaluation model. The TAM centers on perceived usefulness (PU) and perceived ease of use (PEOU) as determinants [51]. The PU measures perceived improvement in work performance, while the PEOU reflects ease of use perception [51].These factors are instrumental in forecasting the adoption of emerging technologies. However, to our knowledge, no meta-analysis has yet been conducted to evaluate the user acceptance of AR/MR in medical education, such as PU, PEOU, and enjoyment.

In both knowledge learning and skill training, AR/MR technology has been widely valued for its user-friendly interface and engaging experience. Student feedback indicates high ratings for perceived PEOU and enjoyment. Consistent with AR/MR's impact on learning outcomes, its PU remains moderate in knowledge learning. However, in skill training, learners show stronger recognition of AR/MR's effectiveness in enhancing training outcomes, aligning with the findings of previous meta-analyses that highlighted AR’s efficacy in enhancing learner satisfaction and proficiency in medical training exercises [12]. A large proportion of heterogeneity in user experience metrics remains unexplained, and this variability may be attributed to factors such as qualitative differences in instructional design, software usability, and learners’ prior experience that are not captured.

To effectively implement AR/MR in medical education, it is important to carefully design the course curriculum. Design courses tailored to the characteristics of AR/MR, fully tapping into the unique value and potential of AR/MR technology in the field of medical education. Implementing assessment models that accurately reflect the unique advantages of AR/MR can facilitate a clearer understanding of learning outcomes among students, thereby increasing their acceptance of AR/MR-integrated curricula. By addressing these key considerations, medical schools can optimize the integration of AR/MR technology and create a more effective and engaging learning environment for their students. By doing so, institutions can leverage the potential of AR/MR to enhance both the theoretical comprehension and practical skill development of future healthcare practitioners, thereby accelerating the development of medical education.

Future research should focus on integrating AR/MR with VR and AI technologies to develop hybrid immersive learning environments and assess their long-term educational outcomes. Moreover, comparative studies in diverse teaching scenarios are needed to generalize findings and refine implementation strategies.

Limitations

The main limitation of this study is the high heterogeneity among the included studies. There is significant heterogeneity in terms of research methods, outcome measures, data collection, and reporting due to the different research topics covered by the included articles. This is not surprising because there is considerable variation in medical curricula and educational interventions are typically designed based on specific teaching content and targeted populations to meet specific educational needs. High heterogeneity has also been observed in meta-analyses of other medical and health sciences education-related interventions. The high heterogeneity observed across studies highlights that the effectiveness of AR/MR interventions is highly context-dependent. Differences in technological implementations (e.g., display types, interaction modalities), instructional designs, and assessment protocols can significantly impact learning outcomes. This underscores the importance of interpreting our findings within specific educational and technological contexts. Additionally, there is a pressing need for future research to develop a standardized evaluation framework for augmented reality (AR) and mixed reality (MR) technologies in medical education. It is also important to note that our findings should be interpreted with caution due to evidence of publication bias, which may lead to an overestimation of AR/MR's true effect on reducing failure rates.

Conclusion

This meta-analysis demonstrated that AR/MR has been applied in medical education in Europe, America, and other regions. AR/MR improves medical skill performance in terms of accuracy and speed, particularly for spatial-visual tasks such as surgery, but shows no significant impact on knowledge acquisition. The technology enhances perceived ease of use (PEOU) and enjoyment compared to traditional methods; however, there is a need for improved assessment tools and further validation. However, it should be noted that these estimates may be influenced by heterogeneity and publication bias.

Abbreviations

AR

Augmented reality

MR

Mixed reality

PEOU

Perceived ease of use

PU

Perceived usefulness

RD

Risk difference

SMD

Standardization mean difference

WMD

Weighted mean difference

TAM

Technology acceptance model

Authors’ contributions

R.Z.: Conceptualization, Data curation, Formal analysis, Methodology, Software, Writing – original draft, Writing – review & editing. X.J.: Conceptualization, Data curation, Project administration, Writing – review & editing. M.L.: Conceptualization, Methodology, Project administration, Supervision, Writing – review & editing. H.T.: Methodology, Writing – review & editing. All authors reviewed the manuscript.

Funding

This study was funded by Macao Polytechnic University-funded research project (Grant number RP/AE-04/2022).

Data availability

The datasets used and/or analyzed during the current study are available from the corresponding author on reasonable request.

Declarations

Ethics approval and consent to participate

Not applicable. (This manuscript is a meta-analysis, and does not report on or involve the use of any animal or human data or tissue.)

Consent for publication

Not applicable.

Competing interests

The authors declare no competing interests.