Our research was a prospective, single-center, randomized controlled PoC trial. According to the planned surgical approaches, participants were stratified into the frontal approach group (group 1), parietal approach group (group 2), and occipital approach group (Group 3), with randomization within each stratum to the experimental group (AR group) or the control group (physical-model group). Randomization within each surgical-approach stratum used a 1:1 allocation ratio (AR group vs physical-model group). The allocation sequence was generated using a computer-generated random-number list with simple randomization. To ensure allocation concealment, the sequence was implemented using a centralized web-based system prepared and maintained by an independent statistician. Baseline assessments (eg, State-Trait Anxiety Inventory [STAI] and Montreal Cognitive Assessment [MoCA]) were completed before group assignment. Participants were enrolled at the First Affiliated Hospital of Xiamen University, and assignment to interventions was implemented through the concealed allocation mechanism described above. Because of the nature of the interventions, clinicians and communication recipients were not blinded to group allocation. All recipient-reported outcomes (comprehension, anxiety, and satisfaction) were completed by the communication recipient. Neurosurgeons’ communication skills were evaluated using the SEGUE scale based on video recordings by independent assessors who were not involved in intervention delivery (assessors were not informed of allocation and scored recordings labeled only with study IDs).

This study aimed to preliminarily validate the feasibility and advantages of AR technology in preoperative risk communication for neurosurgery. By superimposing an individualized 3D virtual model from the AR headset onto the real environment and demonstrating it to the communication recipient, we compared this approach with the traditional physical model–assisted communication method across both subjective and objective dimensions. This comparison aimed to explore the impact of different communication-assistance technologies on recipients’ understanding of surgery-related issues (eg, surgical methods, surgical risks), communication satisfaction, and anxiety levels.

This study included 67 patients who visited the Department of Neurosurgery at the First Affiliated Hospital of Xiamen University between January and September 2025 and were diagnosed with intracranial space-occupying lesions through auxiliary examinations after admission. As this is a communication intervention study, the enrolled participant was defined as the “communication recipient,” that is, the person who received the standardized preoperative risk communication and was primarily involved in SDM and consent for surgery. This was the patient when decisional capacity was intact; otherwise, an LAR served as the communication recipient and decision maker. Unless otherwise stated, all psychometric outcomes (STAI and MoCA) and postcommunication assessments (comprehension and satisfaction) refer to the communication recipient (patient or LAR).

All communication recipients (patients with decisional capacity or LARs, when applicable) provided written informed consent before enrollment. Participants were explicitly informed that the preoperative communication session would be video-recorded for research purposes (SEGUE-based assessment of communication quality), that refusal to be recorded would not affect clinical care, and that they could request discontinuation of recording and withdraw from the study at any time without penalty. Video recordings, questionnaire data, and study logs were deidentified by assigning unique study codes and stored on a password-protected, secure institutional drive with access restricted to the research team. Any patient imaging data used to generate the 3D models were processed in deidentified form for research use. Textboxes 1 and 2 detail the inclusion and exclusion criteria, respectively.

Head-mounted AR device (SeerLens B50R AR Glasses; Xvisio Technology) and a physical anatomical model (Hongyu Medical Education Equipment Co Ltd).

The AR intervention used the SeerLens head-mounted AR device. Key manufacturer-reported specifications relevant to clinical use (eg, display type, resolution, field of view, tracking method, weight, and computing/connection configuration) are summarized in Multimedia Appendix 1.

At enrollment (before the communication session), we collected demographic characteristics of the communication recipient, including age, sex/gender, and educational attainment (recorded as the highest completed level and, for reporting in Table 1, collapsed into 3 categories to reflect the Mainland China compulsory education context and to reduce sparse subgroup cells). For LARs, we additionally recorded their relationship to the patient. When specified in the study protocol, baseline psychological and cognitive characteristics were also collected, including baseline anxiety assessed using STAI Y-1 and STAI Y-2 (before the communication session) and cognitive screening assessed using the MoCA (before the communication session). Baseline clinical and surgical variables (eg, diagnosis category and planned surgical approach) were extracted from the medical record.

Considering the characteristics of preoperative risk communication scenarios, this study used the open-source software 3D Slicer (The Slicer Community) to premodel classic lesion locations during digital 3D modeling. To protect patient privacy, deidentification procedures were implemented at the premodeling stage, including a comprehensive review of patient images to exclude any models exhibiting identifiable facial features. Visual inspection by 2 independent neurosurgeons (ZY and ZS) confirmed that none of the included prefabricated models were amenable to facial-feature recognition. As strict precision is not required for risk communication scenarios, the premade models were selected and adjusted according to the patient’s 2D imaging data before use (Figure 1A-C). Model adjustments can be made on a computer and quickly transferred to the SeerLens AR head-mounted device, maximizing time saved in preparation before communication.

During the model demonstration phase, the AR device used in this study (SeerLens B50R AR Glasses) was developed and manufactured by a domestic company (Multimedia Appendix 1). During the demonstration, the communication recipient uses the head-mounted AR device to superimpose a virtual projection onto the real patient’s head and, in combination with the neurosurgeon’s explanation, observes spatial information such as lesion size, depth, location, and adjacency. If necessary, the communication recipient can interact with the virtual model in real time through voice or gestures, under the neurosurgeon’s guidance, based on their needs (Figure 1D).

During AR-assisted sessions, the communication recipient viewed the patient-specific 3D anatomical model through a head-mounted AR display. The neurosurgeon did not wear a headset; instead, the neurosurgeon used a computer-based interface synchronized in real time with the recipient’s AR view. This workstation allowed the neurosurgeon to manipulate the model (eg, rotation, zoom, translation), adjust visual properties (eg, transparency/occlusion), and emphasize specific structures (eg, isolating the lesion and adjacent anatomy, highlighting/labeling), while monitoring the same model state shown to the recipient. In this way, the neurosurgeon could verify that the explanation referred to the correct virtual anatomical structures and direct attention through synchronized, on-model cues (Figure 1E).

Before randomization, eligible patients and their families were informed about the trial. Written informed consent was obtained from the patient when capable; otherwise, consent was provided by the patient’s LAR. During this process, 5 eligible patients declined participation. Ultimately, 62 communication recipients (patients or LARs) completed the Chinese version of the STAI-Form Y, which was used to evaluate state anxiety (State Anxiety Inventory, STAI Y-1) and trait anxiety (Trait Anxiety Inventory, STAI Y-2), with higher scores indicating higher levels of anxiety [24]. To evaluate the comprehension abilities of the communication recipients, cognitive function was assessed using the MoCA (Chinese 7.1-Beijing) [25]. Randomization was then performed within 3 strata. In the experimental group (AR group), communication recipients wore the SeerLens AR device when necessary, and the neurosurgeon superimposed the digital 3D model onto the patient’s real head, demonstrating the corresponding spatial structures based on the patient’s lesion location and explaining them to the communication recipient. Here, “real-time support” denotes the in-session use of the AR overlay as a shared spatial reference aligned to the patient’s lesion location, allowing the explanation to progress in step with the conversation and recipients’ questions. This wording describes the workflow and does not imply that the AR arm isolates a single mechanism (eg, spatial registration vs enhanced interactivity), nor that any observed effects can be attributed to AR hardware alone. In the control group (physical-model group), the neurosurgeon provided explanations with the assistance of a physical anatomical model. To minimize between-arm differences in informational content, clinicians followed a prespecified checklist of key information items derived from the SEGUE framework. The checklist covered (1) diagnosis and lesion location, (2) relevant neuroanatomical relationships, (3) the proposed treatment plan and surgical corridor/approach, (4) major risks and high-risk regions, (5) alternative options and the expected postoperative course, and (6) an opportunity for questions and teach-back (Multimedia Appendix 2). The checklist was identical for both arms; the only planned difference was the visualization tool used during explanation (AR-based patient-specific 3D model vs conventional physical model). Although informational items were standardized via the checklist, the AR aid inherently differs from the physical model in patient-specific anatomical specificity and software-enabled, dynamically adjustable visualization (eg, structure isolation/toggling and transparency adjustment); therefore, the between-arm contrast should be interpreted as a bundled comparison of these visualization-support features. In both arms, recipients were encouraged to ask questions and to interact with the tool jointly with the clinician.

After the communication session, the communication recipient completed postsession assessments, including (1) the Medical Knowledge Comprehension Questionnaire, which included a subjective understanding rating scale (Cronbach α=.88; Multimedia Appendix 3) to assess subjective comprehension and an objective understanding assessment (Multimedia Appendix 4): we used the Objective Assessment of Medical Knowledge Comprehension Following Risk Communication, a structured, directed questionnaire covering 4 domains (anatomical localization, treatment surgery, potential risks, and expected benefits), plus an open-ended item, developed to provide a pragmatic, objective quantification of recipients’ comprehension of the core information delivered during preoperative risk communication; (2) anxiety assessment (STAI-Form Y, Chinese version: STAI Y-1 and STAI Y-2); and (3) communication satisfaction assessment, using a self-developed rating scale (Cronbach α=.86; Multimedia Appendix 5). All questionnaires were self-administered by the communication recipient without assistance; no investigator administered the objective questionnaire or guided responses. Multiple-choice items were scored using a prespecified answer key.

Communication duration was measured in minutes by the neurosurgeon conducting the session and was defined as the face-to-face risk communication time from the start of the standardized explanation to the completion of the conversation. This metric did not include any presession technical preparation required for the AR workflow, which was completed before the communication session. The entire communication process was video-recorded so that independent researchers could later evaluate neurosurgeons’ communication skills using the SEGUE scale (Chinese version) [26,27].

The primary outcome was objective understanding, assessed immediately after the communication session using the directed multiple-choice knowledge questionnaire (Multimedia Appendix 4). The key secondary outcome was subjective understanding, assessed using the subjective understanding rating scale (Multimedia Appendix 3). Other secondary outcomes were communication satisfaction (Multimedia Appendix 5), pre-to-post changes in anxiety (STAI Y-1 and STAI Y-2), communication duration, and neurosurgeons’ communication skills, assessed using the SEGUE scale based on video recordings.

We considered a 30% difference between the control and experimental groups to be clinically significant, based on the score range of the primary outcome measure (knowledge questionnaire scores). We assumed an expected effect size of d=1.50 (preset coefficient of variation=0.2), α=.05, and power (1–β)=0.85. The sample size was computed using G*Power (version 3.1.9.7). The required sample size was 10 participants per arm (20 total) for a single comparison. As we prespecified 3 anatomical strata and planned stratified analyses, we targeted approximately 20 participants per stratum (≈10 per arm), corresponding to a minimum total sample size of 60 across all strata (Figure 2A shows the per-stratum calculation). Although this sample size meets statistical requirements, as a single-center PoC study, it may still be constrained by the actual case volume. The primary analysis was conducted on the pooled cohort; powering per stratum was used as a conservative design choice to ensure adequate precision for the prespecified stratified analyses.

Patients, caregivers, or members of the public did not serve as partners in the design, conduct, analysis, or reporting of this PoC randomized trial. Recruitment and communication occurred in time-sensitive preoperative settings, and the study focused on feasibility and measure refinement; therefore, establishing and supporting a Patient and Public Involvement and Engagement advisory group was not feasible at this stage.

All 62 enrolled communication recipients completed the risk communication, with no dropouts. The primary communication recipient (who completed all study assessments) was the patient in 30 cases and an LAR in 32 cases (Table 1). Specifically, there were 20 participants in the frontal approach cohort (group 1), 21 in the parietal approach cohort (group 2), and 21 in the occipital approach cohort (group 3). Within the frontal approach group, the AR group (group 1A, experimental) comprised 9 patients, and the physical-model group (group 1B, control) comprised 11. Within the parietal approach group, the AR group (group 2A, experimental) included 11 patients, and the physical-model group (group 2B, control) included 10. Within the occipital approach group, the AR group (group 3A, experimental) included 12 patients, and the physical-model group (group 3B, control) included 9 (Figure 3 and see Multimedia Appendix 6 for the CONSORT [Consolidated Standards of Reporting Trials] checklist).

After assessing normality (Figure 2B), we processed the data according to their underlying distributions. No baseline differences in demographics were observed among communication recipients across groups (Table 1). Before risk communication, there were no statistically significant pooled between-arm differences in baseline anxiety (STAI Y-1, P=.16; STAI Y-2, P=.67) or cognitive function (MoCA score, P=.06). Similarly, no statistically significant within-stratum baseline differences were observed (Table 1; Figure 4).

Recipient-type distribution was similar between groups (AR: 17/32, 53%, patients and 15/32, 47%, LARs; physical model: 13/30, 43%, patients and 17/30, 57%, LARs; Fisher exact test P=.46; Table 1). Recipient-type robustness analyses did not suggest that anxiety or satisfaction outcomes were driven by recipient-type composition. Interaction tests were not statistically significant for subjective understanding, satisfaction, or anxiety outcomes (P_interaction=.22 for subjective understanding; P_interaction=.07 for communication satisfaction; P_interaction=.71 for ΔSTAI Y-1; and P_interaction=.93 for ΔSTAI Y-2). However, for communication satisfaction, the group × recipient-type interaction approached conventional significance (P_interaction=.07), with a substantially larger absolute AR – control difference among LARs (37.40 vs 24.71; Δ≈12.7 points) than among patients (37.29 vs 32.92; Δ≈4.4 points; Table 3).

Adjusting for recipient type did not change the overall conclusions (objective understanding P=.006). Robustness results are presented in Table 3.

In the prespecified pooled primary confirmatory analysis, AR-assisted preoperative risk communication in neurosurgery—compared with conventional aids centered on physical models—significantly improved recipients’ objective understanding of the current condition and surgery-related information. Although gains in self-perceived understanding were modest, recipients reported high subjective satisfaction with the communication. These effects are consistent with AR’s intuitive, 3D, and comprehensive spatial visualization. In addition, user-friendly head-mounted AR devices support real-time human-computer interaction, which may further enhance both subjective satisfaction with the process and objective understanding of the content. This pattern aligns with prior work showing that digital 3D planning tools improve the quality of surgeon-patient communication during informed consent and objectively promote patients’ understanding of their condition [4,19].

Notably, objective understanding did not differ significantly between groups in the frontal lobe approach subgroup (P=.57). This null finding should be interpreted cautiously. First, our a priori sample size calculation was based on the primary overall comparison and assumed a very large effect size (Cohen d=1.50). The frontal subgroup included only 9 and 11 participants and was therefore likely underpowered to detect moderate effects; the nonsignificant result may represent a type II error rather than evidence of no effect. Second, a clinically plausible explanation is that frontal approaches may involve relatively intuitive anatomical landmarks and a more straightforward surgical corridor; in such situations, a generic 3D physical anatomical model may already provide sufficient spatial cues for lay recipients, leaving limited room for additional improvement with AR (ie, a potential ceiling effect). By contrast, when lesions are deeper, risk-relevant structures are less visually accessible, or the planned corridor is more circuitous, AR may offer greater incremental value through multiangle exploration and dynamic, structure-specific highlighting and annotation. Taken together, the location/approach subgroup findings (including both significant and nonsignificant results) should be considered exploratory and hypothesis-generating.

Moreover, the control group demonstrated relatively low objective understanding scores (mean 9.40 and 9.28 out of 20 for the parietal and occipital regions, respectively), raising the question of whether physical model–based explanations are sufficiently effective for communicating complex surgical concepts. This low baseline performance may reflect the inherent challenges of explaining intricate surgical procedures using static models, or it may indicate that the standard physical model fails to adequately convey 3D anatomical relationships. This finding highlights the potential benefit of AR-assisted visualizations, which may better engage patients and provide more intuitive, spatially accurate representations.

Our study design could not fully disentangle the effect of personalization from that of the immersive technology. Specifically, the AR arm provided a patient-tailored 3D visualization based on the communication recipient’s routine imaging and adjusted to the lesion location, whereas the control arm used a generic physical anatomical model; therefore, the observed improvement in objective understanding may partly reflect content specificity rather than the AR delivery method itself. We did not include a patient-specific 2D imaging arm because this PoC trial was designed as a pragmatic study aligned with routine clinical practice in our mainland China setting, where generic physical anatomical models are among the most commonly used adjuncts for anatomy teaching and preoperative communication in many surgical departments. Moreover, in real preoperative neurosurgical risk communication, patient-specific 2D images alone are often difficult for nonexperts to interpret and may be insufficient to convey complex 3D relationships critical for risk discussion (eg, lesion depth, adjacency to eloquent regions, and surgical corridors). Finally, it is important to acknowledge that the use of AR in preoperative neurosurgical risk communication inherently combines 2 inseparable features—patient-specific personalization and an immersive, interactive experience. In real-world applications, these attributes do not operate in isolation but function as an integrated whole, jointly shaping communication outcomes. We therefore believe that the improved objective understanding of the surgical plan and related risks among communication recipients reflects the synergistic effects of personalization and immersive interaction. In addition, because head-mounted spatial presentation (in situ overlay) and enhanced interactivity (dynamic view manipulation) co-occurred in the AR workflow, this PoC design was not intended to determine which of these components primarily drove the observed differences.

We emphasize that AR’s 3D, immersive, and interactive features provide spatial anchors for apprehending complex intracranial anatomy and surgical pathways. In this study, “enhanced interactivity” refers to tool affordances that allow rapid switching between views (eg, rotation/zoom, structure toggling, transparency adjustment, and clipping planes) during joint clinician-recipient exploration of the 3D model; physical models also support interaction (eg, holding, rotating, and pointing), but with fewer view-manipulation affordances. In addition to conventional verbal explanations and physical anatomical models, AR head-mounted displays can provide an optional patient-specific 3D visualization. During the session, the clinician and the communication recipient can jointly orient to key structures, the target lesion, and clinically relevant risk regions using the same visual reference.

By rendering abstract spatial information visible and manipulable, AR affords an immersive, participatory experience that improves objective understanding and increases satisfaction. It is worth noting that although the mean changes in state anxiety (∆STAI Y-1) did not differ significantly between groups, the data revealed high variance across all strata (eg, SDs ranging from 7.35 to 19.32 in AR groups and reaching 13.52 in the parietal control group). This substantial variability suggests that high-fidelity visualization—whether via AR or physical models—may exert a “polarizing effect” on recipients’ emotional states. This interpretation is consistent with recent meta-analytic evidence indicating that, while virtual modalities consistently improve understanding, their impact on anxiety remains variable and nonsignificant compared with standard care [28]. For some individuals, the “unveiling” of the lesion’s precise location reduces the fear of the unknown, providing reassurance. Conversely, for others, the vivid and realistic depiction of the pathology may overwhelm their coping mechanisms. This aligns with modern psychological frameworks on “information avoidance,” which posit that detailed disclosure of threatening health information can paradoxically heighten distress in individuals prone to blunting coping styles, despite the objective gain in knowledge [29]. Therefore, the nonsignificant difference in mean anxiety scores likely reflects these opposing reactions canceling each other out. Meanwhile, self-rated understanding is strongly influenced by prior knowledge, expectations, and emotional state; over short intervals, it may not rise in parallel with objective measures, a finding consistent with our results [30]. These observations suggest that a single preoperative discussion is a necessary component of SDM but not a complete intervention chain. A comprehensive SDM process should integrate emotional support, psychoeducation, and expectation management into multi–time point, multichannel, continuous communication, coupled with counseling and follow-up across the preoperative-intraoperative-postoperative continuum. Within this chain, AR should function not merely as a display tool but as a vehicle for preference clarification and value prioritization. Integrating humanistic care with emerging assistive technologies is a likely future direction for clinician-patient communication.

Finally, there is another interesting result in this study. Our recipient-type robustness analyses also suggest a potentially meaningful trend in communication satisfaction. Although the group × recipient-type interaction did not reach the conventional threshold for statistical significance (P_interaction=.07), the magnitude of the AR-associated improvement was substantially larger among LARs (≈12.7-point higher mean satisfaction vs control) than among patients (≈4.4-point higher vs control; Table 3). This pattern is hypothesis-generating, and the present PoC trial was not powered to test effect modification by recipient type; nevertheless, it is clinically plausible that surrogate decision makers may derive greater reassurance and perceived communication quality from patient-specific, manipulable visualization that clarifies anatomical localization and risk-relevant structures in real time. Future multicenter trials with adequate power should prespecify and evaluate recipient-type subgroup effects on satisfaction and related decisional outcomes.

During surgical consultations, clinician-patient dialogue often fails to establish a shared understanding of the anticipated care experience or alignment with patients’ overarching health goals [31,32]. Although surgeons must provide information to support autonomous decision-making, both surgeons and patients may treat informed consent as perfunctory. With its emphasis on 1-way information flow, the process affords limited real-time feedback from patients and often does not prompt reflection on the personal value of surgery or preparation for potential complications. Medicolegal reviews indicate that failure to explain surgical risks and adverse events is the most frequent allegation raised by plaintiffs [33]. SDM is designed to align therapeutic choices with patients’ objectives and, in principle, addresses this deficit [34]. Clinically, SDM commonly involves the neurosurgeon outlining to the patient or LAR the pros and cons of 2 or more therapeutic alternatives [35]. In this process, the mode of communication largely determines effectiveness; the judicious use of assistive technologies in preoperative risk communication can increase patient engagement in decision-making and strengthen both risk communication and SDM. At present, many artificial intelligence–designed decision-support tools (eg, decision aids and question-prompt lists) emphasize information analytics and behavioral prediction while paying relatively little attention to the bidirectional interactions needed to elicit, refine, and apply patient preferences or to support the communication process itself [36,37].

AR can help address these shortcomings by three-dimensionally visualizing personalized virtual models and overlaying them onto the physical clinical environment. Head-mounted AR devices allow real-time interaction via gestures or voice commands, enabling clinicians and patients to bridge the cognitive gap created by abstract spatial concepts. Under illness-related vulnerability and persistent information asymmetry, merely presenting options and describing their pros and cons does little to reduce information and power differentials—particularly in highly anxious contexts [38]. By overlaying virtual models onto the real-world setting, AR gives recipients access to otherwise latent anatomical relationships, substantially improving spatial understanding and the accuracy of risk localization. This, in turn, narrows—at least in the short term—the knowledge gap between clinicians and patients, as confirmed in our study.

Within the SEGUE framework, we minimized confounding from within-group variation in neurosurgeons’ communication competence and quantified skills using SEGUE scores to ensure appropriate assessment and control. This exploratory trial focused on the acceptability of AR assistance in real-world preoperative neurosurgical communication, the feasibility of implementation, and preliminary effects. The findings can guide industry partners in pursuing targeted product refinements for specific clinical scenarios and provide a practical foundation for improving models of neurosurgical preoperative risk communication.

From a clinical perspective, our results suggest that AR-assisted, patient-specific visualization may be a useful adjunct to standard preoperative risk and treatment plan communication, particularly when spatial understanding of neuroanatomy and surgery-related risk regions is critical. Importantly, AR should be viewed as a communication aid rather than a substitute for clinician communication skills; structured, empathic, and comprehensible explanations remain essential regardless of the visualization modality. At present, domestic head-mounted AR devices have not achieved full mass production, and versions for medical scenarios remain under continuous optimization. Critical components, including device reliability, hospital information system interfacing, data security, and legal scrutiny, are still evolving, which objectively limits short-term adoption. Nevertheless, with the maturation of the supply chain and software ecosystem, and with declining device costs, domestic AR devices are likely to be used more frequently in local health care contexts, enabling more targeted services and greater benefit for patients.

Implementation considerations such as usability, staff training burden, workflow integration, and privacy/data security compliance are important for real-world deployment; however, these factors (including costs and cost-effectiveness) were not evaluated in the present PoC trial. Therefore, our findings should not be interpreted as evidence that AR is ready for widespread adoption; rather, they motivate larger, multicenter, pragmatic studies, together with implementation and economic evaluations, to clarify feasibility and value across diverse health care settings.

Our study has several limitations. First, this was a single-center study with a modest sample size, which may limit generalizability. Multiple comparisons are an important limitation of this PoC study. We prespecified a single primary confirmatory comparison (pooled objective understanding) and treated all other comparisons as exploratory and unadjusted; nominally significant exploratory findings should be interpreted cautiously and confirmed in adequately powered trials with prespecified multiplicity-control strategies. Second, clinicians and participants could not be blinded to the visualization modality, introducing potential performance and novelty effects. Although SEGUE scoring was conducted by independent assessors who were not informed of group allocation, complete blinding may have been compromised because the AR headset could be visible in the video recordings; therefore, observer bias in SEGUE ratings cannot be excluded. Third, the costs of learning and process change deserve attention. Despite no significant difference in communication time between the AR and traditional model groups, widespread routine use of AR as a communication aid will require clinicians to spend time beforehand learning and adapting to the associated hardware and software processes. To maximize personalized and targeted communication effects, clinicians also need additional time for preparation before the conversation. We recognize that, for neurosurgical services with heavy workloads, such upfront commitments can create implementation challenges. Even so, additional time ought to be considered a normal element of precommunication preparation, because preoperative rehearsal is beneficial for surgeons, while its effect on care cadence should be balanced through improved processes and a clear division of labor. Furthermore, although we standardized the core informational items delivered in both arms, AR enables dynamic view manipulation and patient-specific visualization, which may alter the salience and perceived amount of information (ie, “effective information load”). Therefore, the observed improvement in objective understanding may reflect not only the technological modality itself but also the integrated features of an AR-assisted visualization package (patient-specific reconstruction and adjustable views). Future studies should include component-isolating comparators (eg, patient-specific 3D visualization delivered on a conventional 2D display, or generic 3D content delivered via AR with standardized interactivity levels) or factorial designs to disentangle personalization, spatial presentation, and interactivity. Fourth, participant heterogeneity should be noted. In keeping with real-world neurosurgical consent processes, some assessments were completed by LARs rather than patients. Although our intervention targets the communication recipient and preoperative decision maker, anxiety and comprehension may differ between patients and proxies. This PoC study was not powered to formally compare effects across recipient types.

In this PoC randomized trial, AR-assisted preoperative risk communication was associated with higher objective understanding of surgery-related information and greater communication satisfaction than communication supported by a conventional physical anatomical model. As the AR arm inherently bundled patient-specific anatomical content, head-mounted spatial presentation, and dynamic view manipulation, these findings should be interpreted as evidence for an integrated visualization-support package rather than for AR hardware alone. AR’s interactive, 3D visualizations provide a more intuitive representation of complex neuroanatomical structures, improving spatial comprehension and enhancing communication satisfaction. While AR did not significantly reduce recipients’ anxiety compared with the physical model, it facilitated a more engaging and participatory experience, aligning with the principles of SDM. Within a standardized SEGUE-informed communication protocol, AR-assisted, real-time, patient-specific visualization may serve as a useful adjunct to SDM by improving recipients’ objective understanding of key anatomical and risk-related information.