This study conducted disaster response simulation training at Yongin Severance Hospital, Yonsei University, an academic medical center designated as a digital health innovation hospital in South Korea [16], using a digital twin–based virtual hospital platform to simulate a code white (level 1 IT system failure) scenario. The study proceeded through four phases: (1) virtual hospital platform development, (2) IT outage response scenario design, (3) training operation with participation from multiple clinical departments and support staff, and (4) training outcome and participant feedback analysis. The implemented simulation used 7 patient scenarios reflecting outpatient care processes, enabling us to assess quantitative indicators, including completion rates and time requirements. Following training completion, we analyzed participant experiences (collected through surveys) using text mining and word cloud techniques for survey data assessment. Through this approach, we explored platform improvement directions and future expansion possibilities while validating the feasibility and effectiveness of the digital twin–based IT outage response training. The insights we gained were used to propose a new model for disaster response training.

The digital twin–based virtual hospital platform was designed to replicate the actual outpatient environment, incorporating 317 clinical spaces and 6 building entrances (Figure 1). The platform was developed as a web-based application accessible through standard desktop computers, allowing for simultaneous multiuser access during training sessions. Detailed platform technical specifications, including the 9 development modules, system architecture, and 2024 updates, are provided in Multimedia Appendix 1.

For the training exercise, all participants gathered in the hospital’s computer room, where each was assigned a dedicated workstation. The digital twin application ran on the hospital’s local network infrastructure, with participants logging in through individual user credentials corresponding to their assigned roles (eg, physician, nurse, administrative staff, and pharmacist). This centralized physical arrangement facilitated real-time coordination and immediate troubleshooting during the simulation. The training participant organizational structure is shown in Figure S1 and Table S1 in Multimedia Appendix 1. The 9 development modules and their specifications are listed in Table S2 in Multimedia Appendix 1.

This simulation-oriented digital twin platform used preprogrammed patient avatars that entered through designated building entrances at preset times based on outpatient appointment schedules rather than incorporating real-time sensor data or live patient flow integration. The simulation engine adjusted patient movement speeds according to age and severity parameters. Escalator and elevator functions were incorporated to enable realistic interfloor movement patterns matching actual hospital workflows.

Each participant’s interface displayed a third-person view with an integrated dashboard showing real-time patient positions, departmental progress, and task completion rates (Figure 2). This allowed participants to track both individual patient scenario progression and overall training status simultaneously. The dashboard interface was role specific, displaying relevant information based on each user’s assigned department and responsibilities.

The training scenario was designed to simulate a hospital-wide level 1 IT outage (code white) with complete HIS paralysis due to potential hacking or program deployment errors. The scenario assumed that the order communication system and electronic medical records (EMRs) would be completely suspended, whereas network and internal communication infrastructure would remain operational. Under these conditions, emergency prescription programs based on internal messaging systems and rapid patient information lookup systems were designated as alternative means for clinical operations until system recovery.

Seven simulated patient scenarios were designed by complexity level (high, medium, and low) reflecting clinical department, visit type, and presenting symptoms. Scenario design was informed by the typical daily outpatient volume of 3500 visits to provide operational context. The scenario development process, including clinical validation procedures and complexity level criteria, is detailed in Table S3 in Multimedia Appendix 1). Each simulated patient completed the entire outpatient process of registration, consultation, testing, prescription, and payment on the virtual hospital platform, with detailed elements including patient movement paths and waiting times also reflected.

Participants were recruited through purposive sampling of staff directly involved in outpatient IT outage response at Yongin Severance Hospital. Department heads from clinical, nursing, laboratory, radiology, pharmacy, and administrative divisions were asked to identify personnel whose roles would be affected during a hospital-wide IT system failure. Multidisciplinary personnel were enrolled across 7 functional groups: administrative support, nursing support, clinical support, laboratory testing support, imaging support, pharmacy team, and patient services support. No formal sample size calculation was performed as the study aimed to include all relevant staff for a full-scale simulation exercise rather than test a specific hypothesis. All identified personnel agreed to participate; there were no refusals or nonresponses. The training was conducted as a single session on July 3, 2024. All participants received pretraining instruction on digital twin platform access and use of emergency prescription programs and rapid lookup systems based on internal messaging. Details of the 2 pretraining education sessions and the emergency prescription program development are provided in Multimedia Appendix 1. On training day, participants assembled at the training venue and completed real-time training on the platform according to assigned roles. During training, patient locations, departmental progress status, and task completion rates were monitored in real time through the dashboard.

Following training, quantitative data, including completion rates and time requirements for each patient scenario, were collected through the digital twin platform’s recorded logs. Official hospital documents, including IT outage response manuals and the previous year’s (2023) disaster response training records, were also collected. Operational readiness was assessed by a panel of 5 training facilitators (2 IT specialists, 1 nursing administrator, 1 clinical faculty member, and 1 emergency management coordinator) using a 7-item checklist covering system availability, user access functionality, prescription processing, billing accuracy, scenario continuity, emergency broadcast operations, and interface usability (Table S4 in Multimedia Appendix 1). Panel members independently scored each item as “pass” or “fail” based on direct observation during the training exercise, and the overall operational readiness score was calculated as the percentage of items rated as “pass.” Full-time equivalent (FTE) requirements were calculated as the number of participants multiplied by total training duration in hours divided by annual working hours per employee (2080 hours based on 8 hours per day, 5 days per week, 52 weeks per year). Detailed FTE calculation parameters and a comparison of the 2 training approaches are provided in Tables S5 and S6 in Multimedia Appendix 1. To compare resource efficiency between the 2024 digital twin–based training and the 2023 conventional training, we performed a descriptive comparison of aggregated training duration and FTE values as the 2023 records were available only as institutional summary documentation rather than individual-level data. The overall training execution workflow is illustrated in Figure S2 in Multimedia Appendix 1.

Open-ended survey data were collected immediately following the training session on July 3, 2024, using a structured paper-based feedback form distributed to all 60 participants who had provided prior consent. The survey consisted of a single open-ended item (Table S7 in Multimedia Appendix 1), and all 60 participants returned completed forms (100% response rate). Responses were classified into 7 groups (administrative support, nursing support, clinical support, laboratory testing support, imaging support, pharmacy team, and patient services support). Two researchers (SAB and YJK) independently read all responses and generated initial descriptive codes, which were organized in a spreadsheet and grouped into similar categories through discussion [17]. Text mining and word cloud techniques were applied in parallel focusing on core statements aligned with the research objective of “training advancement” to increase analysis validity and derive practical improvement measures.

This study received approval from the institutional review board of Yongin Severance Hospital (9-2026-0044). All participants provided written informed consent before training participation. No patient data were used in this study; all scenarios were based on simulated patients. This study adhered to the principles of the Declaration of Helsinki.

The disaster response simulation training was conducted at Yongin Severance Hospital on July 3, 2024. This training, conducted without prior rehearsal, implemented simulation replacing core IT-based medical services, including consultation, prescription, testing, and payment during medical information system (order communication system and EMR) suspension. A total of 60 participants from multiple clinical departments and support services organized into 7 multidisciplinary groups (outpatient nursing teams, laboratory services, radiology, pharmacy, and administrative support) completed all 7 simulated patient scenarios (Table 1). The 7 simulated patient scenarios were structured according to difficulty level (high, medium, and low), with the complete process of registration, consultation, testing, payment, and pharmacy dispensing performed within the virtual environment, reflecting actual outpatient care flow.

Training time varied according to patient scenario contents, interdepartmental movement routes, and prescription entry procedures, with the medium-difficulty patient group having the longest execution time (patient 4: 46 minutes 47 seconds; patient 5: 42 minutes 18 seconds; Table 2). Meanwhile, the high-difficulty scenarios averaged approximately 35.7 (SD 3.9) minutes, and low-difficulty scenarios were completed within 29.7 (SD 3.6) minutes. In particular, one of the medium-difficulty simulated patients (patient 4) showed extended execution time due to testing wait times and repeated prescription entry (Figure 3). These results suggest that differences in clinical flow and task complexity by patient type can directly affect training time requirements.

The comparison of medical billing between emergency prescription programs and HIS programs showed full concordance (100%); prescription and test result entry timeliness (100%) and scenario consecutive execution rate (100%) also demonstrated full completion quality. The emergency response broadcast system (code white) operation status was also confirmed at 100%. In contrast, digital twin platform operational readiness was rated at 70%, reflecting some technical limitations, including insufficient user permission–based menu structure and inadequate interface intuitiveness. While the emergency prescription program was rated at 90%, needs for improvement in prescription classification item display and target patient designation functions were mentioned (Table 3).

Resource efficiency analysis showed that the digital twin–based training reduced training time requirements by approximately 70 minutes compared to the previous year’s conventional training approach. This translated to a 0.034 reduction in FTE staffing from 0.072 in 2023 to 0.038 in 2024 (Figure 4).

Structured content summarization of participant responses to a single open-ended question (administered immediately after training) identified 5 content categories. Each category is described below; the full set of categorized responses is summarized in Table 4.

Participants expressed positive perceptions regarding training realism and platform effectiveness. Text mining and word cloud analysis of open-ended responses revealed that the most frequently mentioned keywords were “realism,” “collaboration,” and “prescription,” indicating high engagement with the simulation environment and interdepartmental coordination processes.

Regarding future training directions, participants proposed expansion to high-risk scenarios, including inpatient simulation, emergency department simulation, evacuation of ventilated patients, fire evacuation drills, and radiation exposure training. Additional suggestions included code red preparedness training and anticancer medication simulation. From an operational perspective, participants recommended outpatient workflow optimization, development of standardized evaluation metrics, and refinement of order classification functions to better mirror actual EMR systems. Specific technical improvements included verification mechanisms for external prescriptions and enhanced patient search functions within the emergency prescription program.

This paper reports the implementation of a simulation-oriented digital twin platform for IT outage disaster response training at a single academic medical center, engaging 60 participants across multiple clinical departments to simulate complete outpatient workflows during a level 1 IT system failure. The achievement of 100% completion rates across all 7 patient scenarios, combined with perfect accuracy in medical billing (100%) and prescription entry timeliness (100%), demonstrates that digital twin technology can reliably reproduce complex clinical processes in a virtual environment. These quantitative results support the feasibility of simulation-oriented digital twin platforms for disaster response training as a proof of concept, with specific technical refinements identified for future development.

The clinical significance of these findings extends beyond simple task completion metrics. Electronic health record downtime creates substantial patient safety risks, with documented cases of medication errors, delayed treatments, and compromised clinical decision-making during system unavailability [5]. Recent systematic reviews emphasize that health care organizations must develop robust business continuity protocols specifically for prolonged IT outages as recovery times have increased substantially in recent years [18]. Our training platform’s capacity to simulate interconnected workflows across departments prepares staff not just for isolated system failures but for the system-wide disruptions that characterize modern IT disasters, where cascading effects can overwhelm entire regional health care systems.

The 70-minute reduction in training duration (compared to traditional training approaches), translating to a 0.034 FTE decrease (from 0.072 to 0.038), should be interpreted as a descriptive operational comparison rather than a causally attributable efficiency gain given that all 60 participants in 2024 were identical to the 2023 cohort (100% overlap; potential learning effect) and the virtual format inherently eliminated physical transit time. Nevertheless, the elimination of physical redeployment represents an inherent advantage of digital twin–based training as it preserves clinical staff availability during training. In health care environments, where every hour of clinical staff time directly impacts patient care capacity, this efficiency gain enables institutions to conduct more frequent training iterations without compromising operational capacity. Recent digital twin applications in health care have demonstrated similar operational improvements: Siemens Healthineers reported 19-minute reductions in computed tomography and magnetic resonance imaging wait times with 32% and 26% capacity increases, respectively, alongside significant overtime reductions [19]. Studies of hospital digital twin implementations show that workflow optimization through virtual modeling can reduce bottlenecks while simultaneously improving staff satisfaction and reducing burnout [20]. Our resource efficiency data suggest that digital twin training platforms can achieve comparable operational optimization while simultaneously strengthening disaster preparedness—a dual benefit that traditional training approaches cannot match.

The operational readiness scores of 70% for the digital twin platform and 90% for the emergency prescription program, assessed by a 5-member panel using a 7-item checklist (Table S4 in Multimedia Appendix 1), provide concrete targets for system refinement. Participant feedback specifically identified the absence of EMR prescription classification features (“L,” “P,” and “R” coding for laboratory, pharmacy, and radiology orders) as the primary operational constraint. This finding aligns with broader implementation science literature demonstrating that successful health IT adoption requires precise replication of existing clinical workflows and careful attention to user experience design [21]. The identified need for role-based access control and department-specific interfaces reflects fundamental principles in digital twin design: virtual replicas must mirror not just physical spaces and processes but also the cognitive and operational patterns of actual clinical environments [22]. These technical limitations, rather than representing failures, highlight the specific development pathways toward production deployment.

The scale of participation (60 staff members across 7 functional groups) distinguishes this implementation from prior digital twin applications in health care [15], which have predominantly focused on operational optimization rather than disaster response training. Digital twin technology has proven effective in manufacturing and aerospace engineering for testing scenarios too dangerous or expensive to implement in reality, and health care applications are increasingly adopting similar approaches [23]. By extending this capability from workflow optimization to disaster preparedness, our study demonstrates that digital twins can serve dual functions: improving daily operations while simultaneously building organizational resilience against catastrophic failures. This convergence of operational efficiency and crisis preparedness suggests a potential shift in how health care organizations approach both quality improvement and disaster readiness.

Health care IT disasters demonstrate why crisis-specific training is essential. The 2024 Change Healthcare ransomware attack affected 190 million Americans, halted claims processing across thousands of organizations, and forced health care organizations to exhaust personal funds to maintain operations [4]. Health care facilities experienced significant operational disruptions lasting weeks to months, forcing health care organizations to revert to paper-based workflows and, in severe cases, requiring complete system restoration. These events affect entire integrated systems simultaneously (not isolated departments), requiring coordinated multi-department response capabilities that existing training approaches fail to develop.

Notably, existing digital twin implementations focus on operational optimization rather than crisis response. Digital twins can optimize clinical operations by analyzing workflows and resource allocation under normal conditions, supporting real-time monitoring and predictive analytics [8,14]. Conventional disaster training faces different constraints: full-scale exercises require 2 to 8 hours and extensive physical setup, limiting training frequency [24]. Although a Swiss pharmacy study showed task completion improving from 69% to 84% following conventional disaster training, 4-month intervals were required between training sessions [25].

Our platform addresses these limitations through 3 design features. The first is crisis-oriented simulation. Unlike operational optimization tools, our system trains staff for complete system absence rather than process improvement. The second is multi-department integration. Our platform simulated coordinated workflows across administrative, clinical, laboratory, pharmacy, and imaging departments, reflecting the cascading interdependencies of real IT disasters. The third is scalability. Accommodating 60 participants simultaneously without physical setup enables weekly rather than quarterly training sessions, supporting the frequent practice necessary for crisis skill retention.

Finally, our implementation demonstrates rapid deployment potential that previous approaches have not achieved. The Saskatchewan Hospital digital twin integration in Canada focused on emergency department optimization with continuous stakeholder engagement over extended time frames [26]. Our approach of deploying a functional training platform within months and conducting full-scale exercises with 60 participants demonstrates that digital twin technology has matured sufficiently for rapid deployment in urgent crisis preparation contexts, suggesting that disaster preparedness platforms can scale across health care systems more readily than gradual implementation approaches [27].

This study demonstrates that digital twin platforms can transform health care disaster preparedness from infrequent, resource-intensive exercises to routine, scalable training. By safely simulating complete system failures with 60 participants across multiple departments, we show that crisis-specific training can overcome historical limitations preventing frequent, realistic preparation. Our findings suggest that digital twin technology offers a practical pathway for coordinated crisis response training, with potential extending beyond IT disasters to other high-stakes scenarios requiring multi-departmental coordination.

First, this study’s single-institution implementation limits the generalizability of the findings to different health care settings. However, major digital twin initiatives similarly began with single-site pilots before widespread adoption [28]. Additionally, the 100% participant overlap between the 2023 and 2024 cohorts introduces a learning effect that confounds the FTE comparison; future multisite studies with independent participant groups are needed. Implementation science frameworks emphasize phased evaluation, with proof-of-concept studies identifying barriers before scaling. Our study establishes feasibility for disaster-specific digital twins, offering a foundation for multi-institutional validation.

Second, the results showing 70% digital twin platform operational readiness and 90% emergency prescription program operational readiness demonstrate that the platform requires further refinement. However, the absence of EMR prescription classification and role-based access control represents an implementation gap rather than a design flaw. Early electronic health record implementations faced similar challenges requiring iterative development [29]. These are solved problems in commercial systems: our identification of specific requirements provides a concrete development road map. We also acknowledge that the operational readiness assessment was not based on a validated usability instrument (eg, System Usability Scale or Post-Study System Usability Questionnaire); future studies should incorporate such instruments to capture individual-level usability perceptions.

Third, only 7 patient scenarios were implemented; thus, the simulation had limited variation in complexity. Training literature demonstrates that focused scenarios enable foundational skill development and recommends iterative introduction of complexity. Our scenarios established pragmatic range boundaries. Future iterations should incorporate scenarios common in actual cyberattacks: multiple simultaneous high-acuity patients, equipment failures, communication degradation, and incomplete information decision-making. The absence of these variables constrains the generalizability of our findings to real crisis conditions, where cognitive overload and resource exhaustion are defining challenges.

This study demonstrates that digital twin–based virtual hospital platforms can effectively train staff for IT outage disaster response. Achieving 100% scenario completion validates functional feasibility as a proof of concept for crisis preparation training. Our findings suggest that simulation-oriented digital twin platforms may help address limitations in disaster preparedness training by enabling repeatable exercises without disrupting operations, although further validation using standardized usability instruments and higher-fidelity scenarios is needed. This approach has potential applications extending beyond IT outages to other disaster scenarios.