Needs assessment and development of an EMR-integrated AI system to enhance nursing handover: NurSync

1. Introduction

Population aging and the increasing societal demand for healthy living have become prominent global challenges. ¹ In Korea, however, the number of active nurses remains critically low, with only 3.8 nurses per 1,000 population recorded in 2018, a figure substantially below the OECD average of 7.4. Although the number of newly licensed nurses per 100,000 population stood at 35.0—exceeding the OECD average of 26.5—this discrepancy indicates a persistent gap between the overall nurse supply capacity and the actual workforce engaged in healthcare delivery. ² The deterioration of the nursing work environment has further exacerbated this issue, leading to frequent resignations and leaves of absence among nursing staff. ³ A 2021 survey by the Korean Health and Medical Workers’ Union revealed that more than half of nurses were dissatisfied with their workload, work intensity, and work-life balance, with many unable to secure minimal rest periods and routinely skipping meals during extended overtime shifts.^4,5 In this context of systemic understaffing and unsustainable workload demands,^3–5 inefficiencies in routine yet safety-critical clinical processes—such as nursing handover—are significantly amplified, making workflow optimization not merely desirable but urgently necessary.^6,7

Nursing handover, performed repeatedly across every shift change by nurses already burdened by understaffing3-5 and excessive documentation demands, ⁸ constitutes a pivotal process for safeguarding patient safety and ensuring continuity of care. ⁹ Most healthcare institutions employ a combination of written and verbal handover practices, which are prone to frequent problems such as omitted information, incomplete documentation, and subjective reporting. ¹⁰ Given that inaccurate or missing information can precipitate medical errors, the necessity for precise and comprehensive documentation is paramount to patient safety.^6,11 Nonetheless, several impediments—including time pressures, non-standardized terminology, outdated information, frequent interruptions, and ambiguous role delineations—undermine the quality and reliability of handovers.^6,7,11,12 The average duration of a handover ranges from 20 to 30 minutes,^6,7 during which novice nurses frequently experience difficulties in identifying and organizing critical information, thereby contributing to heightened stress and job dissatisfaction.^13,14 The reliance on unstructured handover formats further compromises documentation integrity, ⁶ ultimately detracting from nursing efficiency, mental well-being, and patient safety. Consequently, there is an urgent need for standardized training and procedural enhancements to optimize the handover process.^7,11,13

The Electronic Medical Record (EMR) system, which electronically captures, stores, and manages medical information, serves as a fundamental tool for healthcare institutions to enhance care quality and ensure patient safety. ¹⁵ Despite the recognized advantages of EMRs, such as real-time accessibility and improved data accuracy, many nurses report that these systems are slow, complex, and unintuitive, impeding their workflow. ⁸ Moreover, dissatisfaction with EMR-based handover tools is widespread, with 50% of nurses finding printed reports cumbersome and 69% criticizing them for containing excessive and irrelevant information. ⁸

Advances in artificial intelligence (AI) technologies, particularly in natural language processing and large-scale data analysis, have opened new avenues for innovation in healthcare. ¹⁶ AI applications in nursing are expanding across domains such as clinical decision support, patient monitoring, task automation, and education, with growing acceptance among nursing professionals.^17–19 Recent evidence indicates that AI integration alleviates nursing workload, promotes patient safety, and supports patient-centered care models. ²⁰ However, the application of AI specifically to nursing handover remains limited; a scoping review found that most existing technological solutions rely on low-complexity tools, with advanced approaches such as natural language processing largely unexplored and the majority of systems still in pilot stages. ²¹ Although recent evidence suggests that generative AI can reduce handover documentation time substantially, ²² few systems have been developed through systematic needs assessment or designed for direct EMR integration with structured handover protocols, underscoring the need for a clinically grounded approach.

Given the severity of Korea’s nursing shortage and the resulting workload pressures described above,^3–5 interventions that reduce the burden of routine yet safety-critical clinical tasks are urgently needed.^17,20AI-based nursing handover systems are expected to improve efficiency, accuracy, and consistency by automatically summarizing records and visualizing patient status, thereby allowing nurses to focus more on direct patient care. Therefore, this study aimed to identify the requirements for and develop an EMR-integrated, AI-based web and mobile handover system grounded in the PASS-BAR protocol, framed as a development and feasibility study. The findings are intended to provide a foundation for future empirical evaluation of the system’s effectiveness in enhancing handover efficiency and safety, reducing nursing workload, and improving the quality of nursing care.

2. Methods

3. Results

3.1. System needs assessment

3.1.1. Survey

3.1.1.1. General and handover-related characteristics

All participants were female registered nurses employed in the oncology ward. The mean age was 28.2 years, with seven participants (77.8%) holding a bachelor’s degree and two (22.2%) possessing graduate-level or higher qualifications. The average clinical experience was 48.89 months.

The average time spent preparing for handover was 23.89 ± 7.84 minutes, and the average duration of handover itself was 27.78 ± 4.41 minutes. All handovers were currently conducted verbally. Preferred handover methods varied: one participant (11.1%) utilized a documented checklist; four participants (44.4%) relied on verbal handover using the Kardex; three participants (33.3%) employed verbal handover referencing the EMR; and one participant (11.1%) used only written forms such as handover sheets. None reported the presence of formal written handover guidelines, and all handovers occurred at the nurses’ station. Regarding training, three participants (33.3%) acquired handover skills through observation of colleagues, eight participants (88.9%) learned from senior nurses, and one participant (11.1%) received formal education. The perceived importance of handover for patient safety yielded a mean score of 1.22 ± 0.44, while satisfaction with current handover methods scored 2.11 ± 0.60 (Table 1).

OPEN IN VIEWER

Table 1.

General and handover-related characteristics (N=9).

Variables	Category	N	Percentage	Mean	SD
Age (years)				28.2	3.16
Education	Bachelor’s degree	7	77.8	​
	Graduate-level or higher	2	22.2	​
Work experience				48.89	33.47
Time spent preparing for handover				23.89	7.84
Time spent on handover				27.78	4.41
Preferred handover methods	Use of a documented checklist	1	11.1	​
	Verbal handover using the Kardex	4	44.4	​
	verbal handover referencing the EMR	3	33.3	​
	using only written forms such as handover sheets	1	11.1	​
Handover training	Observation of colleagues	3	33.3	​
	Training of senior nurses	8	88.9	​
	Formal education programs	1	11.1	​
Importance of handover for patient safety				1.22	0.44
Satisfaction with the handover methods				2.11	0.60

3.1.1.2. Handover evaluation

Participants evaluated the handover process using a 7-point Likert scale. The highest-rated item was the perception of having received the most recent patient information (6.11 ± 0.78). Participants also reported high levels of agreement regarding the sufficiency of patient information (5.89 ± 0.78), the clarity of the handover delivery method (5.89 ± 0.60), and the opportunity to ask questions about unclear content (5.89 ± 0.93). They indicated that they were able to focus on the information provided (5.78 ± 0.83) and that patient information was delivered in a timely manner (5.78 ± 0.44). Efforts to minimize interruptions during handover were also positively rated (5.78 ± 1.30). Participants considered the explanation of handover content to be clear (5.67 ± 0.71), and they reported that the handover addressed various aspects of nursing care (5.67 ± 0.71) and prioritized upcoming treatments (5.67 ± 1.00). Consideration of patients’ experiences during handover was similarly rated (5.67 ± 1.40), while discussions on potential risks and complications were slightly lower (5.56 ± 0.53). Opportunities to discuss challenging clinical situations were rated 5.44 ± 1.13, and the appropriateness of the time allocated for handover received the same score (5.44 ± 1.05). The logical structure of the handover process was rated moderately (4.56 ± 1.13). However, the ability to visually check patients during handover (3.44 ± 1.00) and the use of consistent tools such as checklists (3.11 ± 1.41) were rated relatively low. The overall quality of the handover process was evaluated at 4.67 ± 0.50 (Table 2).

OPEN IN VIEWER

Table 2.

Evaluation of nursing handover and EMR system satisfaction (N=9).

Category	Variables	Mean	SD
Handover Evaluation	Recency of received handover	6.11	0.78
	Sufficiency of patient information	5.89	0.78
	Clarity of handover content	5.67	0.71
	Ease of understanding handover method	5.89	0.60
	Ability to focus on handover information	5.78	0.83
	Opportunity to discuss difficult clinical situations	5.44	1.13
	Coverage of nursing aspects during handover	5.67	0.71
	Opportunity to ask questions	5.89	0.93
	Appropriateness of handover duration	5.44	1.05
	Timeliness of patient information	5.78	0.44
	Ability to check patient during handover	3.44	1.00
	Prioritization of upcoming treatments	5.67	1.00
	Discussion of risks and complications	5.56	0.53
	Consideration of patient experiences	5.67	1.40
	Logical structure of handover	4.56	1.13
	Use of structured tools (e.g., checklist)	3.11	1.41
	Efforts to minimize interruptions	5.78	1.30
	Overall quality of handover	4.67	0.50
User satisfaction with the electronic medical record (EMR) system	Screen layout and composition	3.78	0.83
	Font	4.11	0.73
	Speed of image/text retrieval	3.22	1.10
	Speed of test result retrieval	3.89	1.10
	Data entry method and time	3.44	0.83
	Security and safety	4.22	0.97
	Use of standardized terminology	3.44	1.03
	Relevance of terminology to tasks	3.67	0.98
	Consistency of terminology	3.44	1.03
	Utilization of necessary information	3.56	0.83
	Provided clinical information	3.00	1.03
	Accessibility of desired information	3.78	1.06
	Interdepartmental data sharing	3.00	1.25
	Speed of task performance	3.78	0.83
	Accuracy of task performance	3.89	0.96
	Work procedure	3.78	0.83
	Workload	3.44	1.04
	Time to focus on professional work	3.44	0.98
	Reflection of user needs	3.56	0.78
	Overall work satisfaction	4.00	0.67

3.1.1.3. User satisfaction with the EMR system

User satisfaction with the EMR system was evaluated using a 5-point Likert scale. The highest satisfaction was reported for system security and safety (4.22 ± 0.97), followed by satisfaction with the fonts used (4.11 ± 0.73), and overall satisfaction with work supported by the system (4.00 ± 0.67). Satisfaction with the accuracy of work (3.89 ± 0.96) and the speed of accessing test results (3.89 ± 1.10) were also relatively high. Moderate satisfaction was noted for screen layout and composition (3.78 ± 0.83), the speed of task performance (3.78 ± 0.83), work procedures (3.78 ± 0.83), and accessibility of desired information (3.78 ± 1.06). The clinical relevance of standardized terms scored 3.67 ± 0.98, followed by satisfaction with how well user needs were reflected in the system (3.56 ± 0.78) and the ability to utilize necessary information (3.56 ± 0.83). Lower satisfaction was observed for the method and time required for data entry (3.44 ± 0.83), consistency of terminology (3.44 ± 1.03), use of standardized terminology (3.44 ± 1.03), workload (3.44 ± 1.04), and available time to focus on professional tasks (3.44 ± 0.98). The speed of retrieving images and text records was rated 3.22 ± 1.10. The lowest satisfaction scores were observed for the quality of clinical information provided (3.00 ± 1.03) and interdepartmental data sharing (3.00 ± 1.25) (Table 2).

3.1.2. Focus group interviews (FGIs)

Analysis of nurses’ experiences with EMR-based handover revealed four main themes and eleven subthemes encompassing effective handover characteristics, difficulties with EMR-based communication, inefficiencies in Kardex documentation, and expectations and concerns regarding AI-based systems. Perspectives varied by clinical experience (Table 3, Figure 1).

OPEN IN VIEWER

Table 3.

Themes and subthemes.

Theme	Subtheme	Participant	Quotation
1. Characteristics of Effective Handover	1.1. Handover Focused on Priority Issues and Recent Changes	Senior nurse	We should prioritize what happened today and what has changed.
		Mid-level nurse	It’s better to start with the current issue and what’s coming next.
		Junior nurse	If I follow the fixed Kardex order, I’m less likely to miss anything.
	1.2. Emphasis on Brevity and Efficiency in Communication	Senior nurse	We need to save time. In the end, efficiency is what matters.
		Mid-level nurse	Ten minutes is fine, but it becomes overwhelming when you have many patients.
		Junior nurse	If seven patients are admitted, it takes almost three hours. I have to do it outside of working hours.
2. Difficulties in EMR-Based Handover	2.1. Fragmented EMR Structure Hinders Information Retrieval	Senior nurse	The EMR is so disorganized that I don’t even know where to start when handing over.
		Mid-level nurse	There are too many tabs, and the names are too similar—it’s confusing.
		Junior nurse	I don’t know where the integrated record is. I have to click through every tab one by one.
	2.2. Lack of Edit Functions Adds Workload	Senior nurse	If the Kardex entry disappears, I have to retype everything—it’s infuriating.
		Mid-level nurse	The readability is poor, and it’s hard to make edits.
		Junior nurse	If I make a mistake while typing, I can’t undo it. I wish there was an undo button.
	2.3. Manual Entry Burden Due to Lack of Automation	Senior nurse	ADRs and past history are not automatically carried over, so I have to re-enter them every time.
		Mid-level nurse	I wish I could copy and paste, but since I can’t, I have to write everything by hand.
		Junior nurse	I have to write everything manually, and I’m not even sure what or how much I need to write.
3. Inefficiencies in Kardex Preparation	3.1. Excessive Time Required	Senior nurse	When we have many patients, it easily takes over 10 minutes per person. Sometimes I can’t finish it all.
		Mid-level nurse	If I have to prepare for more than five patients, it really drains my energy.
		Junior nurse	For newly admitted patients, it takes more than 20 minutes per person. That’s up to three hours for seven patients.
	3.2. Ambiguity in Determining the Scope of Information	Senior nurse	Sometimes it’s just too long, but it’s hard to summarize everything clearly.
		Mid-level nurse	I want to include only meaningful content, but I’m afraid I might leave something important out.
		Junior nurse	I don’t know what part to write or where to start, especially for complex patients.
	3.3. Frequent Revisions Delay Completion of Kardex	Senior nurse	I write it, then check the integrated records, and then I end up editing again and again.
		Mid-level nurse	It takes a long time because I have to revise things repeatedly.
		Junior nurse	I keep checking and editing to make sure everything’s correct—it takes forever.
4. Expectations and Concerns about AI-Based System	4.1. Automation to Reduce Workload and Improve Accuracy	Senior nurse	If Kardex prep time is reduced, I can focus more on assessing the patient.
		Mid-level nurse	If AI could summarize imaging and lab results and consultations, that would be so helpful.
		Junior nurse	I’d love to be able to review it on my phone before work—it would really help.
	4.2. Suggestions for Essential Functions	Senior nurse	Information like ADRs must be included no matter what.
		Mid-level nurse	It would be great if lab results and consult summaries were automatically integrated.
		Junior nurse	We missed a chemo order once because there was no alert. That kind of notification is essential.
	4.3. Concerns about AI Errors and the Need for Human Verification	Senior nurse	Even if AI handles the basics, nurses still need to verify the content. Otherwise, it’s risky.
		Mid-level nurse	Free-text entries should record who wrote them so that accountability is clear.
		Junior nurse	AI-generated records should always be editable by humans.

OPEN IN VIEWER

Figure 1.

Key themes from focus group interview.

3.1.2.1. Theme 1. characteristics of effective handover

3.1.2.1.1. Subtheme 1.1. handover focused on priority issues and recent changes

Participants emphasized that effective handover prioritizes current patient problems and recent clinical changes. Senior nurses advocated concise communication focused on day-to-day updates rather than repeating entire patient histories. Mid-level nurses concentrated on communicating current problems and future care plans, while junior nurses preferred structured formats to ensure completeness.

3.1.2.1.2. Subtheme 1.2. emphasis on brevity and efficiency in communication

All participants agreed that handovers must be conducted efficiently within limited time. Senior nurses preferred summarizing key points based on clinical judgment. Mid-level nurses stressed minimizing time per patient when handing over multiple patients. Junior nurses reported spending excessive time verifying and organizing information, sometimes during off-duty hours.

3.1.2.2. Theme 2. difficulties in EMR-Based handover

3.1.2.2.1. Subtheme 2.1. fragmented EMR structure hinders information retrieval

Participants consistently reported that information was scattered across multiple EMR tabs, impeding prompt access and retrieval. Senior nurses expressed frustration with the lack of a structured interface; mid-level nurses noted tab similarities causing confusion; junior nurses found the layout challenging to understand, leading to excessive time spent searching.

3.1.2.2.2. Subtheme 2.2. lack of edit functions adds workload

Participants reported difficulty due to limited or absent undo and edit functions. Senior nurses explained that minor errors often necessitated rewriting entire sections. Mid-level nurses described the system as difficult to revise and poorly readable. Junior nurses emphasized the absence of an “undo” button, leading to lost entries after input errors.

3.1.2.2.3. Subtheme 2.3. manual entry burden due to lack of automation

The absence of automation required manual entry of large volumes of data. Senior nurses reported repeatedly entering adverse drug reactions, past histories, and treatments. Mid-level nurses cited lack of copy-and-paste functionality causing redundant entries. Junior nurses struggled with uncertainty regarding required information and lacked standardized guidance.

3.1.2.3. Theme 3. inefficiencies in kardex preparation

3.1.2.3.1. Subtheme 3.1. Excessive Time Required

Participants reported significant time demands in Kardex preparation, especially with multiple patients. Senior nurses noted that time per patient often exceeded ten minutes on busy days. Mid-level nurses described increased physical and mental strain when managing over five patients. Junior nurses often spent two to three hours outside work hours completing documentation for new admissions.

3.1.2.3.2. Subtheme 3.2. ambiguity in determining the scope of information

Participants expressed uncertainty about the amount and type of information to include. Senior nurses found condensing lengthy histories into summaries difficult; mid-level nurses worried about omitting meaningful details; junior nurses struggled especially with patients having extensive histories or ICU backgrounds.

3.1.2.3.3. Subtheme 3.3. frequent revisions delay completion of kardex

Participants noted Kardex documents required frequent updating as new information emerged. Senior nurses reported repeated revisions after reviewing integrated records. Mid-level nurses experienced delays due to ongoing edits. Junior nurses found continuous double-checking and modifications significantly slowed completion.

3.1.2.4. Theme 4. expectations and concerns about AI-Based system

3.1.2.4.1. Subtheme 4.1 automation to reduce workload and improve accuracy

All participants expressed optimism about AI automating repetitive documentation. Senior nurses hoped AI would reduce Kardex preparation time and free more time for patient care. Mid-level nurses believed automated summaries of tests and consultations would facilitate handovers. Junior nurses were particularly enthusiastic about AI-generated content review on mobile devices before shifts.

3.1.2.4.2. Subtheme 4.2. suggestions for essential functions

Participants identified key desired features for an AI handover system. Senior nurses emphasized inclusion of critical information such as adverse drug reactions, current medications, and surgical plans. Mid-level nurses advocated automatic integration of medical history, treatment records, and lab results. Junior nurses requested auto-population of patient complaints, procedural history, and alert systems for chemotherapy or imaging orders.

3.1.2.4.3. Subtheme 4.3. concerns about AI errors and the need for human verification

Despite high expectations, participants voiced concerns about relying solely on AI. Senior nurses warned erroneous data might propagate through AI, risking patient safety. Mid-level nurses stressed the importance of traceability and accountability for AI-generated entries. Junior nurses emphasized the need for manual editing capabilities to correct AI errors.

3.2. System development

Based on the results of the needs assessment, we developed a dual-platform (web and mobile) AI-based nursing handover system, hereby referred to as NurSync (Figure 2). The system pipeline operates as follows: (1) structured nursing records (17 types) are retrieved from the data source; (2) each item is processed by the on-device AI model to assign a PASSBAR rank and mapped to protocol categories (Figure 3); (3) nurses may add supplementary notes; (4) items are formatted into structured inference prompts with record-type-specific context; and (5) ranked items are assembled into a handover summary under four PASSBAR categories, representing a reorganization of existing EMR data rather than newly generated content. To mitigate the risk of AI errors, three architectural safeguards were implemented: (1) every AI-prioritized item maintains a direct link to its source nursing record for instant verification; (2) AI-generated handover summaries are presented on a dedicated screen, while all 17 original record types remain fully accessible in their unaltered form; and (3) on-device inference ensures no patient data is transmitted to external servers. The system operates as a decision-support tool, with nurses retaining full authority to review, override, or supplement AI prioritizations.OPEN IN VIEWER

Figure 2.

System development flow.

OPEN IN VIEWER

Figure 3.

Components of the study based on the PASS-BAR framework.

3.2.1. Planning and design

Based on survey and focus group findings, this study targeted major challenges in nursing handovers, such as repetitive documentation, omission of critical information, and inefficiencies from dual recording. Four objectives guided system development: automating data input, enabling AI-based extraction and summarization, reducing last-minute revisions, and integrating handovers with EMR documentation. Accordingly, web- and mobile-based platforms with unified structures were designed for both desktop and mobile use. The system adopted a streamlined PASS-BAR model, reorganized into four categories—Background & Situation, Assessment & Intervention, Safety & Recommendations, and Communication—allowing AI to analyze 17 nursing record types and provide standardized, efficient handover support. In the current development phase, the system was tested using a simulated dataset based on first-cycle chemotherapy nursing records, designed to reflect the structure and content of actual clinical EMR data without accessing live hospital databases. Data privacy and security were ensured throughout development by using de-identified, simulated data.

3.2.2. Software design (UI/UX design)

The UI/UX design was guided by a needs assessment to optimize handover efficiency and reduce cognitive load. The web application employed a table-based layout reflecting EMR structures, enabling simultaneous viewing of multiple records and categorization of content into four clinically relevant sections: Background & Situation, Assessment & Intervention, Safety & Recommendations, and Communication. Visual features, such as keyword chips, vital sign trends, and consultation notifications, enhanced clarity and real-time recognition. The mobile application adopted a card-based interface with expandable bottom sheets and threaded memos, supporting ward rounds and decentralized care through real-time synchronization. Prototyping with Notion and Figma allowed iterative refinement to resolve issues of fragmented information, lack of editing, and documentation burden. Overall, the design ensured usability, clinical relevance, and workflow integration to facilitate accurate, timely, and efficient communication.

3.2.3. AI model development

An AI model was developed to classify and prioritize critical nursing information for clinical handovers. A 2.1B-parameter language model (Kanana-nano) was fine-tuned via a two-stage sequential LoRA strategy: Stage 1 (domain adaptation) used 63,953 instruction-response pairs of Korean language and medical corpora (90/10 split); Stage 2 (task-specific tuning) used 28,055 pairs in which experienced clinical nurses manually classified handover items by PASSBAR rank (1 = routine, 2 = conditionally important, 3 = essential; 95/5 split). All training data comprised entirely fictitious patient scenarios authored by clinical nurses; no real patient records were used. Model validation used a separate, independently sampled dataset of 320 items derived from real clinical records, with full de-identification applied. The model achieved 77.9% accuracy on the KorMedMCQA-Nurse benchmark (17 percentage points above the base model) and 96.28% accuracy on PASSBAR rank classification against expert nurse-annotated ground truth; misclassifications occurred predominantly between adjacent ranks, carrying minimal clinical risk. The model underwent independent external verification by the Korea Artificial Intelligence Verification Institute.

3.2.4. Web application development

The web application was built with a reusable component-based architecture, integrating specialized tables for 17 nursing record types. AI inference was supported through a custom Web Assembly (WASM) module with cache optimization, while real-time synchronization using Supabase ensured consistency between nursing records and AI outputs (Figure 4).OPEN IN VIEWER

Figure 4.

Sample page of NurSync system (webpage).

3.2.5. Mobile application development (iOS)

The iOS mobile app maintained functional and data consistency with the web version while optimizing for touch-based interactions. Core components and handover screens were implemented with Supabase for data storage and llama. cpp for embedded AI inference. Iterative quality assurance addressed UI and performance issues, enhancing usability and system stability (Figure 5).OPEN IN VIEWER

Figure 5.

Sample page of NurSync system (mobile application).

4. Discussion

The findings of this study indicate that nurses spent an average of 23.89 ± 7.84 minutes preparing for handovers, with the handover process itself taking an average of 27.78 ± 4.41 minutes. These results underscore that handovers impose a considerable time burden on nurses, potentially reducing work efficiency. ²⁹ Given that handovers are conducted multiple times daily, non-standardized processes can exacerbate inefficiencies and increase the risk of information omissions and communication errors. ³⁰ To address such issues, structured communication protocols such as SBAR and ISBAR have been widely implemented, demonstrating effectiveness in reducing handover time, minimizing information omissions, and enhancing nurse satisfaction.^31,32 In this study, NurSync was developed based on the PASS-BAR protocol and integrated AI technology to automatically summarize and classify handover information into clinically relevant categories. This approach is expected to effectively reduce time burdens, prevent information omissions, and enhance the standardization and efficiency of nursing handovers.

Survey findings revealed moderate satisfaction with the existing EMR system, with scores ranging from 3.00 to 4.22. Notably, data retrieval speed (3.00) and editing functions (3.11) received particularly low ratings. These findings align with previous studies indicating that limitations within EMR systems not only hinder nursing work efficiency but also increase time consumption through repetitive data entry and corrections, thereby contributing to heightened work-related stress. ³³ In response, this system was developed to integrate AI-based functionalities that facilitate rapid data retrieval and streamline editing processes, thus reducing nurses’ workload and potentially improving overall usability and satisfaction.

Focus group interviews revealed four major thematic areas: characteristics of effective handover, challenges in EMR-based communication, inefficiencies in Kardex documentation, and expectations and concerns regarding AI-based systems. Nurses emphasized the need for concise, priority-focused handovers; however, they reported difficulties arising from fragmented EMR interfaces and the repetitive nature of Kardex documentation. While participants expressed positive expectations regarding AI’s potential to reduce workload and improve accuracy, they also highlighted concerns related to transparency and the need for manual editing capabilities. These concerns are consistent with previous research indicating that although AI contributes to enhancing efficiency and accuracy in nursing tasks—such as automated documentation and clinical decision support—it also raises ethical and practical issues, including algorithmic bias, opacity, and the necessity for human oversight.^17,18,34

To address these multifaceted challenges, this system was developed as an AI-based handover system structured on the standardized PASS-BAR protocol. Specifically, the system reorganized PASS-BAR into four clinically relevant categories—Background & Situation, Assessment & Intervention, Safety & Recommendations, and Communication—and employed AI to automatically classify and summarize 17 types of nursing records. This design mitigates information fragmentation by enabling the simultaneous review of critical data on a single interface, thereby improving both the speed and accuracy of handovers. By minimizing repetitive documentation and providing structured, real-time overviews of patient information, this system is expected to reduce the likelihood of information omissions and communication errors during shift changes while enhancing system usability.

Furthermore, to support diverse work environments, this system was developed as both a web-based and a mobile-based platform. The web application was optimized for desktop use within nurses’ stations, while the mobile version was designed for quick access during ward rounds or shift transitions. This dual-platform approach enhances flexibility and accessibility, enabling nurses to efficiently retrieve structured handover information in real time regardless of their location. The unified design and consistent structure across platforms minimize the learning curve, thereby improving usability, particularly in dynamic clinical settings.

These findings align with the broader trajectory of AI-driven handover research. Recent evidence demonstrates that generative AI integration into nursing information systems can substantially reduce handover documentation time, ²² consistent with the efficiency gains observed in the design of NurSync. Furthermore, the limited adoption of NLP-based approaches in existing handover technology underscores the contribution of this study, ²¹ which combines a fine-tuned large language model with direct EMR integration and a structured handover protocol, offering a clinically grounded direction for advancing the field.

This study has several limitations. First, generalizability is constrained by the single-center design and small sample size (N = 9). Although participants were purposively stratified by clinical experience to maximize variation in perspectives, and data collection continued until theoretical saturation was achieved, the findings reflect the specific context of a single oncology ward at one university hospital in Seoul and may not be transferable to other clinical settings, specialties, or healthcare systems. Second, as pre- and post-implementation evaluations were not conducted, the actual impact of this system on workflow efficiency, user satisfaction, and patient safety remains unassessed. In particular, no experimental validation of the AI model’s outputs in real clinical environments was conducted in this phase; while model accuracy was evaluated against expert-annotated ground truth and verified externally, the potential impact of AI misclassification on clinical decision-making has not been empirically assessed. Although architectural safeguards were implemented to minimize risk, the safety and reliability of the system in live clinical use requires rigorous prospective evaluation. Future research should systematically evaluate these outcomes following system deployment to determine its effectiveness and inform further improvements.

5. Conclusion

This study developed NurSync, an AI-based, EMR-integrated web and mobile system designed to enhance the efficiency and effectiveness of nursing handovers. By automatically extracting and summarizing key information from 17 types of nursing records, this system substantially reduces the time and effort required for handover preparation and documentation.

In conclusion, this study developed NurSync as a development and feasibility study, providing a structured AI-based EMR-integrated system designed to streamline nursing handovers through automatic extraction and summarization of key information from nursing records. NurSync is designed to reduce documentation burden, support accurate information delivery, and enhance shift-to-shift communication, particularly for novice nurses and those managing heavy caseloads. The system’s dual-platform design ensures flexible, real-time access, ultimately improving efficiency, accuracy, and patient safety. Future studies should further evaluate its long-term effects on workflow, nurse satisfaction, and patient outcomes.

Supplemental material