This qualitative interview study explored ICU nurses’ experiences of continuous monitoring, frequent alarms, and vicarious trauma–related emotional strain in a digital critical care environment. We used a qualitative interpretive design with one-on-one semistructured interviews and analyzed the data using reflexive thematic analysis. The study is reported in accordance with the COREQ (Consolidated Criteria for Reporting Qualitative Research) [21] (Checklist 1).

The study was conducted in the integrated ICU of a tertiary hospital in Hangzhou, Zhejiang Province, China. The unit is characterized by high-acuity care supported by continuous monitoring and frequent alarms from multiparameter bedside monitors, infusion pumps, ventilators, and other life-support devices. To minimize interruptions and ensure privacy, interviews were conducted in a private meeting room within the ICU department.

We used purposive maximum-variation sampling to recruit ICU nurses with varied roles and monitoring responsibilities. Sampling aimed to capture diversity in years of ICU experience, professional title (junior, intermediate, or senior), clinical role (eg, staff nurse, team leader, and specialized roles such as extracorporeal membrane oxygenation team member), sex, and exposure to high-intensity rescue situations or end-of-life care. Participants were eligible if they were registered nurses currently working in an adult ICU, routinely exposed to physiological monitoring and alarms, and able to complete a 40‐ to 60-minute interview in Chinese. Nurses on long-term leave during the study period or unable to complete an interview were excluded. Recruitment and preliminary analytic review proceeded iteratively. After the initial interviews, the research team reviewed whether the sample captured sufficient variation in ICU experience, professional title, clinical role, and exposure to high-impact monitoring or end-of-life events. Additional participants were recruited to strengthen variation and examine emerging patterns. A total of 15 ICU nurses participated. Recruitment ended when the dataset was judged sufficient to support coherent theme development in reflexive thematic analysis, with reference to the principle of information power [22,23].

Recruitment occurred between November and December 2025. Potential participants were approached through direct invitation based on the purposive maximum-variation sampling strategy described above. To reduce perceived pressure to participate, the invitation emphasized that participation was voluntary, that declining would have no consequences for their employment, evaluations, or workplace relationships, and that participants could withdraw at any time without explanation. To support spontaneity and reduce rehearsed responses, only the general study aims were provided before the interview, with full topic prompts introduced during the conversation. Two nurses declined participation due to scheduling constraints; no participants withdrew after consenting.

The interview guide was developed based on the literature on alarm-intensive work, alarm management, nursing surveillance, vicarious trauma, and occupational distress in critical care settings; the study aim; and iterative team discussion to ensure clarity, sensitivity, and clinical relevance [24]. Interviews were conducted by 2 researchers: YW (female, MSc in Nursing, clinical research nurse) and XG (female, BSc in Nursing, clinical nurse). Both interviewers had qualitative research experience and no supervisory, managerial, or evaluative relationship with the participants.

Interviews were conducted in Chinese using a semistructured format and lasted 40 to 60 minutes. No nonparticipants were present during the interviews. The interview guide elicited accounts across 4 domains: everyday monitoring routines and alarm exposure; responses to alarms, patient deterioration, and death; perceived pressure related to visible physiological data and residual experiences after shifts, such as intrusive recall or lingering alarm sounds; and coping strategies and suggestions for improving alarm and monitoring systems. Interviews were audio-recorded with consent and transcribed verbatim. Transcripts were not returned to participants for comment or correction. Identifying information was removed during transcription. Transcripts were checked by the research team for completeness and accuracy before analysis, and each transcript was assigned a participant ID (P1-P15). The finalized interview guide is provided in Multimedia Appendix 1.

Data were analyzed using Braun and Clarke reflexive thematic analysis [23,25], with NVivo 15 used to support data organization and retrieval. Analysis proceeded iteratively alongside data collection. The first author led transcript familiarization, coding, and theme development, supported by analytic memos, reflexive notes, and field notes recorded after interviews. Initial coding was conducted inductively while remaining attentive to the study aim and participants’ accounts of monitoring routines, alarm-related events, emotional responses, perceived accountability, and recovery after distressing events. The theoretical concepts introduced in the Introduction informed the interview guide and later interpretation, but they were not used as predefined codes or themes; coding remained primarily inductive and grounded in participants’ accounts. Team meetings were used to discuss emerging codes, candidate themes, divergent accounts, and alternative interpretations. Coding differences were treated as prompts for reflexive discussion rather than as a basis for intercoder reliability, consistent with reflexive thematic analysis. Exemplar quotations were selected to illustrate recurrent patterns and meaningful variation across participants. Quotations were translated into English for reporting, with attention to preserving meaning, nuance, and professional tone.

Trustworthiness was supported through reflexive documentation, field notes, peer debriefing, and an audit trail of recruitment, coding, theme development, and analytic decisions. Team discussions were used to examine emerging interpretations, divergent accounts, and theme boundaries. Thick description of the ICU setting, participant characteristics, monitoring context, and illustrative quotations supported transferability. Participants were not asked to provide feedback on the final themes or findings.

Reflexivity focused on how the researchers’ nursing and clinical research backgrounds may have shaped data generation and interpretation. These backgrounds helped build rapport and supported understanding of ICU monitoring routines but may also have sensitized the team to alarm burden, accountability, and distress after deterioration or death. To manage these influences, interviewers used open-ended prompts, and the team used memos and discussions to examine assumptions about monitoring, responsibility, emotional endurance, and good nursing.

Ethical approval was obtained from The First Affiliated Hospital of Zhejiang Chinese Medical University (2025-KLS-184‐02). All participants provided informed consent. Participation was voluntary, and participants could pause or terminate the interview at any time. No financial compensation or other incentives were provided. Given the emotionally sensitive nature of vicarious trauma narratives, interviews were conducted with attention to distress cues, and breaks were offered as needed. All data were deidentified and stored securely.

Participants described working in ICU environments organized around continuous physiological monitoring, central displays, frequent alarms, and work-related digital communication. Reflexive thematic analysis generated 4 themes and 12 subthemes: (1) digital gaze, (2) digitally mediated vicarious trauma, (3) after-shift residue, and (4) recovery strategies and system-level needs. Demographic characteristics of the participants are presented in Table 1.

Continuous monitoring functioned not only as a safety technology but also as an infrastructure that organized attention, visibility, responsibility, and the sensory experience of deterioration. The findings suggest that alarm-intensive care should be understood as a sociotechnical design issue in which displays, alarm behavior, communication platforms, and emerging AI-supported tools shape both patient safety work and clinicians’ recovery. The discussion therefore focuses on how monitoring systems should represent clinical transitions, govern alarms, protect recovery time, and avoid creating new forms of surveillance.

Our findings suggest that continuous monitoring functioned not simply as a safety technology or a source of alarm fatigue but as an infrastructure that organized attention and work rhythms. Nurses described maintaining a peripheral watch while performing bedside tasks, while frequent alarms fragmented care, documentation, and decision-making. Consistent with alarm-fatigue research on mental workload and alarm relevance [26-28], the monitor-alarm system redistributed attention among the bedside, screens, infusion devices, and emerging alarms.

Monitoring also made patient instability and nursing work simultaneously visible. In our data, when deterioration was difficult to reverse, visible values could heighten nurses’ sense that their bedside responses were open to scrutiny despite their limited control over treatment decisions or patient outcomes. Although shared monitoring displays can support situation awareness and interdisciplinary coordination [29,30], our findings suggest that the same visibility may also create accountability pressure. Digital gaze, therefore, helps explain how display design, alarm behavior, team hierarchies, and critical illness jointly shape visibility and perceived accountability [9-11].

Beyond these everyday work effects, monitoring also mediated nurses’ encounters with deterioration and death. In this study, monitor-generated sensory traces formed part of how deterioration was anticipated and remembered. This extends evidence that ICU professionals exposed to suffering, deterioration, and death may experience trauma-related symptoms or secondary traumatic stress [16,31] and that intrusive memories may persist after work-related traumatic events [32]. Rather than independently causing vicarious trauma, monitoring technologies may shape the sensory and interpretive conditions under which high-stakes events are witnessed.

Participants’ accounts suggest that monitoring systems are not neutral during high-impact clinical transitions. In our data, weakening waveforms, escalating alarms, numerical decline, and subsequent silence marked the movement from active deterioration or resuscitation toward a postevent or end-of-life state. This raises a distinct design question: how should monitoring systems behave when clinical status and care goals are changing? Human factors research on patient-monitor interfaces emphasizes designs that support real-time interpretation of evolving physiological information and safe, efficient monitor use [12,33]. Our findings extend this concern by suggesting that interfaces should make changes in clinical and care status legible rather than simply presenting successive abnormalities.

This design challenge is particularly salient at the end of life. Participants’ suggestion of a postresuscitation buffer mode reflects a perceived mismatch between routine alarm behavior and the clinical context after resuscitation termination or death. End-of-life alarm research supports this concern: arrhythmia alarms may remain audible after comfort-care initiation [34], and ongoing telemetry alarms can interfere with end-of-life conversations and family interaction [35].

The proposed buffer mode should therefore be regarded as a data-grounded design hypothesis rather than a ready-to-deploy solution. Future participatory work could examine a clinician-initiated, workflow-governed postevent mode following resuscitation termination, confirmed death, or other end-of-life transitions, with users, workflows, and real-world data informing its development [2]. Because withdrawal of life-sustaining treatment and end-of-life transitions require protocolized clinical processes, any such mode should follow clinical confirmation rather than automatically inferring death [36]. It should preserve safety-critical alarm escalation while reducing repetitive low-priority alerts and making relevant device status and immediate next tasks more visible [5].

At the system level, the implication is not simply to reduce alarms. Participants described competing alerts, repeated interruptions, and uncertainty about alarm relevance. Studies of patient-specific alarm customization and intelligent alarm management likewise indicate that alarm burden can be reduced when configurations are adapted to the clinical context while preserving safety [37,38]. Our findings add that effective alarm governance should be aligned with clinical workflows: alarm settings should be patient- and context-sensitive, alarm burden may need periodic review, and responsibility for reviewing, adjusting, and responding to alarms should be explicit across the care team.

Digital support should also respect the temporal boundary of work. Participants described work-group messages and phone notifications as drawing attention back to unstable patients and resuscitations after shifts. This aligns with recovery research identifying psychological detachment as central to recovery from job stress and with evidence linking work-related smartphone use during off-job hours to greater work-life conflict [39,40]. Communication platforms should therefore use role-sensitive escalation pathways and clear expectations for after-hours messaging so that urgent information reaches the appropriate on-duty clinician without routinely extending nonurgent monitoring into recovery time.

Finally, AI-enabled alarm support or wellbeing tools should be designed as assistance rather than another layer of observation. Although participants did not discuss AI directly, their accounts of visible data, scrutiny, and accountability indicate that systems inferring workload, emotional state, or performance may be experienced as surveillance, particularly when their outputs are accessible to managers or linked to appraisal. This concern is consistent with workplace surveillance research showing that perceived surveillance is associated with psychological distress and lower job satisfaction through increased job pressure, reduced autonomy, and privacy concerns [41]. Consistent with health AI ethics and governance guidance, such tools should clearly disclose their purpose, data sources, access arrangements, and intended uses, and should preserve clinicians’ ability to question or override outputs [42]. In this context, wellbeing data should be kept separate from performance management, and nurses should be involved in co-design and evaluation [43]. Evaluation should assess perceived surveillance and staff experience alongside technical performance and patient safety.

Several limitations should be noted. This study was conducted in a single tertiary ICU in China, which may limit transferability to other organizational, technological, and cultural contexts. Participants were recruited through direct invitation, which may have introduced self-selection bias, as nurses who agreed to participate may have been more willing to discuss emotionally sensitive experiences or more reflective about alarm-related work. As with interview-based research, accounts may be shaped by recall bias, social desirability, or recent emotionally salient events, although confidentiality and the absence of supervisory or evaluative relationships likely supported openness. Finally, quotations were translated from Chinese into English, and some linguistic or cultural nuances may have been altered despite efforts to preserve their meaning.

This qualitative interview study suggests that continuous monitoring and frequent alarms shape ICU nurses’ attention, perceived accountability, emotional strain during monitored deterioration, and recovery after shifts. Alarm-intensive care should therefore be approached as a digital health and sociotechnical design issue as well as a patient safety concern. Alarm governance, display design, work-related communication, and future AI-supported monitoring should be co-designed and evaluated for patient safety, alarm reduction, clinicians’ attention and recovery, and perceived surveillance.