This study is reported following the SRQR (Standards for Reporting Qualitative Research) guidelines (Checklist 1) [25]. A qualitative design, using semistructured individual interviews, was selected to explore the subjective and nuanced nature of alert fatigue in depth [26,27].

Eligible participants were practicing junior doctors, also referred to as junior medical officers (JMOs) in Australian hospital systems, defined as interns (postgraduate year [PGY] 1), residents (PGY2), and registrars (PGY3+), who had interacted with alerts in clinical information systems (eg, EHR systems) in hospital settings in Australia. Junior doctors were selected for the study as they are frequent recipients of alerts in hospital settings. A convenience sample of participants was recruited by disseminating details about the study through email distribution lists, posters, and presentations at JMO training sessions, at 2 hospitals within the same hospital network. Additional participants were recruited via snowball sampling through the authors’ professional networks, including word of mouth, LinkedIn posts, and emails to hospital contacts who circulated the study to JMO networks. Prospective participants were encouraged to contact the researchers via email if they were interested in participating.

A semistructured interview guide was developed using insights from existing research on alert and alarm fatigue and discussion among all authors [4,14,30]. Two pilot interviews were conducted by 1 author (NN) with JMO coresearchers via Microsoft Teams, which resulted in minor refinements to the interview guide, such as adding new questions and changing the order of questions to enhance flow. Questions centered around participants’ use of alerts, perceptions and experiences of alert fatigue, factors influencing alert fatigue, the impact of alert fatigue, and recommendations to reduce experiences of alert fatigue in practice (Multimedia Appendix 1). Individual interviews were conducted with participants by one author (NN) using Microsoft Teams, Zoom (Zoom Communications, Inc), or a phone. Three participants were known to the interviewer prior to the interviews through their professional and personal networks.

Interviews were audio recorded, transcribed using Otter AI software (Otter.ai, Inc) [31], and checked for accuracy. Transcription and analysis occurred in parallel to data collection, allowing for further refinement of the interview guide to explore themes in depth as interviews progressed. Interviews continued until data saturation was reached; that is, no new themes emerged in the final interviews, and only new examples or perspectives related to existing themes were identified [32].

Transcripts were imported into NVivo (version 14.24.1; Lumivero) [33] and thematically analyzed using a combined inductive and deductive approach [34]. The first 6 transcripts were independently analyzed by 2 authors (NN and MB or AB or RF) who inductively coded quotes. Themes were deductively mapped to 4 categories: defining alert fatigue, factors influencing experiences of alert fatigue, the impact of alert fatigue, and strategies to reduce alert fatigue. Each pair of authors met to discuss alignment and resolve discrepancies.

Following discussions, 3 of the 4 original categories were restructured under the SEIPS model, where codes related to defining alert fatigue were moved to processes, factors influencing experiences of alert fatigue were moved to the work system, and the impact of alert fatigue was moved to outcomes [24]. Themes related to the process of alert fatigue were further interpreted using the information processing model [28].

Remaining transcripts were analyzed by 1 author (NN) using the SEIPS model, adding new codes and themes as they emerged. Upon completion of the analysis of all interviews, 4 authors (NN, MB, AB, and RF) participated in a workshop to discuss the categorization of codes and themes, which resulted in some codes being renamed, merged, and/or recategorized. The coding framework was then presented to all authors and select participants for member checking, who provided input and agreed upon the final thematic structure.

Ethical approval was received from the Australian Capital Territory Health Low Risk Ethics Committee (protocol: 2024.LRE.00024). All participants provided written informed consent and did not receive compensation for participating in the study. Interview data were deidentified at the point of transcription, with all potentially identifying details removed. Demographic characteristics are presented in aggregate, and no identifiable participant information is included in this study.

A total of 20 interviews were conducted between August 2024 and March 2025. Participants included 9 interns (PGY1), 6 residents (PGY2), and 5 registrars (PGY3+), working at 10 hospital sites, across 3 Australian states (Table 1). Interviews lasted between 18 and 56 minutes, with an average length of 33 (SD 11; range 18-56) minutes. Two interviews were conducted via phone and the remainder via videoconference.

Participants received alerts from EHR, electronic medical record (EMR), and/or electronic medication management systems. All participants were users of commercially available EHR or EMR systems (eg, Epic, Cerner, Sunrise, and MedChart), while 2 participants also used homegrown systems. Alert types included passive and interruptive alerts such as those for medications (eg, allergy, drug-drug interactions, and dosing), and best practice (eg, venous thromboembolism prophylaxis, advanced care directives, and resuscitation plans). Alerts popped up during tasks such as medication ordering, opening patient charts, and opening the EHR system. Alerts outside of standard clinical information systems (“system”) were also described.

In the following sections, we present the clinicians’ perceptions of alert fatigue (process), contributing factors (work system), impacts (outcomes), and potential strategies for addressing alert fatigue. Figure 2 shows a schematic of themes aligned to the SEIPS model.

Factors were coded to all work system domains of the SEIPS. Multimedia Appendix 2 shows the distribution of coded references within each domain.

Strategies to reduce alert fatigue included technical changes such as improving the interface design, personalizing and increasing the relevance of alerts; organizational strategies such as monitoring and evaluation initiatives and educating clinicians about alert fatigue; as well as personal strategies such as improving overall well-being and using workarounds. A full list of strategies is presented in Multimedia Appendix 3.

This study provides a comprehensive exploration of junior doctors’ experiences of alert fatigue in hospital settings. Importantly, alert fatigue was found to manifest at different stages of information processing where participants failed to detect alerts, superficially processed alerts, developed mental models, and used excessive cognitive effort to interpret alerts, which led to different response patterns. These experiences were perceived to impact patient safety, such as increasing the risk of errors and contributing to patient harm, and clinician well-being, such as workflow interruptions and reduced efficiency.

The SEIPS model highlighted complex interactions between factors contributing to experiences of alert fatigue, including alert logic and design, organizational culture, and individual traits and emotional states. We observed junior doctors at the same hospital, who received similar alerts, to experience alert fatigue differently depending on their experience, role, and personality. For example, where junior doctors had positive attitudes and alerts were tightly aligned with responsibilities, even frequently disregarded alerts were not perceived as fatiguing. While poor clinical relevance remains a fundamentally important factor in the development of alert fatigue, our study reveals that doctors may have different levels of tolerance depending on these personal, organizational, and task factors. We also observed feedback loops between factors contributing to alert fatigue and its outcomes. For instance, repeated exposure to low-value alerts could lead to negative attitudes, subsequently reducing trust and influencing attitudes toward future alerts.

In our study, alert fatigue was rarely experienced in an all-or-nothing fashion. Instead, participants often described alert fatigue to arise in specific contexts, such as when under time pressure or for certain alert types. While some participants reported generalized negative perceptions of alerts, none described being universally fatigued or entirely unaffected. Similar to existing studies, we also found alert fatigue to develop for both passive and interruptive alerts [35].

Our findings support the view that alert fatigue is not simply an action, but a mental state that is subjectively experienced at conscious and subconscious levels [16,29]. Participants described 3 distinct information processing pathways in which alert fatigue occurred: one where alert stimuli were not perceived at all, another in which alerts were recognized but superficially processed based on mental shortcuts, and a third where alerts required high cognitive effort to interpret and respond to. The first 2 pathways, in line with existing conceptualizations of alert fatigue, were often associated with alerts being rapidly dismissed [4,12,13]. Despite this, mental shortcuts were perceived as an efficient way to process information and were often not associated with conscious experiences of fatigue. While many participants believed they were able to rapidly interpret alerts, patient safety risks could however still arise due to the potential to misinterpret information, particularly when the alerts lacked visual distinctiveness.

In contrast, the third pathway was consistently described as fatiguing. Participants reported feeling frustrated due to the cognitive burden of alerts and related impacts on efficiency, which arose particularly where alerts were ambiguous, time-consuming, or difficult to action. For certain alert types or individuals, this burden was described to gradually lead to cognitive disengagement, with clinicians eventually defaulting to rapid override as a strategy to avoid further mental effort. This finding mirrors alarm fatigue literature that suggests fatigue develops as a coping mechanism for cognitive burden [30,36]. While burnout is frequently linked to alert fatigue [37], our findings suggest that rapidly overriding alerts may conversely serve as a protective mechanism to manage cognitive load and avoid burnout, consistent with literature linking longer EHR use to increased fatigue and burnout [38,39].

These findings suggest that override rates and response times alone are insufficient measures of alert fatigue. Aligning with prior critiques, our work confirms that frequently overridden alerts are sometimes still attended to and can be perceived as helpful [13,40]. Similarly, although quick dismissals can indicate alert fatigue, they can also represent efficient pattern recognition, while longer interactions could both indicate appropriate engagement with alerts and signal cognitive burden. We therefore recommend supplementing system metrics with clinician self-reports to better understand both conscious and subconscious experiences of fatigue and applying solutions that appropriately address the specific type or types of fatigue experienced.

In line with existing literature, we found exposure to clinically irrelevant, repetitive, and poorly timed alerts contributes to alert fatigue [4,15,21,41,42]. Consistent with quantitative studies, the total volume of system alerts was rarely identified as a cause of alert fatigue alone, but instead became problematic when clinicians received many clinically irrelevant and repetitive alerts, and multiple alerts within a single clinical encounter [4]. We also identified novel influences, including personality traits, mood, social learning, and exposure to external alerts. Many participants described cognitive overload as a result of not just system alerts, but also notifications from pagers, emails, phone calls, and verbal prompts, which cumulatively contributed to experiences of fatigue and dismissal of system alerts. While existing literature has sought to improve total alert burden to reduce alert fatigue [43,44], our study highlights the need to consider system alerts within the broader ecosystem of alerts and tasks that clinicians receive.

Social and organizational factors are underexplored in existing alert fatigue literature but were found to collectively shape experiences in our study. Junior doctors often looked to peers to guide their own responses to alerts and expectations of their roles, affecting not just alert responses but also the framing of alerts as helpful or annoying. Organizational and cultural expectations of efficiency further increased alert fatigue, as managing alerts was often seen as a barrier to achieving other clinical tasks. While alerts are likely to offer the most benefit to junior doctors, our study substantiates prior research that suggests this group experiences more fatigue than their senior counterparts [45]. Participants with more clinical experience had greater confidence and therefore could interpret and respond to alerts with less cognitive effort, whereas those with less experience reported more ambiguity and higher effort expended in interpreting information.

Our results highlighted several strategies for reducing alert fatigue that align with prior literature, including removing ineffective alerts and increasing clinical relevance [46], incorporating contextual awareness [47], applying human factors design principles [48,49], and implementing alerts in line with the 5 rights of CDS (right information, person, format, channel, and time) [6,7,50,51]. Our findings build on this literature by providing insight into how and why strategies are likely to be effective, and how they can be targeted to the different causes and impacts of alert fatigue. Several studies testing these strategies have reported their effectiveness in decreasing the overall volume and increasing the acceptance rate of alerts [50]. However, we argue that future research must also evaluate whether clinicians’ experiences of fatigue are reduced. Given that our results showed junior doctors are likely to experience greater levels of alert fatigue than more senior doctors, we emphasize the importance of including this group in redesign efforts and evaluation of fatigue reduction.

We also identify and present new strategies that could improve experiences of alert fatigue. These include allowing clinicians to personalize alerts, education on alert fatigue, supporting overall emotional well-being, and streamlining the system and other alerts. Due to the individual variability in experiences of alert fatigue, personalization may be particularly valuable in fostering a greater sense of control and responsibility, thus improving engagement with alerts that individuals deem to be relevant. Further research is needed, however, to evaluate the effectiveness of personalization and other strategies in reducing alert fatigue and improving patient safety.

Interestingly, in some cases, experiences of fatigue were not associated with the presence of alerts, but by their absence, where critical information was difficult and time-consuming to locate. This finding underscores the need for balance between over- and underalerting, ensuring alerts effectively address the clinical or safety need. Implementers should carefully consider the issue they aim to target, evaluating whether CDS and alerts specifically are an appropriate solution, or if different strategies or forms of CDS, such as passive alerts or search functionality, should instead be considered [3,6].

Our participants were junior doctors in urban Australian hospitals who primarily used commercial information systems. Findings therefore may not generalize to other clinician groups, specialties, or settings, such as rural hospitals or those using homegrown systems. Additionally, we did not incorporate observational or system usage data, which may limit insights into subconscious behaviors. Future studies should incorporate mixed methods, that is, both clinicians’ experiences of alert fatigue and actual responses to alerts and explore experiences of alert fatigue among different user groups.

To our knowledge, participants in our study used only knowledge-based alerts, limiting conclusions regarding experiences of alert fatigue for smarter alerts, such as those that are artificial intelligence (AI)-based. Our results can, however, offer important considerations. Although AI-based alerts can significantly improve clinical relevance and specificity [52], they may also increase cognitive burden if information is complex or constantly changing. Alternatively, given the fast-paced nature of the setting in which alerts are presented, AI-based alerts may be superficially processed and, due to their high positive predictive value, be automatically accepted rather than rejected due to automation bias [53]. Further research is needed to explore how alert fatigue manifests for smarter alerts in hospital settings and its impact on patient safety.

This study highlights alert fatigue as a dynamic, context-dependent experience that is not easily captured through studying behavior alone. Our findings reinforce the need to move beyond using singular metrics to measure alert fatigue and instead address alert fatigue using tailored strategies that target the specific causes and outcomes involved. While reducing low-value alerts and improving interface design remain important, broader approaches that support clinician well-being, foster positive organizational culture, and consider the impact of competing priorities and tasks may be equally critical.