Patterns of Alarm Fatigue and Their Predictors Among Nurses in Intensive Care Units: A Latent Profile Analysis

ABSTRACT

The clinical alarms from various medical instrument and equipment are becoming a new challenge for staff in intensive care unit workplace. This study revealed the latent profiles of alarm fatigue among intensive care unit (ICU) nurses and to explore their predictors. A cross‐sectional survey was conducted from 2023 to 2024 using the nurses' alarm fatigue questionnaire (NAFQ) to measure the alarm fatigue of 725 ICU nurses in China. The overall alarm fatigue of nurses in ICUs is at a medium‐high level. Three classes of alarm fatigue were identified by latent profile analysis and influenced by multiple factors (i.e., whether working overtime, whether having alarm management protocol in the wards, clinical false alarm experience, and alarm function cognition). This study suggests that administrators provide tailored supports and interventions to reduce alarm fatigue of nurses in ICUs based on the characteristics of their alarm fatigue profiles.

Key Points

• Clinical alarms of monitoring devices in ICU workplaces are one of the most important methods to warn personnel directly of potential threats to patients' health and lives.

• ICU nurses, as the primary caregivers for critically ill patients, play a crucial role in monitoring and promptly responding to diverse alarms, thereby highlighting the growing concern over alarm fatigue among nursing staff in ICUs.

• The use of person‐centered approach for the connotation of ICU nurses' alarm fatigue is less studied and still needs to be further explored.

1. Introduction

Intensive care unit (ICU), an important workplace for treating critically ill patients, houses the hospital's highest density of medical instruments and equipment. Continuous monitoring through these devices is paramount to detect subtle changes and risks in clinical practices (Sowan et al. 2016), particularly for patients with extremely unstable clinical conditions and scarce effective communication abilities. Hence, clinical alarms in ICU workplaces stand as a vital method for directly warning personnel to potential threats to patients' health and lives (Shaban Aysha and Sayed Ahmed 2019).

Studies have reported a high prevalence of clinical alarms (Li et al. 2018; He et al. 2021; Liu, Liu, et al. 2021; Liu, Xu, et al. 2021), with one reporting up to 400 alarms per patient during one nurse shift (Keller Jr. 2012) and another finding that ICU nurses spent 35% of their worktime responding to alarms (Bitan et al. 2004), respectively. While these visual and acoustic alarms promptly alert healthcare staff to clinical changes and any devices failure (Lewandowska et al. 2020), 72% to 99% of them are false alarms (Simpson and Lyndon 2019) and can't exactly report worsening conditions for patients (Ruppel et al. 2018; Muroi et al. 2020). Moreover, only 5%–15% of alarms are needed clinical response or intervention in the monitoring systems (Paine et al. 2016). Medical staff are particularly prone to fatigue from repeatedly responding to unfiltered, insignificant and/or false alarms in long time (Fernandes et al. 2020), as shown in a study published in 2017 (Petersen and Costanzo 2017).

Alarm fatigue, a desensitized state among healthcare personnel due to excessive sensory overexposure from a great deal of alarm sounds over time (Harris et al. 2017; Akturan et al. 2022), is associated with reduced safety and quality of health care, increased risk of adverse outcomes and even death of patients (Simpson and Lyndon 2019; Nyarko et al. 2023; Carelli et al. 2022). The desensitization contributes to nurses delaying or lack of response to clinical monitoring alarms (Cvach 2012; Leigher et al. 2020; Lewandowska et al. 2020; Purbaugh 2014). And this phenomenon is significantly associated with medical errors tendencies, with one study finding the error tendency increase by 0.263 unit for every unit increase in alarm fatigue level (Gündoğan and Erdağı Oral 2023). In addition, alarm fatigue adversely affects physical and mental health among nurses (Storm and Chen 2021; Ding et al. 2023; Nyarko et al. 2023; Lewandowska et al. 2020), as exacerbated during the Coronavirus Disease 2019 (COVID‐19) pandemic (Akturan et al. 2022). Alarm fatigue is a problem of growing global attention, and alarm hazards are included in the Emergency Care Research Institute (ECRI)'s annual list of top hazards related to medical technology for many years (Lewandowska et al. 2020). Furthermore, its management has been designated an international goal for patient safety by the accreditation organization for excellence joint commission (2022).

ICU nurses, as the primary caregivers for critically ill patients, play a crucial role in monitoring and promptly responding to diverse alarms, thereby highlighting the growing concern over alarm fatigue among them. It is of great importance to further research on this topic. Previous studies on the phenomenon of alarm fatigue have investigated using traditional correlation or regression methods which focus on analyzing the relationships of alarm fatigue with other variables (e.g., metal health, safety and ethic; Seok et al. 2023; Gündoğan and Erdağı Oral 2023; Storm and Chen 2021; Asadi et al. 2022). Despite such approaches insight into valuable information about the direct and unique association of alarm fatigue with other variables, they tend to ignore the differences among individuals and are limited in their ability to reflect the multifaceted nature of alarm fatigue and intervene. Latent profile analysis (LPA) is a person‐centered method that categorizes individuals with similar patterns of personal and professional characteristics, traits, or behavior into profiles according to their responses (Teng et al. 2024), which has been applied in psychology, education, management, marketing, medical and health research. Compared to traditional statistical methods, person‐centered approaches allow for a deeper, multifaceted understanding of the relationships between variables and facilitate the classification of heterogeneous characteristics within this population. This, in turn, helps guide the development of tailored interventions. Therefore, it can shed new light on nurses' perceptions of their work characteristics and health outcomes. Moreover, it would be highly interesting to explore what latent patterns of alarm fatigue can be observed among ICU nurses and which specific characteristics enhance or impair their alarm fatigue. Knowledge in this area remains limited and warrants further research. The main hypotheses of our research were as follows: (a) the alarm fatigue of nurses in ICUs is medium; (b) the distinct profiles of alarm fatigue exist in the subjects; and (c) there are differences of the reaction characteristics of the population in different items. Testing the three hypotheses will increase the understanding of the alarm fatigue among ICU nurses and permit more tailored guidance for developing interventions to promote the health of nurses and improve the safety of patients.

2. Methodology

2.1. Aims

This study aimed to explore the distinct profiles of alarm fatigue among ICU nurses using the LPA method, that is, whether the sample of nurses in ICUs was heterogeneous in terms of alarm fatigue. We also examined the relationships between the profiles and sociodemographic and management‐related variables to enrich the current knowledge on the predictors of alarm fatigue.

2.2. Design

A cross‐sectional self‐report study was conducted in this study.

2.3. Sample/Participants

According to the related literature, one entry corresponds to a sample size of 5–10 cases, and 15%–20% of the loss rate in the sample and the inefficiency of the questionnaire are considered. In total, 725 nurses were recruited conveniently from 36 hospital sites (86 ICUs) in China. The inclusion criteria for ICU nurses were nurses that: (1) had at least 1 year of working experience in ICU; (2) agreed to participate in this study. Nurses on leave (sick leave, maternity leave, or holiday), healthcare assistants, and other non‐nursing staff were excluded.

2.4. Ethical Approvals and Ethical Description

Written approval was granted by the Ethics Committee of Shanghai General Hospital (no. 2024KS059).

2.5. Data Collection

With the permission of nursing administrators, we recruited hospitals and intensive care units between December 2023 and February 2024. A manager (e.g., supervisor of nursing care, ward head nurse) was appointed to be the single point of contact with the research team in each hospital, who was given additional explanation about the data collection process or could ask for more details about the study. All nurses received a quick response (QR) code generated an online questionnaire platform (Wenjuanxing, www.wjx.cn) by WeChat from the manager with information about the study and an invitation to participate. This invitation included a fully anonymized link (QR‐code). The respondents voluntarily agreed to take part in the study and filled out the questionnaire. And the questionnaires were excluded were as follows: (1) questionnaires with a filling time of less than 2 min; (2) the inconsistent questionnaire with personal information; and (3) the answer regularity (consistent or wavy) scale. The participants did not receive any incentive or payment for their participation in the study. To further improve the quality of the study, we set each internet protocol (IP) address to only fill out the questionnaire once, and the respondent could submit questionnaire successfully when all options were completed.

2.6. Measures

The self‐report instruments in this study included the following information:

2.7. Personal Characteristics Form

This is a form including two parts that: (1) demographic variables, such as age, gender, education and qualification level, years of expertise within the working/ICU, type of ICU, shift length, and so on; and (2) information on one nurse's staff, such as the habit of setting alarms for medical devices, training related to the use of medical devices in the ward, and so on.

2.8. The Nurses' Alarm Fatigue Questionnaire (NAFQ)

The validity and reliability of 13‐item questionnaire, which was developed (Torabizadeh et al. 2017) to measure the alarm fatigue of ICU nurses. This questionnaire is scored on a 5‐point Likert scale (0 = always, 4 = never), except for items 1 and 9, which are scored the other way round. The total score of the questionnaire ranges from 8 (lowest impact of fatigue) to 44 (highest impact of fatigue), where the higher score indicated a greater impact of alarm fatigue on nurses' performance. According to Brislin's translation model (Brislin 1970), the NAFQ was translated and cross‐cultural adapted by Liu, Liu, et al. (2021), and the validity of the questionnaire using content validity and the reliability applying Cronbach's alpha coefficients (0.771) were confirmed in Chinese ICU nurses. The Cronbach's alpha was 0.807 in this study.

2.9. Nurses' Recognition of Medical Device Alarms Questionnaire (NR‐MDAQ)

The original version of the NR‐MDAQ was developed by the Healthcare Technology Foundation (HTF) (2006) and modified and validated by Cho et al. (2016) in Korean ICU nurses. This is a 23‐item questionnaire covering nurses' recognition of clinical alarms and its obstacles to clinical alarm management. It encompasses three different dimensions of nurses' recognition to clinical alarms subscale with 14 items, such as alarm parameter setting, clinical false alarm experience, and alarm function cognition. Participants respond to these items on an ordinal score (1 = strongly disagree, 5 = strongly agree). Responses regarding the obstacles to clinical alarm management subscale with 9 items are rated on a negative 9‐point Likert scale (1 = strongly important, 9 = strongly unimportant). The Chinese version of the NR‐MDAQ was translated and validated by Wang and Zheng (2018). Cronbach's alpha for the scale was 0.920, and it was widely used to measure the NR‐MDAQ of Chinese ICU nurses (Wang and Zheng 2018; Liu et al. 2022). The Cronbach's alpha of each subscale in our study was 0.824 and 0.931, respectively.

2.10. Data Analysis

LPA was conducted to identify the alarm fatigue profiles of ICU nurses by Mplus version 8.3. Models were tested with one class initially and additional classes added incrementally until a unique solution could be determined using robust maximum likelihood estimation (Muthén and Muthén 2012). Based on the guidelines for fit indices in the literature, we adopt a combination of informational criteria including Akaike information criterion (AIC), Bayesian information criterion (BIC), and sample size‐adjusted BIC (aBIC), in which lower values indicated superior fit and likelihood ratio tests (Lo–Mendell–Rubin, LMR, and bootstrap likelihood ratio test, BLRT) which compared the fit between a k‐class solution with a k‐1‐class solution. A significant LMR or BLRT (with p < 0.05) would indicate that the more complex model (k‐class model) outperformed the simpler model (k‐1 class model) with the increased model fit. The relative entropy was reported to evaluate the classification accuracy of participants, which was estimated ranging from 0.0 to 1.0, with higher values indicating greater accuracy. Values of entropy above 0.80 indicated a good profile solution.

Descriptive statistics were conducted for all variables by SPSS (version 19.0). For between‐class comparisons, we reported the standardized mean differences (SMDs) as the measure of effect size instead of relying on p‐values from traditional hypothesis tests (e.g., t‐tests and chi‐square tests) (Austin 2011). This approach was adopted to mitigate the inflation of Type I error associated with multiple comparisons and to shift the focus from statistical significance to the magnitude of the observed differences, which aligns with contemporary methodological recommendations for observational studies (Austin and Stuart 2015; Franklin and Schneeweiss 2017). And SMDs were calculated using an Excel template based on the analysis of variance (ANOVA) method for continuous variables and the distribution for categorical variables. An SMD of less than 0.10 was considered indicative of a negligible difference between groups, suggesting a good balance (Austin 2011). Multiple logistic regression was used to analyze the predictors of potential profiles. The selection of variables for inclusion in the multivariable model was based on a priori clinical knowledge, existing literature, and the theoretical framework of this study, rather than on the statistical significance of univariate analyses (Sauerbrei et al. 2020). Results are presented as adjusted odds ratios (aORs) with their 95% confidence intervals (CIs) and corresponding adjusted P values (adj P), which serve as the primary effect size measure to illustrate the direction, strength, and precision of the associations (Lang and Altman 2015). The differences were statistically significant at p < 0.05.

3. Results

3.1. Participant Characteristics

In total, 809 nurses participated in this study, and the number of valid responses was 725. The mean age of participants was 32.37 ± 6.35 years, and 88.6% of the sample were female. The majority had an undergraduate education in Nursing (77.8%) and junior professional title (74.1%), worked 8 h shifts (71.0%), and had experience of working overtime (81.2%) (Table S1). Themean length of professional experience was 10.42 ± 7.01 years, and the experience in ICU was 7.80 ± 5.77 years. Among the 725 intensive care nurses, the score of alarm fatigue was 26.89 ± 6.71. The score of obstacles to clinical alarm management was 44.67 ± 16.73. And three‐dimensional scores for Nurses' recognition of clinical alarms was 16.30 ± 2.54, 22.03 ± 6.81, and 13.41 ± 2.04, respectively (Table S2).

3.2. Latent Profile Analysis (LPA)

To identify the best model, we separately estimated six models in this study and displayed the fit statistic results of each model in Table 1. The 3‐profile model showed the best fit. Although the AIC, BIC, and aBIC values continued to decline, they tended to stabilize after the 3‐profile model. Meanwhile, the LMR test (p < 0.001) and BLRT (p < 0.001) indicated that the 3‐profile model was statistically significant. This model also had the highest entropy (0.985), and the proportion of members in the smallest subtype was 22.2% (larger than the 5% threshold).

TABLE 1.

Table fit statistics for profile structures (n = 725).

Number of profiles	LL	FP	AIC	BIC	aBIC	LMR (p)	BLRT (p)	Entropy	Smallest class%	Probability
1	−14631.145	26	29314.291	29433.531	29350.973	NA	NA	NA	NA	NA
2	−14104.785	40	28289.571	28473.018	28346.006	0.0000	0.0000	0.959	25.24%	0.25241/0.74759
3	−13467.589	54	27043.178	27290.831	27119.365	0.0000	0.0000	0.985	22.21%	0.22207/0.37793/0.40000
4	−13308.325	68	26752.650	27064.509	26848.589	0.0025	0.0026	0.974	9.66%	0.09655/0.40000/0.12690/0.37655
5	−13118.059	82	26400.118	26776.184	26515.810	0.0001	0.0002	0.977	9.79%	0.11724/0.14483/0.09793/0.19448/0.44552
6	−13035.819	96	26263.638	26703.910	26399.082	0.3524	0.3516	0.969	4.83%	0.11862/0.09655/0.04828/0.44552/0.15724/0.13379

Abbreviations: AIC, Akaike information criteria; aBIC, sample‐size‐adjusted BIC; BIC, Bayesian information criteria; BLRT, bootstrapped likelihood ratio test; FP, free parameters; LL, log‐likelihood; LMR, Lo–Mendell–Rubin.

Figure 1 shows three latent profiles identified by the LPA. Profile 1 was the group with low alarm fatigue (22.2%), showing relatively low incorrect alarm behaviors and negative attitude. Members of Profile 2 (37.8%) had an average of alarm fatigue combined with medium negative alarm behaviors and attitude (“medium alarm fatigue”). Profile 3 (40.0%) grouped staff with high incorrect alarm behaviors and negative attitude (“high alarm fatigue”).

FIGURE 1.

Latent profile analysis result.

3.3. Comparison of Demographic and Management‐Related Variables Between the Three Profiles

Table 2 presented the sociodemographic characteristics and NR‐MDAQ of the three latent profiles. Overall, gender, whether having overtime experience, whether having management protocols for medical device alarms, alarm parameter setting, clinical false alarm experience, and alarm function cognition showed imbalance between profiles (SMD > 0.10). In contrast, demographic factors such as age, work experience, educational level, professional title, and other relevant variables showed negligible differences between profiles (SMD < 0.10).

TABLE 2.

Comparison of alarm fatigue of the profiles based on sociodemographic characteristics and NR‐MDAQ (n = 725).

Variables		Profile 1 (n = 161)	Profile 2 (n = 274)	Profile 3 (n = 290)	SMD
Age, years (n, %)	20–30	72 (44.7%)	111 (40.5%)	124 (42.8%)	*
	31–40	71 (44.1%)	139 (50.7%)	133 (45.9%)	0.055
	41+	18 (11.2%)	24 (8.8%)	33 (11.4%)	0.040
Gender (n, %)	Female	140 (87.0%)	254 (92.7%)	248 (85.5%)	0.103
	Male	21 (13.0%)	20 (7.3%)	42 (14.5%)
Professional seniority, years (n, %)	1–5	47 (29.2%)	74 (27.0%)	90 (31.0%)	*
	6–10	46 (28.6%)	83 (30.3%)	76 (26.2%)	0.040
	11+	68 (42.2%)	117 (42.7%)	124 (42.8%)	0.005
ICU seniority, years (n, %)	1–5	65 (40.4%)	110 (40.1%)	132 (45.5%)	*
	6–10	49 (30.4%)	86 (31.4%)	77 (26.6%)	0.049
	11+	47 (29.2%)	78 (28.5%)	81 (27.9%)	0.011
Education level (n, %)	College education	39 (24.2%)	50 (18.2%)	60 (20.7%)	*
	Undergraduate education	120 (74.5%)	220 (80.3%)	224 (77.2%)	0.053
	Postgraduate education	2 (1.2%)	4 (1.5%)	6 (2.1%)	0.029
Professional title (n, %)	Junior	118 (73.3%)	203 (74.1%)	216 (74.5%)	*
	Intermediate	42 (26.1%)	65 (23.7%)	66 (22.8%)	0.029
	Senior	1 (0.6%)	6 (2.2%)	8 (2.8%)	0.059
Shift length (n, %)	8 h shifts	121 (75.2%)	189 (69.0%)	205 (70.7%)	0.052
	12 h shifts	40 (24.8%)	85 (31.0%)	85 (29.3%)
Shifts nurses (n, %)	Yes	57 (35.4%)	123 (44.9%)	109 (37.6%)	0.082
	No	104 (64.6%)	151 (55.1%)	181 (62.4%)
Experience of working overtime (n, %)	Yes	119 (73.9%)	219 (79.9%)	251 (86.6%)	0.125
	No	42 (26.1%)	55 (20.1%)	39 (13.4%)
The habit of setting alarms for medical devices (n, %)	Yes	149 (92.5%)	255 (93.1%)	262 (90.3%)	0.046
	No	12 (7.5%)	19 (6.9%)	28 (9.7%)
The ward has management protocol for medical device alarms (n, %)	Yes	141 (87.6%)	218 (79.6%)	210 (72.4%)	0.141
	No	20 (12.4%)	56 (20.4%)	80 (27.6%)
The ward has alarm management system for monitoring devices (n, %)	Yes	134 (83.2%)	217 (79.2%)	217 (74.8%)	0.079
	No	27 (16.8%)	57 (20.8%)	73 (25.2%)
Experience of received professional training related ventilators (n, %)	Yes	129 (80.1%)	215 (78.5%)	230 (79.3%)	0.016
	No	32 (19.9%)	59 (21.5%)	60 (20.7%)
Experience of received training in medical device alarm management (n, %)	Yes	145 (90.1%)	231 (84.3%)	244 (84.1%)	0.069
	No	16 (9.9%)	43 (15.7%)	46 (15.9%)
Insight of concept related alarm fatigue (n, %)	Yes	96 (59.6%)	139 (50.7%)	169 (58.3%)	0.079
	No	65 (40.4%)	135 (49.3%)	121 (41.7%)
ICU type (n, %)	GICU	51 (31.7%)	86 (31.4%)	90 (31.0%)	*
	SICU	58 (36.0%)	93 (33.9%)	89 (30.7%)	0.045
	MICU	27 (16.8%)	49 (17.9%)	54 (18.6%)	0.018
	EICU	16 (9.9%)	31 (11.3%)	37 (12.8%)	0.035
	Other	9 (5.6%)	15 (5.5%)	20 (6.9%)	0.028
Nurse‐to‐patient ratio (n, %)	1: ≤ 2	30 (18.6%)	46 (16.8%)	38 (13.1%)	*
	1: 3	45 (28.0%)	86 (31.4%)	85 (29.3%)	0.029
	1: 3+	86 (53.4%)	142 (51.8%)	167 (57.6%)	0.053
Alarm Parameter Setting, scores (Mean ± SD)		16.81 ± 2.69	16.49 ± 2.23	16.20 ± 2.71	0.117
Clinical False Alarm Experience, scores (Mean ± SD)		20.01 ± 8.16	20.91 ± 6.27	24.21 ± 5.82	0.330
Alarm Function Cognition, scores (Mean ± SD)		13.63 ± 2.17	13.82 ± 1.74	12.89 ± 2.11	0.280
Obstacles to Clinical Alarm Management, scores (Mean ± SD)		44.13 ± 20.89	45.32 ± 15.67	44.36 ± 15.05	0.037

Note: Between‐group differences were assessed using the standardized mean difference (SMD). An SMD < 0.100 was considered indicative of good balance. For multi‐categorical variables, SMDs were calculated separately for each non‐reference category against the reference category, which is marked with an asterisk (*).

3.4. Predictors of Latent Profile Membership

Consistent with the pre‐specified analytic plan, the final multinomial logistic regression model adjusted for priori variables (e.g., age, work experience, educational level, professional title, shift) and environmental variables (e.g., ICU type, nurse‐to‐patient ratio). Moreover, additional variables that demonstrated meaningful imbalance (SMD > 0.10) in baseline comparisons in the subsequent multivariable model to examine the predictors of these factors with profile membership.

The multinomial logistic regression analysis results were shown in Table 3. It can be seen that experience of working overtime, management protocol for medical device alarms, clinical false alarm experience and alarm function cognition independently impacted profile membership. In terms of clinical false alarm experience, ICU nurses with higher scores were more likely to be placed in profile 3 (all p < 0.05). Comparing between profile 1 and 3, ICU nurses with no overtime experience were more likely to belong to profile 1, whereas those working in wards without alarms management protocol were more likely to belong to profile 3. In addition, lower alarm function cognition was also associated with a higher likelihood of being in profile 3. However, there was no tendency to profile membership for comparisons between profile 2 and 3 in terms of overtime experience and alarms management protocol.

TABLE 3.

Multinomial logistic regression analysis of predictors of the latent profile memberships (n = 725).

Predictors	Profile 1 vs. Profile 3				Profile 2 vs. Profile 3
	B	aOR	95% CI	adj p	B	aOR	95% CI	adj p
Age (ref: ≥ 41 years)
20–30 years	0.705	2.024	0.653–6.268	0.222	0.563	1.756	0.663–4.654	0.258
31–40 years	0.341	1.406	0.623–3.173	0.411	0.574	1.775	0.869–3.626	0.115
Gender (ref: male)	−0.146	0.864	0.447–1.671	0.664	0.556	1.744	0.936–3.252	0.080
Professional seniority (ref: ≥ 11 years)
1–5 years	−0.092	0.912	0.299–2.788	0.872	0.375	1.455	0.579–3.655	0.425
6–10 years	0.002	1.002	0.450–2.233	0.995	0.326	1.386	0.720–2.669	0.329
ICU seniority (ref: ≥ 11 years)
1–5 years	−0.540	0.583	0.243–1.400	0.227	−0.467	0.627	0.306–1.284	0.202
6–10 years	−0.081	1.002	0.450–2.233	0.995	−0.155	0.857	0.467–1.572	0.617
Education level (ref: postgraduate)
College education	0.339	1.404	0.226–8.719	0.716	−0.012	0.988	0.222–4.397	0.988
Undergraduate education	0.240	1.272	0.217–7.445	0.790	0.116	1.123	0.270–4.677	0.873
Professional title (ref: senior)
Junior	1.441	4.227	0.448–39.915	0.208	0.101	1.106	0.318–3.846	0.874
Intermediate	1.942	6.971	0.768–63.293	0.085	0.462	1.588	0.472–5.343	0.455
Shift nurse (ref: yes)	0.287	1.333	0.848–2.095	0.213	−0.203	0.816	0.559–1.193	0.295
Experience of working overtime (ref: yes)	0.641	1.899	1.109–3.253	0.020	0.243	1.275	0.780–2.082	0.332
The ward has management protocol for medical device alarms (ref: yes)	−0.750	0.472	0.268–0.831	0.009	−0.220	0.803	0.526–1.225	0.309
Alarm Parameter Setting	0.055	1.056	0.920–1.213	0.438	0.052	1.053	0.933–1.189	0.403
Clinical False Alarm Experience	−0.117	0.890	0.854–0.926	< 0.001	−0.920	0.912	0.880–0.945	< 0.001
Alarm Function Cognition	0.116	1.123	0.962–1.312	0.143	0.201	1.222	1.066–1.401	0.004

Note: Profile 1: low alarm fatigue, Profile 2: medium alarm fatigue, Profile 3: high alarm fatigue (reference subtype); 95% CI: 95% Confidence Interval; aOR: adjusted Odds Ratio, adj P: adjusted p value, odds ratio and p values were adjusted for the potential environmental confounders (e.g., ICU type and nurse‐to‐patient ratio) in the multivariable logistic regression model. Ref: reference.

4. Discussion

This study was conducted to establish whether latent profiles of alarm fatigue among ICU nurses, and if so, to examine sociodemographic and alarm management‐related predictors of profile membership. To our knowledge, there has been no such study in the literature. Hence, the findings were discussed with the relevant literature. In this study, the total mean scores of alarm fatigue in ICU nurses were 26.88, indicating that their alarm fatigue is at a moderate or even high level, which was consistent with findings from other studies (Bourji et al. 2020; Carelli et al. 2022; Seok et al. 2023; Lewandowska et al. 2023). Fatigue might be a stressful aspect of ICU nurses' profession owing to exposure highly to the clinical alarms (Carelli et al. 2022). Attention of this phenomenon, as well as improvement of its knowledge, may benefit to the health service, especially relating to patient safety.

Three latent profiles were identified based on the model fit indicators and interpretability, namely, “low alarm fatigue” (Profile 1, 22.2%), “medium alarm fatigue” (Profile 2, 37.8%), and “high alarm fatigue” (Profile 3, 40.0%). The average scores of each profile were 18.85, 27.07, and 31.17, respectively. This categorization proved the heterogeneity of ICU nurses' alarm fatigue in every latent profile, complementing previous studies that treated ICU nurses as a homogeneous whole and providing guidance for developing targeted interventions in further research to reduce their alarm fatigue.

The members of Profile 3 had largest proportion of working overtime experience among all the profiles. Despite little previous research involving overtime experience with alarm fatigue exists in the literature, this result sheds some light on the role of overtime in alarm fatigue. Unsurprisingly, ICU nurses tended to work more overtime due to an overload stemming from their higher responsibility and more care demands (Lewis and Oster 2019), leading to increase work‐related stressors and job burnouts. Compared with other nursing staffs, nurses with heavy workloads were more likely to experience alarm fatigue, as evidenced in previous studies (Lewandowska et al. 2020; Nyarko et al. 2024). In present study, Profile 3 was characterized by a higher proportion of 12 h shifts and nurse‐to‐patient ratio up to 1:3 than Profile 1 and 2 (see Table 2), indicating that they practice under conditions associated with a demanding workload. Length of shifts and nurse‐to‐patient ratio would directly lead to greater alarm fatigue, which was consistent with earlier studies based on the variable‐centered approach (Bonafide et al. 2017; Claudio et al. 2021; Storm and Chen 2021; Lewandowska et al. 2023; Nyarko et al. 2024). Overtime undermined nurses' capacity for timely and accurate clinical alarm response through physical and cognitive strain (Bonafide et al. 2017; Storm and Chen 2021). This finding suggests that hospitals should address unit environment‐induced alarm fatigue and avoid overloading ICU nurses to ensure their physical readiness for decision‐making to address clinical monitor alarms. And the results guide managers' efforts towards supporting personal characteristics and providing resources to address factors that might impact patient care among ICU nurses. It is possible that nurses working overtime in ICUs with conditions of 12 h shifts and nurse‐to‐patient ratio up to 1:3 may need more resources and mentoring for preventing alarm fatigue. Apart from that, further research on overtime hours and alarm fatigue degree is highly recommended.

What is more, members of Profile 3 had fewer having alarm management protocols compared with Profile 1 and 2. There were also some other studies in the literature reporting protocols on alarm management was pivotal to minimizing alarm fatigue (Bourji et al. 2020; Sliman et al. 2020; Nyarko et al. 2024). There was information in the literature that optimizing workflow and developing protocols for alarm management had a positive effect on improving alarm fatigue (Kim and Kim 2021; Varisco et al. 2021; Gorisek et al. 2021; Sliman et al. 2020). Therefore, it is imperative for each hospital to adopt and implement specific policies and procedures regarding clinical alarms to reduce alarm fatigue. It is crucial to develop alarm management standards to assist nurses in coping with alarm fatigue (Alkubati et al. 2024). Following this, it is of the utmost importance to optimize alarm management and corresponding training mechanisms and needs further research, including standardization of protocols, the definition of guidelines for alarm handling and its workflow. Besides that, a culture of alarm safety management could be established in which ICU nurses feel free to express or release their alarm fatigue concerns. In our study, the prevalence of nursing staff on the insight of concept related to alarm fatigue was less than 60% (see Table 2). It is prerequisite to strengthen the awareness of ICU nurses on the consequences of alarm fatigue and prevention methods.

In addition, members of Profile 3 had higher level of clinical false alarm experience than those in Profile 1 and 2. The evidence on the role of false alarm experience in alarm fatigue among ICU nurses had been certain (Petersen and Costanzo 2017; Fernandes et al. 2020; Gündoğan and Erdağı Oral 2023; Jeong and Kim 2023; Lewandowska et al. 2023). False alarm was a widespread phenomenon (Ruppel et al. 2018; Simpson and Lyndon 2019; Muroi et al. 2020). Excessive exposure to false alarm in long time can contribute to alarm fatigue in medical staff (Petersen and Costanzo 2017; Fernandes et al. 2020). In our study, the average scores of clinical false alarm experience in Profile 3 were 24.21 (see Table 2), which were higher than the levels reported in domestic studies (Wang and Zheng 2018; Liu et al. 2022). As we known, the clinical alarms were counted automatically via setting a range of upper and lower limits in systems. Literature suggested that the vast majority of false alarms originated from technical problems and differences in categorization of alarms. To reduce alarm fatigue, it is imperative that hospitals should address alarm management techniques, including reassessing and standardizing their alarm rules and procedures, using softer alarms and prioritizing alarms. Moreover, false alarms were associated with the application of a uniform alarm range to every patient. To solve alarm fatigue, artificial intelligence technologies would be used to accurately assess patients' current conditions based on big data and automatically calculate and set personalized alarm parameters in the future.

Interestingly, members of Profile1 had alarm function cognition scores that were higher than those of Profile3 but lower than those of Profile 2. This indicated that ICU nurses with lower alarm function cognition were more prone to being classified into the high alarm fatigue group. Alarm function cognition might significantly affect ICU nurses' abilities of alarm management, including setting alarm range and parameters that accurately reflected patients' present conditions, setting or handling alarms properly, controlling and responding to alarms effectively. Its deficiency led to a higher occurrence of false alarms and incorrect alarm behaviors. In turn, alarm fatigue developed and might be further aggravated among ICU nurses. This result suggested that improving alarm function cognition was an important factor in minimizing alarm fatigue among ICU staff, which could be achieved through education and training in alarm management. Studies confirmed that alarm fatigue was significantly lower in nurses who were trained in alarm management (Lewandowska et al. 2023; Asadi et al. 2022; Nyarko et al. 2023; Shaban Aysha and Sayed Ahmed 2019; Bi et al. 2020). It is therefore imperative that nurses' leaders should develop a targeted, evidence‐based, and sustainable training mechanism and strategies regarding alarm management. They are beneficial for reducing alarms and alarm fatigue without any adverse effects.

4.1. Limitations

This study had certain considerable strengths: identifying various constellations of alarm fatigue and their sociodemographic and alarm management‐related predictors among ICU nurses, collecting a relatively large sample with an adequate response rate (Timmins et al. 2023), and using a standard, widely known and specificity questionnaire to assess alarm fatigue. Despite these strengths, the present research also has several limitations that should be discussed.

First, owing to the cross‐sectional nature of the design, we can only identify associations between the different profiles but cannot draw definite conclusions on the directionality correlation between the identified constructs and covariates. But for all this, the examined predictors of profile membership are well‐grounded in variable‐centered studies regarding alarm fatigue. Second, convenience sampling in this study may cause potentially biased estimates. Random sampling should be performed in further research. And the data were collected exclusively from ICUs in China. Therefore, the findings may not be directly applicable to other countries with vastly different healthcare infrastructures, staffing models, cultural contexts, and socioeconomic conditions. Future research should aim to replicate our analysis in multinational settings to determine the cross‐cultural validity and boundaries of our proposed latent profiles. Third, no outcomes of latent profile membership were considered in the present study. Future studies may examine potential consequences of profile membership (such as mental workload, quality of life, burnout, compassion fatigue, or satisfaction) and determine the potential influencing factors of mindful self‐care. Fourth, despite certain advantages of using NAFQ to measure alarm fatigue, the results are self‐reported, subjective, and may be affected by participants' mental states. Further studies may be needed to explore additional, more objective measures (e.g., heart rate, blood pressure, sleep, and other physiological indicators).

5. Implications for Nursing and Health Policy

Considering the high prevalence of alarm fatigue in ICU nurses, concerning alarm fatigue in this group is highly needed. Alarm fatigue is a risk factor for nurse health and patient safety (Carelli et al. 2022; Nyarko et al. 2023). In this study, we utilize a person‐centered approach for the identification of the structure of alarm fatigue profiles in nursing practice. We contend these findings offer significant contributions beyond the results of previous studies, which have largely been constrained by a variable‐centered paradigm in the separation perspective. The distinct alarm fatigue profiles from LPA pave the way for precisely targeted strategies better tailored to the needs of nurses than “one size fits all,” for example, to investigate for whom and in what way interventions are beneficial to reduce alarm fatigue. For ICU nurses with high alarm fatigue, nursing managers and researchers should prioritize targeted interventions aimed at reducing overtime, minimizing false alarms, and enhancing training efficacy. By addressing these factors, critical care nurses can create a more efficient work environment while improving patient outcomes. Therefore, this study contributes to the literature by offering insights into alarm fatigue in ICU nurses, which can guide interventions and guidelines for minimizing alarm fatigue.

6. Conclusion

This research identified the subgroup characteristics and predictors of ICU nurses' alarm fatigue through LPA. We found three qualitatively distinct profiles in terms of alarm fatigue. Meanwhile, we revealed potential predictors of profile membership include whether working overtime, whether having an alarm management protocol in the wards, clinical false alarm experience, and alarm function cognition. These findings suggest that nursing administrators develop targeted interventions based on the heterogeneity of nurses' alarm fatigue in future critical care practices, such as reducing environment‐induced alarm fatigue, optimizing alarm management and corresponding training mechanisms, and establishing cultures of alarm safety and innovating alarm management technologies. Further research should be conducted with qualitative investigations and large sample longitudinal studies to understand the reasons behind these differences. To build a robust evidence base, future efforts should employ multi‐center randomized controlled intervention trials to confirm efficacy and generalizability across healthcare systems.

Author Contributions

Study design: F.Y., F.F.; data collection: F.Y., R.T., Q.Y., Q.Q.; data analysis: F.Y., R.T.; study supervision: F.Y., Y.P., QL.W., F.F.; manuscript writing: F.Y., R.T.; critical revisions for important intellectual content: QL.W., F.F.

Funding

This research is supported by scientific research fund of Shanghai Municipal Health Commission of China (no. 202140047 and no. 202340260), nursing development program of Shanghai Jiao Tong University School of Medicine (no. SJHUHLXK2021), and characteristic research project of Shanghai General Hospital (no. CCTR‐2022N03).

Conflicts of Interest

The authors declare no conflicts of interest.

Supporting information

Table S1: The characteristics of intensive care unit nurses.

Table S2: The distribution of mean alarm fatigue and NR‐MDAQ scores of intensive care unit nurses.

Data S2: STROBE Statement—Checklist of items that should be included in reports of cross‐sectional studies.

Yang, F. , Tai R., Yu Q., et al. 2026. “Patterns of Alarm Fatigue and Their Predictors Among Nurses in Intensive Care Units: A Latent Profile Analysis.” Nursing & Health Sciences 28, no. 1: e70304. 10.1111/nhs.70304.

Data Availability Statement

Data sharing not applicable to this article as no datasets were generated or analysed during the current study.