Examining the Utility of Perceptual Noise Categorization in Pediatric and Neonatal Hospital Units

Abstract

Medical and nursing staff working in hospitals often experience exposure to extreme sound environments, and there is growing evidence of the negative impacts. Previous research highlighted various complexities regarding noise sources in hospitals; however, identifications of intrinsic noise categories that can reveal the complex mixture of existing hospital noise is still limited. The objective of this work was to identify intrinsic categories of the noise sources based on staff perceived annoyance and explore clear associations of these categorized noise sources with psychological perceptions. The staff perceptual responses regarding hospital noise were assessed by conducting surveys at the three pediatric and neonatal care units in two hospitals. Using principle component analysis (PCA), the psychological annoyance responses of 94 participants were used to derive the inherent structural patterns of the existing noise sources. The derived PCA categorization was validated on mixed-model analysis of variances, and employed on regression models to explore potential associations between the categorized noise factors and the staff’s psychological perceptions. The results highlighted three intrinsic noise categories and their negative impacts on staff’s psychological perceptions including work/rest disturbance and noisiness. Taken as a whole, the findings better reveal problematic noise source categories and establish a framework for hospital noise control that is less source-specific and more broadly generalizable.

Keywords

perceptual annoyance noise source factor analysis regression model healthcare design

Medical and nursing staff working in hospitals often experience exposure to excessive sound environments. The consequences of this exposure can manifest in various ways. Previous studies have revealed, for example, the impacts of the hospital sound environments on several staff outcomes, including increase in staff stress, disturbance, annoyance, and anxiety (Arnold & Kornadt, 2011, Morrison et al., 2003; Okcu et al., 2011, Ryherd et al., 2008). However, there is still a limited number of studies that have identified unique contributions of specific noise categories to negative staff outcomes.

This study explores the inherent categorizations of the various noise sources existing in healthcare settings based on staff annoyance perceptions and identifies problematic correlations of the categorized noise sources with human psychological responses. By conducting this study, it is possible to better understand problematic types of noise sources which should be preferentially addressed in noise control remediation programs.

Background

Hospital soundscapes do not typically provide the calm and restful atmosphere that would satisfy patient and staff preferences (Busch-Vishniac & Ryherd, 2019). This is a crucial mission to promote more healing spaces for patients as well as efficient working environments for staff members in hospitals. One way to achieve that mission is to improve occupant perceptions (e.g., psychological reactions) regarding acoustical environments in hospitals. A number of studies have investigated healthcare providers outcomes related to acoustical environments. The primary importance of obtaining these outcomes is to understand staff perceptions regarding their working environments and reveal problematic noise categories that adversely affect perceptual responses.

The complex mixture of various acoustical stimuli generated by different noise sources, such as alarms, operating sounds, and staff conversation, has been shown to cause stress and disturbance to medical and nursing staff (Arnold & Kornadt, 2011). Among such negative effects, annoyance is listed as one of the critical effects of noise in hospitals, according to the World Health Organization Guideline for community noise (Berglund et al., 1999). In one study, the perceived annoyance and anxiety were found to be significantly higher and work performance poorer in a noisier adult intensive care unit (ICU) compared to a quieter ICU with similar medical acuity and treatment models (Okcu et al., 2011). Another study found a significant correlation between noise levels and annoyance ratings in a pediatric ICU (Morrison et al., 2003). Although most previous studies have convincingly shown an association between poor acoustical conditions and adverse staff annoyance responses, few have systematically specified the types of the noise sources the staff evaluate as particularly annoying or disturbing.

To investigate existing noise sources in hospitals, some earlier studies focused on specific types of medical equipment alongside overall, equivalent levels (Tsiou et al., 1998), maximum sound levels (Bailey & Timmons, 2005), and the percentages of the occurrences (MacKenzie & Galbrun, 2007) of the noise events. A few studies ranked subjective perceptions due to specific noise sources. For example, Hsu et al. (2010) assessed how much nurses felt bothered by a list of noise sources and discovered that alarm noise was perceived to be the most bothersome sound in their cancer unit. In Gabor et al. (2003), six healthy sleep study subjects reported that conversation and alarms were the most disruptive noises they perceived. These studies have provided useful insights into potential problematic noise sources in terms of either their objective noise levels or occupants’ perceptual responses; however, little is known about the deterministic structures of noise sources that explain how different types of noise sources influence perception.

Identifications of intrinsic noise categories can reveal the complex mixture of existing noise sources that is still generally lacking in the published literature. Due to the complexity of the existing sound environment in hospitals, various types of noise sources should be considered, not just a single noise source; otherwise conclusions would be too narrow and specific (e.g., if examining individual noise sources). Further, occupants do not always perceive noise in a similar way. Therefore, inherent structures of the noise sources that describe the unique contributions of various noise categories should be developed; otherwise conclusions may be too broad and vague (e.g., if relating overall hospital noise levels to annoyance).

Due to the complexity of the existing sound environment in hospitals, various types of noise sources should be considered, not just a single noise source; otherwise conclusions would be too narrow and specific.

The objective of this work was to identify some intrinsic categories of hospital noise sources based on staff’s perceived annoyance and explore clear associations of the categorized noise sources with psychological perceptions. In this work, the psychological responses including self-evaluated work disturbance, patient rest disturbance, and noisiness in staff working environments were analyzed. The findings enable us to provide more definite hospital noise structures and practical modification approaches that more effectively improve staff perceptions of their working environments.

Method

Site

This study was conducted at three pediatric and neonatal care units in two hospitals in the United States—a pediatric intensive care unit (PICU) and a pediatric medical-surgical (Med-Surg) unit in a hospital in the Midwestern United States (denoted as Hospital A), and a neonatal intensive care unit (NICU) in a hospital in the Southeastern United States (denoted as Hospital B). The units were selected based on several factors including previous collaborations and site access and logistics.

Site observations were undertaken prior to the survey administration to understand general architectural configurations of the units where the targeted staff were working. The physical/architectural characteristics of unit configurations as well as some relevant features, including acuity levels and paging systems, were recorded over the observation periods and are summarized in Table 1. The PICU is a critical care setting where patients who need intensive treatments are hospitalized. The Med-Surg unit is a postsurgical unit for patients requiring medical rehabilitation. Both units have triangular-shaped racetrack structure, except the Med-Surg unit has slightly expanded unit volume in horizontal and vertical directions, which accommodates an additional five patient rooms and two nurse stations. The NICU has distinct areas corresponding to patient acuity levels and patient room types. The area with high acuity level consists of private rooms and additional transition/isolation rooms. There was no noise control program or training administrated during the study periods at the three units; thus, conditions measured represent true baseline conditions in these units with no research intervention.

Table 1.

Unit Configurations and Characteristics in Hospitals A and B.

Unit	Hospital A		Hospital B
Unit	PICU	Med-Surg Unit	NICUs
Acuity level	Highly acute	Step-down	Advanced care	Critical
Unit configuration	Triangular-shaped racetrack		Nightingale- and pod-style	Off-beds
Paging system	No overhead paging systems		Overhead paging systems (include voice announcement)
Number of patient rooms/beds	19	24	28	8
Number of nurse stations	2	4	3	1
Type of patient rooms	Private	Private	Open	Private
Patient bed types	Cribs for pediatric	Beds for adults	Incubators/baby cribs
Patient room doors	Sliding metal glass door with two or more glass panels	Solid core wood hinged door with metallic push/pull latch	No doors in patient rooms/bed areas	No doors, except transition/isolated rooms with sliding metal glass doors

Note. PICU = pediatric intensive care unit; Med-Surg Unit = medical-surgical unit; NICU = neonatal intensive care unit.

Survey Design

Participants

A total of 100 participants (91 women, 9 men) completed the surveys. However, six participants reported their primary working units as multiple units or other clinical site, or did not identify their primary work units; these respondents were excluded from further analysis as they were not in our three-unit sample specification. In post hoc power analysis, the sample size was determined to ensure reliable statistical results for detecting large effect size.

Participants’ age ranged from 23 to 67 years with a mean age of 34.02 years (SD = 10.35). The majority of the participants were healthcare providers (71 nurses, 8 physicians, 4 nurse practitioners, 2 respiratory therapists, and 3 certified clinical perfusionists). The other participants included two secretarial staff, one nurse aid, and one chaplain. Two participants did not report their occupations. Of the 94 participants, 33 participants worked in the Hospital A PICU, 26 participants worked in the Hospital A Med-Surg unit, and 35 participants worked in the Hospital B NICU. Almost all the 35 Hospital B NICU staff members marked both open and private areas as their primary working environment. Working year experience differed unit by unit. That is, in Hospital A (PICU and Med-Surg unit), the majority of the participants had between 1 and 5 years of hospital work experience, whereas more than half of the participants in the Hospital B NICU had 11 or more years.

The average response rate to the individual survey items was 85%. The demographic distributions of the units are listed in Table 2.

Table 2.

Numbers of Respondents by Gender, Work Shift, Professional Role, and Hospital Work Experience.

Variable	PICU	Med-Surg Unit	NICU
Participants	33	26	35
Gender
Female	31	26	29
Male	2	0	6
Work shift
Day shift	15	10	14
Night shift	14	15	7
Combination of day/night shifts	2	1	13
Professional role
Nurse	29	23	23
Physicians	2	0	6
Other	2	3	6
Hospital work experience (in years)
<1	3	5	2
1–5	15	14	4
6–10	11	3	9
>11	4	4	20

Note. PICU = pediatric intensive care unit; Med-Surg Unit = medical-surgical unit; NICU = neonatal intensive care unit.

Procedure

The surveys were conducted in a web-based format for the Hospital A PICU and Hospital A Med-Surg unit, and in both web-based and paper-based formats for the Hospital B NICU to increase response rates. Careful attention was made by the researchers to avoid overlapping responses between the paper and web surveys from a single respondent. Staff members working in the three units were invited to the surveys by either receiving invitation emails from their hospital administrators or paper forms from the investigators, using standardized recruitment scripts. Participation in the survey was completely voluntary and anonymous. Participants were provided written informed consent information at the survey introduction page. After participants completed the survey, the answers were compiled into a platform of the Statistical Package for the Social Sciences (SPSS version 24) data set. This research was approved by the institutional review board in the authors’ affiliated institute and the administrated hospitals.

A questionnaire was developed to assess subjective perceptions regarding acoustical environments, disturbance/noisiness within the work environments, and perceived annoyance regarding a variety of noise sources. Some parts of the survey questionnaire were based on previous studies (e.g., Okcu et al., 2011; Ryherd et al., 2008) and adjusted for the objectives of this study. Because some components of questions slightly differed unit by unit (due to hospital preference), only responses to identical or equivalent questions were analyzed. Taking only these identical or equivalent pieces, a total of 51 question items were employed in the field surveys and following five variables were chosen based on their relevance of the research objectives.

Work disturbance/patient rest disturbance

Instructions said to complete the questionnaire “in thinking of your primary work area”. Participants were asked to indicate how frequently they perceived that their work was disturbed due to noise and that patient rest/recuperation was disturbed in their working units on a scale of 1 (never) to 5 (very frequently).

Noisiness

Participants were given an instruction “overall, what is your perception of the sound environment” and asked to rate how frequently they perceived noisiness within their working environment on a scale of 1 (not at all noisy) to 5 (very noisy).

Noise annoyance

Participants rated how annoying they perceived each of the 21 noise sources within their working environment. For example, participants indicated “how annoying (if at all) they perceived staff conversation to be” on a scale of 1 (not at all annoying) to 5 (extremely annoying) or marking as N/A (not applicable).

Hearing sensitivity

Participant self-evaluations of their hearing sensitivity were also assessed. Instructions stated “in general, how sensitive are you to noise”. Participants responded on a scale of 1 (not at all sensitive) to 5 (extremely sensitive).

Statistical Analysis

Principal component analysis (PCA)

PCA, one of the extraction methods in factor analysis, was used to identify inherent structural patterns in the staff perceptions of the annoyance of the 21 noise sources. Pattern and component matrices were obtained based on these inter-correlated quantitative noise variables, and rotated with appropriate method (varimax/quartimax with Kaiser normalization or oblimin method). Eigenvalues and percentages of variance explained by each component in the analysis were computed to determine the appropriate number of components and its corresponding matrix. Reliability of each resulting component was assessed using the Cronbach’s α coefficient.

Mixed model analyses of variance

Mixed model ANOVAs were conducted to identify significant differences in categorized noise source factors among the respective units, as well as to confirm and validate the compositional contrasts among the PCA-categorized noise sources. The analyses examined noise annoyance perceptions as a function of the three unit types (between subjects) and categorized noise factors (within subjects). Because multiple comparisons were tested simultaneously, a Bonferroni corrected α was used to determine statistical significance among a number of equivalent comparisons on the same set of observations, so that the probability of making at least one Type I error would be less than .05 (Judd et al., 2017).

Multiple linear regressions (MLRs)

MLR analyses were used to identify statistically significant relationships between the categorized noise factors and staff psychological outcomes including perceived work disturbance, noisiness, and patient rest disturbance. Factor scores obtained from the PCA were used in the MLRs to eliminate multicollinearities among the PCA-categorized noise factors. Two potential covariates—working years and noise sensitivity—were included as potential predictors. In these analyses, an α of .05 was used. To find out whether the proposed models had enough power to detect effects of interest, post hoc power analyses were conducted.

Prior to those analyses, the residuals were examined to identify potential violations of regression assumptions. Violations of normality of residual distribution as well as homoscedasticity of residuals can be caused by the presence of potential outliers, which were detected by some indices including lever, studentized deleted residuals, and Cook’s D statistics (Judd et al., 2017). The normality of distributions was also inspected using the normal quantile-quantile plot. The presence of nonnormal distributions was addressed by removal of outliers or data transformations (e.g., logarithmic transformation).

Results

Descriptive Statistics

Descriptive statistics including mean, median, and standard deviation for the demographic (i.e., age), perceptual (i.e., noisiness, work disturbance, and rest disturbance), and noise sensitivity variables are shown by unit in Table 3. Notice that the mean age of the Hospital B NICU participants was about 10 years greater than the mean age of the participants in Hospital A PICU and Hospital A Med-Surg unit. Among participants in Hospital A, the mean values of most of the variables in the Med-Surg unit were typically greater than the corresponding values in the PICU. Also, the mean scores for disturbance to patient rest and recuperation due to noise were 0.6 lower in Hospital B NICU than the scores in Hospital A PICU and Hospital A Med-Surg unit. The standard deviations of these variables were fairly comparable among the three units.

Table 3.

Descriptive Statistics for Age, Noisiness, Work Disturbance, Rest Disturbance, and Noise Sensitivity.

Variable	PICU		Med-Surg Unit		NICU
Variable	N	M (SD)	N	M (SD)	N	M (SD)
Age	31	29.71 (6.30)	24	29.21 (7.51)	34	41.35 (10.97)
Noisiness	29	4.79 (0.86)	21	4.81 (1.17)	34	4.59 (1.18)
Work disturbance	27	2.74 (0.59)	21	3.29 (1.01)	35	2.86 (0.91)
Rest disturbance	27	3.63 (0.69)	20	3.60 (0.75)	34	2.94 (0.89)
Noise sensitivity	25	2.44 (0.87)	19	2.58 (1.22)	35	2.97 (0.82)

Note. PICU = pediatric intensive care unit; Med-Surg Unit = medical-surgical unit; NICU = neonatal intensive care unit.

According to the result of the correlation analyses with two-tailed tests of significance, perceived noisiness was associated with greater work disturbance, r(82) = .49, p < .0001, as well as patients’ rest disturbance, r(80) = .58, p < .0001. Greater perceived noisiness was observed by participants with more work experience, r(84) = .25, p = .020. Higher noise sensitivity was significantly associated with greater perceived noisiness, r(78) = .24, p = .033, and more work disturbance, r(77) = .24, p = .036. The results of the correlation analysis confirmed the two covariates—working year and noise sensitivity, suggested previously, and were thus controlled in subsequent multiple regression analyses.

PCA

PCA was applied to derive underlying variables that describe the inherent structural patterns in the staff’s perceived annoyance responses due to specific noise sources. Graphical plots showing factor loadings were created for clear visualization of the data set.

Performing the PCA using quartimax rotation method with Kaiser normalization, five principal components from the 21 noise-source variables and the loadings of each component were obtained. Their eigenvalues along with the cumulative percentages of variance explained are shown as a scree plot in Figure 1. Components with eigenvalues greater than or equal to 1.0 have been shown. The greater the eigenvalue that is associated with a component, the more of the variance in the variables that component explains. As observed, more than half of the variation in the noise annoyance, due to 21 noise sources, can be explained by the first three principle components. To examine the nontrivial cumulative variances, an inflexion point of the curve where the great decrease of the variance and eigenvalues occurs was explored on the scree plot, shown in Figure 1. From the fourth principle component in the scree plot, the change in both eigenvalues and percentage of variance explained seems relatively minor. Therefore, the first three components were extracted. Moreover, the noise source variable operational sounds (e.g., respirators/suction) was loaded onto the first and third principle components comparably. The variable cleaning equipment was highly loaded on the fifth component and moderately weighted on the first and third components. Because these two noise items were clearly loaded on multiple factors, they were excluded. The variable unit doors opening/closing was somewhat loaded on both Factors 1 and 2, and ultimately retained as part of Factor 1, but additional examination of this specific noise source may be warranted in future work.

Figure 1.

A scree plot for the five principle components. Percentage of cumulative variance explained and eigenvalue for each component are displayed. The inflexion point of the curves where the great decrease of the explained variance and eigenvalue occurred after the fourth principle component, which suggests that the first three principle components sufficiently explain the variability of the noise source annoyance in the data.

As a result, 19 noise source variables were successfully classified into the first three components (see Table 4) that are named as three following categories—facility noise (i.e., Factor 1 including footsteps, exterior noise, ventilation noise, and other equipment noise), human speech/activity noise (i.e., Factor 2 including staff and visitor conversation), and alarm noise (i.e., Factor 3 including alarms on medical equipment). Higher loading values indicate stronger associations between the variables and the components. For instance, the variable staff conversation had the highest loading on the second principle component. Similarly, visitor conversation and patient sounds attained their highest loadings on the second component among the generated five components. These results indicate that those variables—staff/visitor conversations and patient sounds—were categorized as one component.

19 noise source variables were successfully classified into the first three components.

Reliability analyses, using Cronbach’s α, which assesses the internal consistency of each component, indicated that facility noise (11 items) had the highest reliability (Cronbach’s α = .892), showing high internal consistency. The other two factors—human speech/activity noise (5 items) and alarm noise (3 items)—likewise had high internal consistency (Cronbach’s αs of .817 and .820, respectively, as shown in Table 4). These two factors are therefore sufficiently reliable for describing human and alarm noise annoyance perceptions even though the number of items loaded into these two factors was less than the facility noise factor. Thus, it can be concluded that the three principle components were sufficiently reliable for describing annoyance perceptions due to individual categorized noise sources in the three units.

Table 4.

Factor Loadings for Exploratory Factor Analysis With Quartimax Rotation of Noise Source Annoyance Scales Along With Reliability Measurements.

	Factor 1	Factor 2	Factor 3
Variable	Facility noise	Human speech/activity noise	Alarm noise
Rolling medical/service carts	.54	.40	.10
HVAC systems	.52	.19	−.01
Unit doors opening/closing	.69	.51	.08
Patient room doors opening/closing	.74	.38	.01
Privacy curtain tracks	.76	.06	−.01
Hand sanitizers	.81	−.21	−.04
Cabinets opening/closing	.84	.05	.08
Falling/dropping objects	.77	.24	.06
Sinks in patient rooms	.59	.06	.29
Footsteps	.55	.29	−.02
Exterior noise from outside	.62	.03	.06
Staff conversation	.33	.71	.19
Visitor conversation	.28	.73	.19
Visitor sounds (e.g., footfall)	.29	.72	.18
Patient sounds (e.g., crying)	.35	.57	−.01
Emergency procedures/events	.40	.55	−.02
Alarms on medical equipment	.29	.06	.77
Paging system	.12	.19	.77
Telephone ringing/conversation	.23	.35	.73
Cronbach’s α (.912)	.89	.82	.82

Note. HVAC = heating, ventilation, and air conditioning. Factor loadings >.40 are in boldface.

Figure 2 shows the 2D or 3D PCA scatter plots showing the distributions of the loading values of the first three components for each of the 19 noise source variables displayed as vectors and shaded circles. According to the factor extraction obtained, each vector and its corresponding circle are indicated within their noise categories as follows—facility noise, human speech/activity noise, and alarm noise. A group of vectors pointing in the same direction correspond to a group of noise sources that have the same general annoyance tendency. As demonstrated in Figure 2, the separation of the groups along the three components shows clear division between human speech/activity noise and facility noise on the one side, and alarm noise on the other side. More precisely, the alarm noise can be naturally isolated perceptually from other types of noise sources as the three 2D plots show its distinct area. In contrast, perceptions of the facility and human speech/activity noise categories depend on their compositional contrasts. That is, when observing the contrasts of the facility factor with the other two noise categories, there were no overlapped areas observed (see Figure 2A and 2B). However, comparing the human speech/activity factor with the other two categories, there was an overlapped region between facility and human noise categories on the alarm-human noise-component plane (see Figure 2C). These results suggest that the perceived annoyance of human noise may have some similarities with facility noise, causing a lack of distinct distributions between the two categories. One reason might be that this plane (as shown in Figure 2C) is the last two principle components and is not as sufficient as the first principle component. This should be further examined in later analyses.

Figure 2.

Two- and three-dimensional scatter plots showing results from 21 examined noise sources according to PCA: (A) Component 1-2, (B) Component 3-1, (C) Component 3-2, and (D) all three components. Distribution along the three components is displayed. Each circle represents the size of its group at the center of the mean PCA score for each group and bars represent the standard deviation for each group on each principal component. The axes: the principle components 1, 2, and 3; the proportion of explained variance is indicated in brackets for each principal component.

These componential explanations would be more reasonable when considering those groups of vectors in terms of their frequency components. A group of the vectors associated with the facility noise contains noise sources with primarily broader frequency bands, and a set of the vectors corresponding with the human speech/activity noise includes noise sources with primary energy concentrated in the speech frequency bands. Because those frequency bands shared parts of their frequency ranges, some of their vectors (categorized noise sources) feasibly pointed to a similar dimension. In contrast, a group of the vectors associated with the alarm noise contained noise sources whose primary energy is concentrated in more narrow and objective-specific frequency bands. Therefore, those vectors indicated a very different direction compared to the other two groups.

Mixed Model ANOVAs

The ANOVAs with hospital unit as an independent factor (three units) and noise category as a within-subject factor (three noise categories) was performed on the mean noise annoyance scores computed according to the three PCA noise categories, as shown in Figure 3. The analyses examined significant differences in the categorized noise factors among the units, which would further validate the noise source categorization obtained by the PCA.

Figure 3.

Mean perception ratings of noise annoyance for the three noise categories in the hospitals.

The results of the mixed-model ANOVAs showed that the overall noise annoyance ratings did not statistically significantly vary across the three units examined, F(2, 64) = 0.5, p = .611, once averaging across the three noise categories, as shown in Table 5. These results indicate that there is no evidence that perceived noise annoyance significantly differed among the unit type or unit characteristics once averaging the types of noise sources.

Table 5.

Mixed-Model Analysis of Noise.

Source	b	df	F	R ²	p
Between-subject main effect
Unit type		2, 64	0.5	.015	.611
Within-subject main effect
Noise category		2, 128	25.45**	.285	<.0001
Grouped categories
Frequency-based(specific vs. broad)	.175	1, 64	48.860**	.433	<.0001
Human-based(human vs. non-human)	.017	1, 64	0.384	.006	.535
Alarm-based(alarm vs. non-alarm)	−.192	1, 64	27.984**	.304	<.0001
Individual contrasts
Human versus alarm	.209	1, 64	12.603**	.165	.001
Alarm versus facility	.366	1, 64	42.641**	.400	<.0001
Human versus facility	.158	1, 64	18.063**	.220	<.0001

Note. Two potential outliers and slightly nonnormal distributions of the residuals were identified; however, removal of the outliers and data transformations did not change the inferential results substantively. Specific = human speech/activity noise and alarm noise; broad = facility noise; human = human speech/activity noise; non-human = alarm noise and facility noise; alarm = alarm noise; non-alarm = human speech/activity noise and facility noise.**p < .0056.

Further analyses revealed that annoyance responses among the three noise categories statistically differed, F (2, 128) = 25.45, p < .0001, and most of their grouped/individual contrasts were statistically significant, as shown in Table 5. Averaging across the three units, alarm noise was statistically perceived as most annoying of the three categories, followed by human speech/activity and then facility noise in all three units. On average, staff perceived more annoyance due to the noise sources whose frequency bands were specific—either narrow or speech bands (M_{alarm + human} = 2.34), compared to those whose frequency bands were broad (M_facility = 1.77), F(1, 64) = 48.860, p < .0001, η² = .424. Similarly, staff were significantly more annoyed by the alarm noise (M_alarm = 2.55) than by the nonalarm noise sources—either human or facility noise sources (M_{human + facility} = 1.95), F(1, 64) = 27.984, p < .0001, η² = .293. In contrast, there was no evidence that perceived annoyance of human-originated noise (M_human = 2.13) differed from non-human-generated noise (M_{alarm + facility} = 2.16), F(1, 64) = 0.384, p = .535, η² = −.010.

This significant main effect of the noise category as well as their significant contrasts confirm the three intrinsic noise categories obtained in the PCA. As described in the PCA findings, the 19 noise sources were inherently categorized into the three components. Specifically, the 3D PCA plot, as shown in Figure 2D, illustrates the clear divisions of the noise categories along each axis. The significant differences between facility, human, and alarm noise categories (i.e., the significant individual contrasts of the noise category) in the mixed-model ANOVAs clearly confirm this distinct PCA categorization. Moreover, as indicated on the 3D and 2D PCA plots, alarm noise was perceived much differently from the other types of noise sources, which is supported by the largest mean annoyance differences between alarm and the other two noise categories (ΔM_{alarm versus human} = 0.42, ΔM_{alarm versus facility} = 0.78, ΔM_{human versus facility} = 0.36) that were statistically significant in the mixed-model ANOVA. Furthermore, the annoyance response due to human noise has some similarities with facility noise, as suggested by the overlapped region on the 2D PCA plot in Figure 2C, which has been partially validated with the non-significant difference in the perceived annoyance between human-originated and non-human-generated noise sources (containing facility noise) in the mixed-model ANOVA.

MLRs

The MLR analyses were employein the perceived annoyance betweend to evaluate potential negative associations of the factorized noise categories with staff’s psychological responses including work disturbance due to noise, disturbance to patient rest, and noisiness in the environment. Regressing work disturbance on the three PCA noise factors and the two covariates, the overall model was statistically significant, F(5, 58) = 6.31, p < .001, as shown in Table 6. Of those variables, both alarm and facility noise categories uniquely and significantly predicted perceived work disturbance, t(58) = 3.39, p = .001, η² = .151, and t(58) = 3.39, p = .001, η² = .151, respectively. These results indicate that alarm and facility noise increase staff’s perceived work disturbance.

Table 6.

Multiple Regression Analyses of Staff’s Psychological Outcomes—Work Disturbance, Perceived Noisiness, and Rest Disturbance.

Source	b	df	F	R ²	p
Staff work disturbance due to noise		5, 58	6.31**	.35	<.001
Facility	.31		11.49**	.17	.001
Human speech/activity	.15		1.99	.03	.164
Alarm	.33		11.49**	.17	.001
Working year	.00		0.00	.00	.970
Noise sensitivity	.16		2.62	.04	.110
Perceived noisiness		5, 58	4.30**	.27	.002
Facility	.29		5.86*	.09	.019
Human speech/activity	.31		5.34*	.08	.024
Alarm	−.01		0.00	.00	.961
Working year	.14		1.00	.02	.321
Noise sensitivity	.25		3.80	.06	.057
Patient rest disturbance due to noise		5, 57	3.17*	.22	.014
Facility	.33		11.02**	.16	.002
Human speech/activity	.14		1.64	.03	.206
Alarm	−.12		1.39	.02	.244
Working year	.01		0.01	.00	.929
Noise sensitivity	.12		1.19	.02	.279

*p < .05. **p < .01.

The overall regression of noisiness on the same five predictor variables was also significant, F(5, 58) = 4.30, p = .002, as shown in Table 6. Both human speech/activity and facility noise categories were significant positive predictors of perceived noisiness above/beyond the other variables, t(58) = 2.31, p = .024, η² = .069, and t(58) = 2.42, p = .019, η² = .076, respectively. These results indicate the potential negative consequences of human speech/activity noise and facility noise categories on perceived noisiness.

The multiple regression of patient rest disturbance on the predictors as a set was also significant, F(5, 57) = 3.17, p = .014, as shown in Table 6. Only facility noise was a significant positive predictor of perceived disturbance to patient rest above/beyond the other variables, t(57) = 3.32, p = .002, η² = .145. This result supported the presence of the potential negative effect of facility noise on perceived disturbance to patient rest (see Table 6).

In sum, the analyses revealed the negative effects of individual noise categories on staff’s psychological responses—perceived noisiness and disturbance to staff’s work and patient’s rest. It can be seen that facility noise was more strongly associated with the staff perceptions, whereas alarm noise was contributed to work disturbance and human speech/activity noise particularly influenced perceived noisiness.

... the analyses revealed the negative effects of individual noise categories on staff’s psychological responses—perceived noisiness and disturbance to staff’s work and patient’s rest.

Discussion

This work developed an intrinsic noise classification based on staff’s annoyance responses in three hospital units and examined unique associations of the noise categories with psychological outcomes. The PCA revealed three inherent noise categories among the various existing noise sources, which was further validated using the mixed-model ANOVAs. Interestingly, there was no difference in overall noise perception across the three units, but significant differences did emerge when utilizing the PCA categories. The subsequent MLR analyses demonstrated significant relationships between the categorized noise factors and staff’s psychological perceptions including work/rest disturbance and noisiness. Taken as a whole, these findings support the usefulness of utilizing inherent noise categorizations in future hospital noise research and remediation.

Several earlier studies explored, and experimentally classified, existing noise sources into several categories. MacKenzie and Galbrun (2007) stated that most of the former studies identified noise sources to be alarms, equipment, and staff conversation, which has been confirmed in this study. Other previous studies categorized noise sources into constant/quasi-constant versus thrust (Tsiou et al., 1998) or steady versus quasi-steady (Hilton, 1985) noise categories. These studies separated noise sources in terms of their impulsiveness and the constancy of the occurrences. Although illuminating, these classifications were somewhat less intuitive and were not empirically derived. Because many existing noises can be assigned to further categorizations, intuitive categorization would not be straightforward or obvious. The three types of noise categorizations discovered in this study would highlight more intuitive generalizations of existing noise types with a reasonable number of types under the analytical justification. Indeed, this result would not be an ultimate conclusion but at least can be fairly reasonable in the examined environments.

Previous studies have largely focused on either overall or individual effects of hospital noise source(s) on human psychological responses (Morrison et al., 2003; Okcu et al., 2011). Although illuminating, these approaches did not generally include statistically defined or categorized hospital noise impacts, so that consequential modifications or recommendations for resolving those effects would be very limited. This study successfully identified the potential associations of the specific noise categories with the human psychological perceptions, which further strengthened the previous research in terms of their consistency. Further studies with more widely validated survey tools and/or patient or family-level data would be valuable for validations of the findings; however, acquisition of these types of data can be challenging due to cost, ethical issues such as patient privacy and capability, and access to adequate sample sizes of patients and families.

The prevalence of greater perceived noisiness due to human speech/activity noise would be explained by consistent occurrence of the noise as well as its distribution among the working units. In contrast, the presence of greater disturbance to work due to alarm-related noise would be justified by the impulsiveness of alarm noise, which has been recognized as potentially disrupting to the staff’s workflow (Busch-Vishniac, 2015). However, as described in the PCA results, the alarm noise was perceived much differently as well as more annoying compared to the other two noise categories. This distinctive perception was reasonable by considering a primary function of alarms—“alerting” the information (otherwise, alarm noise did not function as it should be). This finding, therefore, suggests that some different treatments should be considered for alarm noise sources in hospitals. The facility noise was significantly associated with all the psychological outcomes examined in this study because it included most of the building-related noise sources, such as heating, ventilation, and air conditioning (HVAC), unit/room doors, and footstep noise and, therefore, accounted for the largest percentage of variance in the PCA model. Therefore, it would be reasonable to conclude that the combination of those noise sources inherently contributed to negative human outcomes. This well-defined characterization of noise sources would enable us to identify intrinsic structures of existing hospital noises and potential noise-related perceptions and influences, which should be preferentially addressed to improve human perceptions in hospitals.

This well-defined characterization of noise sources would enable us to identify intrinsic structures of existing hospital noises and potential noise-related perceptions and influences, which should be preferentially addressed to improve human perceptions in hospitals.

The greatest limitation of this study is the small number of participants (N = 94) that certainly affected the probability of errors and the generalizability of the results. Based on a series of post hoc power analyses, a sufficient statistical power (i.e., greater than .80) was found at the large effect size level, whereas less power at the small to medium effect size level was found for most of the regressions and ANOVAs. With limited statistical power and effect sizes, we can only observe associations that are substantial, which may limit the significance of some of the trends examined in this study. Additionally, this small sample size dropped the subjects-to-items ratio, which is a critical factor in developing PCA models. Considering that the ratio in this study (4.475:1) was slightly below the recommended minimum ratio of 5:1 (Gorusch, 1983), the PCA result might be moderately reliable and stable. Future studies should consider a larger sample size as well as a sufficient subject-to-item ratio when applying the PCA technique, which would be necessary to find some “well-defined” soundscape characteristics that can be applied for the majority of healthcare environments. Additionally, it would be useful to validate results in other unit types or with other populations, such as patients and family members. Despite some sample limitations, the analytical approaches presented in this study provide a good framework for future research on hospital soundscapes. In practice, this framework allows hospital designers and staff to better understand categories of noise sources and how they are perceived by occupants, which in turn can help them to create more pleasant and comfortable hospital soundscapes.

Implications for Practice

Use of categorized noise sources identified in this study could be helpful for specifying key structures of existing hospital noises

Descriptions of the psychological influences of the noise categories on the principle component analysis (PCA) dimensions suggest potential problematic noise source(s) that should be preferentially modified in practice

Facility noise is the first category to modify for improving staff perceptions

Alarm noise may reduce staff's work disturbance, but the disruption could be a part of its primary functions

Improvements to human speech/activity noises can specifically reduce noisiness perceived by staff

Acoustical modifications regarding the noise source classification would effectively improve staff perceptions of their working environments

Footnotes

Acknowledgements

We are indebted to Dr. Ashley Darcy-Mahoney, Dr. Jonathan Weber, and Mr. Ian Hough for their collaboration, along with the incredible staff at all three study sites.

Declaration of Conflicting Interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This work was partially supported by the Baptist Health Foundation Sherry Kranys Research Innovation Fund.

ORCID iD

Yoshimi Hasegawa, PhD

References

Arnold

Kornadt

(2011). Noise in an intensive care unit. The Journal of the Acoustical Society of America, 130, 3754–3760. https://doi.org/10.1121/1.3655884

Bailey

Timmons

(2005). Noise levels in PICU: An evaluative study. Pediatric Nursing, 17(10), 22–26.

Berglund

Lindvall

Schwelaand

D. H.

Goh

K. T.

(1999). Guidelines for community noise [Technical report]. http://www.bvsde.paho.org/bvsci/i/fulltext/noise/noise.pdf

Busch-Vishniac

(2015). A model of clinical alarm errors in hospital. Biomedical Instrumentation & Technology, 49, 280–291. https://doi.org/10.2345/0899-8205-49.4.280

Busch-Vishniac

Ryherd

(2019). Hospital soundscapes: Characterization, impacts, and interventions. Acoustics Today, 15(3), 11–18. https://doi.org/10.1121/AT.2019.15.3.11

Gabor

J. Y.

Cooper

A. B.

Crombach

S. A.

Lee

Kadikar

Bettger

H. E.

Hanly

P. J.

(2003). Contribution of the intensive care unit environment to sleep disruption in mechanically ventilated patients and healthy subjects. American Journal of Respiratory and Critical Care Medicine, 167, 708–715. https://doi.org/10.1164/rccm.2201090

Gorusch

R. L.

(1983). Factor analysis (2nd ed.). Lawrence Erlbaum.

Hilton

B. A.

(1985). Noise in acute patient care areas. Research in Nursing & Health, 8(3), 283–291.

Hsu

Ryherd

West

Barnhill

Swisher

Levit

(2010). Further studies of hospital noise control at the Johns Hopkins hospital: Part 2. The Journal of the Acoustical Society of America, 127, 1805. https://doi.org/10.1121/1.3384082

10.

Judd

C. M.

McClelland

G. H.

Ryan

C. S.

(2017). Data analysis: A model comparison approach to regression, ANOVA, and beyond (3rd ed.). Taylor & Francis.

11.

MacKenzie

D. J.

Galbrun

(2007). Noise levels and noise sources in acute care hospital wards. Building Services Engineering Research and Technology, 28(2), 117–131. https://doi.org/10.1177/0143624406074468

12.

Morrison

W. E.

Haas

E. C.

Shaffner

D. H.

Garrett

E. S.

Fackler

J. C.

(2003). Noise, stress, and annoyance in a pediatric intensive care unit. Critical Care Medicine, 31(1), 113–119. https://doi.org/10.1097/00003246-200301000-00018

13.

Okcu

Ryherd

E. E.

Zimring

Samuels

(2011). Soundscape evaluations in two critical healthcare settings with different designs. The Journal of the Acoustical Society of America, 130, 1348–1358. https://doi.org/10.1121/1.3607418

14.

Ryherd

E. E.

Waye

K. P.

Ljungkvist

(2008). Characterizing noise and perceived work environment in a neurological intensive care unit. The Journal of the Acoustical Society of America, 123, 747–756. https://doi.org/10.1121/1.2822661

15.

Tsiou

Efthymiatos

Theodossopoulou

Notis

Kiriakou

(1998). Noise sources and levels in the Evgenidion Hospital intensive care unit. Intensive Care Medicine, 24, 845–847. https://doi.org/10.1007/s001340050676