Monitoring time-varying noise levels and perceived noisiness in hospital lobbies

Statistical noise gauges

Unpredictable impulse noise, which blends into background noise, often poses the biggest difficulties for monitoring mixed noise sources in hospital lobbies. Some studies have focused on the variation of L_eq over a 24-h period. ^{1 –5} For road traffic noise, Griffiths and Langdon ⁶ proposed using a statistical level of L_N , with L ₁₀–L ₉₀ as the measure of variability. De Coensel et al. ⁷ used L ₅–L ₉₅ as an effective descriptor of noise observations since they showed much more variation among statistical levels, while Torija et al. ⁸ used a sound-level variance which was calculated using the standard deviation of the sound levels. However, Torija et al. ⁸ provided only limited psychological research regarding their models. The aforementioned indicators provide an idea of the cumulative fluctuation levels in the average background noise. Hence, the statistical approach and the repetitive feature technology were confirmed as useful tools to apply in detailed correlation analysis of the relationships between objective measurements and subjective evaluations of complicated noise as in my previous study. ⁹

However, the results of that study showed lower benefits (e.g. in terms of correlation coefficient (r) values) than were achieved when the tools in question were applied to independent noise sources. This aroused our curiosity regarding the structural details of sound levels and the characteristics of noise sources in individual hospitals, particularly in relation to the emphasis placed by Vardeman and Morris ¹⁰ on data integrity in the statistical analysis process for science-based certifications. For example, Kaczmarek and Preis ¹¹ investigated how the time-varying structure of road traffic levels affect noise-related annoyance judgments. However, their four traffic noise structures were artificially created in a laboratory from single vehicle pass-by recordings, and the stimulus structures of their subjective noisiness research were mostly subjected to artificial random. They were not necessarily need to consider the data normality of measurements of real community noise, which is defined by the WHO ¹² as noise emitted from the various sources of noise in a community, including road traffic, businesses, construction work, and other sources; that is, the real noises that typically and commonly occur in our surrounding environments. In fact, when it comes to measuring this complex type of noise in general, the only feasible option is to carefully choose the noise source data used, giving due consideration to whether the chosen noise sources are essential sources of the prevailing noise. More specifically, the same option applies when measuring noisiness in hospital lobbies. In order to make the required observations more simple, Harman ¹³ utilized subjective factor analysis derived from the observation of patients and those accompanying them in hospital lobbies, which was then applied in categorizing the major noise sources selected. In a more recent study, Sudarsono et al. ¹⁴ utilized focus group data in selecting eight noise sources, with the focus group being asked to identify what kinds of sounds in hospital wards cause patients to feel annoyed. At the same time, the sources of noise in hospital lobbies have rarely been studied in previous surveys, as most past surveys have focused instead on noises that commonly disturb critically ill patients in ICUs or emergency departments. ^{3,15 –17}

This new study examined whether it is necessary to first confirm the normal or non-normal state of a data set before calculating the statistical values for measurements, since this issue was not explored in detail in my previous study. ⁹ Meanwhile, the application of data normality testing could offer various research benefits (such as correlation coefficient r), as we believe that the statistical levels and repetitive feature technology of noise measurements would also serve as an effective descriptor of the subjective noisiness of the complicated noise events in hospital lobbies. Given the two purposes stated above, the new statistical values used in this study consisted of independent single factors and potential compound factors, and a method for measuring subjective noisiness judgments was investigated. Accordingly, the discussion of the previous study was revisited and elaborated upon in this study.

Subjective noisiness

Surveys of dissatisfaction with prevailing noise conditions have been carried out with the object of developing acceptability predictors for the target noise sources. Paiva et al. ¹⁸ reported that noise-related annoyance, which they defined based on interviews with 225 subjects, is a subjective indicator of the perceptions of and sensitivity to road traffic noise in residential areas. Yamada et al. ² performed post-occupancy evaluations of subjective responses to sound environments in hospital wards. Their analysis of a questionnaire survey showed that many in-patients regarded conversations and noisy medical equipment, sounds which were measured by L_Aeq at 15 min intervals, as unacceptable noises. Meanwhile, Aletta et al. ¹⁹ also monitored sound levels at L_Aeq ,_{15 min} and L_N ,_{15 min} from the viewpoint of the quality of acoustic environments, including subjective evaluations thereof, surveying the perceptions of both residents and staff members in nine living rooms of five nursing homes. The results showed that L_Aeq ,_{15 min} and L_N ,_{15 min} were significantly different at different times of the day or due to different sound sources types in the different living rooms, and that there was a positive correlation between L_Aeq ,_{15 min} and the number of persons in a room. More recently, Aletta et al. ²⁰ reported the results of an emailed questionnaire indicating that limited differences in noise sensitivity scores emerged for different staff roles, while greater differences between bedrooms and living rooms were found. Thomas et al. ²¹ surveyed sound levels measured in individual bedrooms, living rooms, and corridors to obtain the typical levels over the course of a day. The results of the staff responses indicated that a number of architectural improvements had resulted in a fine acoustic environment. Considering these and other study results stated above, we believe that psychoacoustic analysis provides an effective process for examining the evaluative efficiency of objective variables integrated over a period of 15 min after acoustic interventions. That said, we define the responses received from residents, in-patients, or their accompanying friends and family members as those of users in order to distinguish between staffers and customers in Taiwan, which is important because the perceptions of employees are usually affected by the specific requirements of their professions.

To clarify the relationship between the objective measures of and subjective responses to complicated noise, this study first conducted a questionnaire survey by the visitors in the lobbies were asked to fill out the questionnaire, since it would be difficult to evaluate such noisiness by surveying the professionals working in the hospitals. ^2,15,22

Loudness

Zwicker’s loudness in hospital lobbies has also been investigated in previous studies, ^23,24 while Tang ²⁵ previously investigated objective and subjective evaluations of the air conditioning noises in offices. Tang ²⁵ found that the less commonly used Zwicker’s loudness level performs better than the commonly adopted noise criterion (NC) curves ²⁶ when predicting the noise perceptions of office workers in air-conditioned landscaped offices. Berglund et al. ²⁷ used Zwicker’s loudness scale to investigate annoyance levels due to environmental noises. They found that Zwicker’s specific loudness based on critical bands matched listeners’ judgments of annoyances more accurately than L_Aeq levels. They presented 15 independent noise samples, such as “passing car,” “office printer,” and so on, in pairs, each with a duration of 4 s, using non-preference judgments. Kaczmarek and Preis ¹¹ concluded that the average loudness, N, is better correlated with annoyance ratings (r = 0.96) than the percentile value of loudness N ₅ and that N can be used a noise index in the laboratory. In referencing the abovementioned studies on loudness, we found that their noisiness research subjects were mostly subjected to specific individual noise sources or artificial noise sources. In contrast, the lobby noisiness in our study consisted of many real noises mixed together, rather than recordings of discrete separate noise sources.

Other measures, in particular the sense of time-varying in the continuity of noise levels, have been investigated by using autocorrelation function analysis. ^28,29 This approach resembles the process used to investigate 1/f noises in the soundscapes survey conducted by De Coensel et al. ³⁰ Both of these models are similar in terms of the process used in detecting fluctuations in signal chains.

Hospital survey

Sixteen urban hospitals located in various Taiwanese cities, all of which included service departments such as internal medicine, surgery, orthopedics, and cardiology departments, among others, were selected for investigation in this study. The relevant characteristics of the 16 hospitals are listed in Table 1. Each of these hospitals had outpatient services, an emergency department, examination rooms, operating rooms connected to their lobbies, and wards with more than 100 beds (Table 1). Furthermore, because of environmental regulations in Taiwan, hospitals are legally required to be located in areas meeting the first-class quiet criterion of 55 dBA from 6:00 a.m. to 8:00 p.m. Nonetheless, the sites for private hospitals are essentially chosen on the bases of cost analyses and other factors besides noise-related issues. ³¹ Meanwhile, hospitals hosting medical schools were excluded from this study, because their service departments, especially their emergency departments, were located in separate buildings. Each of the hospitals surveyed is confined to a single site; large-sized hospitals consisting of multiple buildings were not included. The main entrance of each surveyed hospital is adjacent to a main road with a width of more than 30 m, and the distance from the hospital façade to the road in each case is less than 6 m. The main entrances of these hospitals open into their lobbies, all of which have outpatient registration and dispensary counters. Some of the hospitals surveyed have a drug or convenience store adjacent to the lobby (Table 1).

OPEN IN VIEWER

Table 1.

The sizes of the lobbies and the characteristics of adjacent spaces in 16 hospitals (A–P) surveyed.

HO	NP	LC	AF	AR (m2)	VO (m3)	CH (m)	NS	OD	ED	EV or ES	CD
A	42	TR	2	100	280	2.8#	20	ˆ
B	35	RE	7	295	1033	3.5#	80			ˆ
C	28	TR	5	357	964	2.7#	70	ˆ	ˆ	ˆ
D	44	TR	3	300	900	3.0	60			ˆ
E	39	TR	3	300	900	3.0	140	ˆ	ˆ	ˆ
F	49	RU	2	168	588	3.5#	50				ˆ
G	29	TR	17	197	887	4.5#	70	ˆ	ˆ	ˆ
H	31	RE	3	287	718	2.5#	120	ˆ	ˆ	ˆ
I	22	TR	7	410	12,340	2.8#	150	ˆ	ˆ	ˆ	ˆ
J	32	TE	4	1300	12,900	3.0#	150	ˆ	ˆ	ˆ	ˆ
K	28	TR	10	713	3919	5.5#	108	ˆ	ˆ	ˆ	ˆ
L	44	RE	9	197	788	4.0#	60	ˆ	ˆ	ˆ	ˆ
M	36	IN	3	310	930	3.0#	110	ˆ
N	41	TR	10	300	2250	7.5#	72	ˆ	ˆ	ˆ
O	40	TR	7	276	828	3.0	75			ˆ	ˆ
P	52	BU	4	240	843	7.5#	60	ˆ	ˆ	ˆ	ˆ

HO: hospital order; NP: number of participants for subjective noisiness survey; LC: location primary impacts, including traffic (TR), trees or garden (TE), bustling (BU), industry (IN), rural (RU), tranquil residential (RE); AF: amount of flooring in the lobby building; AR: area of lobby hall excluding hallway, stores, elevator facilities, and so on; VO: volume of lobby hall excluding hallway, stores, elevator facilities, and so on; CH: ceiling height of lobby hall; NS: number of seats in the waiting area of lobby; OD: connected to outpatient department; ED: connected to emergency department; EV: elevator; ES: escalator; CD: connected to drug or convenience store; #: hall ceiling was decorated by some absorptive treatments.

Noise measurements

To measure the noise levels in the hospital lobbies, A-weighted equivalent sound pressure levels (L_Aeq , _T ) using time averages at 15 min intervals were obtained over the daytime period from 9:00 a.m. to 5:00 p.m. and over the nighttime period from 5:00 p.m. to 12:00 a.m. Two microphones (Brüel and Kjær, B&K type 4190) with a fast time constant (0.125 s) were used to record noises at two receiver positions. The first position was in the center of the given lobby’s visitor seating area at least 1 m away from the walls and at a height of 1.2 m, with the goal being to record all the sounds that would be experienced by a visitor surrounded by the lobby’s soundscape. The other microphone was located outside of the lobby’s main entrance and played the role of a road-traffic noise detector. During all the measurements, the patients, staff members, and patients’ companions continued with their normal activities. Data were collected with a B&K PULSE system, and corresponding analysis was conducted using B&K 2-CHs constant percentage bandwidth (CPB) analyzer software.

A further calculation of the sharpness and unbiased annoyance (UBA) ³² was derived from the broadcasting scenarios over durations of 30 s each at the same position as the L_Aeq measurements, with the broadcasting scenarios from 9:00 a.m. to 12:00 p.m. in the 16 lobbies being used. The UBA is a combination of the metrics of sharpness (S), fluctuation strength (F), and N ₁₀ level, and includes a correction factor for time of day/night factor (d) as given below

U B A = d ( N 10 s o n e ) 1.3 [ 1 + 0.25 ( S a c u m − 1 ) log ( N 10 s o n e + 10 ) + 0.3 ( F v a c i l 1 + N 10 s o n e N 10 s o n e + 0.3 ) ]

1

where

d = { 1 1 + ( N 10 5 ) 0.5 from from 6 a.m. 10 p.m. to to 10 p.m. 6 a.m.

A discrepancy was found between daytime and nighttime records of this application in this study. Only selecting daytime for analysis is a good fit in global terms (Figure 1). The hospital lobby UBA survey was specifically arranged due to the results of the psychological questionnaire.

OPEN IN VIEWER

Figure 1.

The age brackets and housing situations of the participants who took part in the subjective questionnaire research.

According to ISO 532-1:2017, ³³ loudness alone may not be useful in assessing the harmful effects of sound events. The harmful effects of such events must instead be determined by excluding the impulse noise in a stationary sound field, as the impulse noise of the noise source may not be modified. However, harmful noises occurred at various times in the 16 hospital lobbies as an aspect of the lobbies’ normal conditions. Darbyshire, ³⁴ in a study of noises in intensive care units (ICUs), reported that noise sounds with higher pitches result in higher loudness than those with lower pitches, at least with respect to loudness as quantified by ISO 532-1:2017. ³³ In the previous study under consideration here, ⁹ the noise spectra of the 16 hospital lobbies throughout the daytime (i.e. from 9:00 a.m. to 5:00 p.m.) were very similar in terms of the shape of each octave band, and in terms of higher levels of noise occurring over the middle octave range of 250–500 Hz. As such, the use of loudness levels was not repeated in this study.

Noise event categories and questionnaire survey of noisiness

The major noise sources in the hospital lobbies

Besides traffic noise, verbal communications in a hospital lobby constitute a necessary behavior at registration desks, dispensary counters, and outpatient service counters. The actual in-take procedures that patients typically go through in hospital lobbies in Taiwan, which normally last longer than 20–30 min, can result in uncomfortable experiences. Even though some hospitals utilize online pre-registration systems and outpatient digital number systems, Wang et al. ³⁵ reported that the average waiting time in outpatient departments in Taiwan was 23.26 min, with averages of 0.75 min spent on registration, 2.38 min for having blood samples taken, 4.05 min for radiology exams, and 7.92 min for receiving prescribed medicines. The focus of this study, meanwhile, was on the noise-related variables affecting busy hospital lobbies, which mainly consist of traffic, facilities, medical appliances, people, daily emergency routines, and various accidental events. Such noise sources cause dissatisfaction among patients and their friends and family members, who generally require environments quieter than those they usually experience in their daily lives. However, there are currently no regulations in Taiwan regarding the noises that patients may experience in hospital lobbies prior to being seen, especially with regard to building construction. The procedure used in categorizing noise events requires the detection of confounding noise-related variables that may distort the subjective noisiness levels of a sound environment in its normal state, and this procedure was applied in the 16 hospital lobbies.

The possible sources of noise during the daytime in hospital lobbies are complicated and overlapping, making their evaluation difficult. However, in a previous study, Park et al. ¹⁷ summarized the top five discriminable noise sources in hospital ICUs by employing an observer in the patient rooms of those ICUs. The daily activities in hospitals result in excessive noise being generated from sources such as enormous air conditioners, medical treatments involving noisy portable carts, conversations, and even the sounds of TVs in lobbies. Recently, many hospitals in Taiwan have also opened a convenience store in their lobby, thus greatly increasing the complexity of the noise profile in these places. Furthermore, hospitals typically have large and complicated mechanical infrastructures, including facilities such as magnetic resonance imaging (MRI) and computerized tomography (CT) facilities. For convenience, to dampen their sounds and vibrations, and for ease of use in emergencies, these pieces of equipment are routinely sited on the first floor (1F) or basement floor (BF). Such facilities, however, drastically augment the noise measured in operating rooms over a frequency range of 500–2000 Hz and with spectral peaks of 250 Hz. ³⁶ However, in our previous survey of such facilities in Taiwan, we found that the noise levels of these magnet facilities measured in the hallways right outside of these examination rooms were under 49 L_Aeq at 1-min intervals, indicating the degree of insulation for such huge mechanical rooms in Taiwan.

Based on the annotations of noise events provided in the preliminary study referred to above, an effort to identify the noise sources relevant to subjective responses was undertaken by employing an observer sitting beside the sound receivers in the hospital lobbies who simultaneously took notes regarding the perceived noise events. Examples of noise event-related notes including those pertaining to outdoor and indoor noises are shown in Figure 2. First, each 24-h recording period (from 9:00 a.m. to 9:00 a.m.) was annotated off-line for a source-specific classification of noise measurements. The results indicated that the noise items of recordings (Table 2) were not clustered along with differences larger than others within certain categories (e.g. during the daytime) in the 16 hospital lobbies. If the sound level of an impulse wave (e.g. due to a rainstorm or thunder) was larger than the background noise by 10 dB during the surveyed period, the taking of subjective and objective measures would be stopped and restarted on another day.

OPEN IN VIEWER

Figure 2.

An example of the notes regarding perceived noise events, which were detected by the peaks in the recordings when the sound levels increased or fell by 3 dB or more.

OPEN IN VIEWER

Table 2.

Factor analysis results of noisiness and mean noisiness scales (N.S.) with standard deviations (SDs) across the 16 hospital lobbies (Kaiser–Meyer–Olkin (KMO) = 0.849, n = 584).

Item (annotation)	Factor loading	Eigenvalue	Percent of total variance explained	N.S.	SD of N.S. among the hospital lobbies
Factor 1: Primary noise (N = 6) (Cronbach’s α = 0.834)
Facilities (air conditioners, elevators, escalators, aeration fans, etc.)	0.769	4.64	38.6%	2.60	0.33
Broadcasting (departmental announcements or occasional channels, etc.)	0.761			2.87	0.25
Medical appliances (carts, sickbeds, medical record cabinets, pulse or fever examination equipment, beds, wheelchairs, printers, etc.)	0.740			2.54	0.29
Roll calls (electronic alarms or oral calls, etc.)	0.696			2.91	0.26
Entrance noises (door leaf opening, people moving, etc.)	0.639			2.78	0.27
Ambulance noises (emergency services, ambulance car bells, stretchers, etc.)	0.497			2.85	0.29
Factor 2: Accidental noise (N = 5) (Cronbach’s α = 0.746)
Construction activity (indoor decoration, water and electricity, outdoor noises around buildings including roadway construction, etc.)	0.779	1.37	11.5%	3.14	0.31
Children playing (hubbub, infants crying, etc.)	0.774			3.17	0.22
Traffic (vehicles starting, braking, and passing each other on the road, etc.)	0.668			2.92	0.23
Conversations (counter, staff, outcoming, etc.)	0.572			3.15	0.18
TV (hygiene instructions, commercials, etc.)	0.524			2.63	0.24
Factor 3: Disregarding noise
Footfall (running, high heels, leather shoe, etc.)	0.307	–	–	2.84	0.28

Depending on the habitual pattern of the sound sorting processes for the care facilities, source classification was finally categorized into 12 source groups (Table 2). The noise events were categorized on the bases of semantic features that integrated perceptual source aggregation as suggested by Guastavino. ³⁷ The point is that a “same” acoustic phenomenon could be classified according to either the source that produced it or according to the action that generated the sound (e.g. the “hubbub sound” which can be categorized either as “children playing” as in “din, loud, chaotic…” or with other prolonged noise). The purpose of this process was similar to that of similar processes used in some previous studies ^17,38 conducted to obtain behavioral observations of the major sound sources in ICUs or emergency departments. Furthermore, Aletta et al. ¹⁹ arranged seven noise source groups (including installation, operational, electric, environmental, and human noise sources) for five nursing homes in Flanders (Belgium) for subjective surveys. However, the environmental noises included many kinds of anthropic noises (e.g. from toys, people outside), such that they were easy to confuse with human sounds and there was some response bias for the different respondent groups in the five independent nursing homes. Factor analysis is an exploratory technique that can be applied to a set of observed noise events that seeks to find underlying items (subsets of events) from which the noise items can be categorized (Table 2). In this study, we hope to reduce the number of sources from which unperceived noise events or the underlying events (e.g. insect sounds inside or outside) may be ignored, ¹³ such that common noise sensations across the 16 hospital lobbies would be determined by the different respondents.

The factor analysis process and the noisiness scale

In order to simplify the subjective responses of all the participants, factor analysis was employed to describe variability among 12 noise source items in the 16 hospital lobbies. The results of factor loading (Table 2) indicated an orthogonal relation compared with the noisiness scale results of the individual hospitals (Figure 3) compiled with recommendations provided by the ISO/TS 15666:2003. ³⁹ The purpose of this procedure was to differentiate how the respondents in this study experienced noisiness in different hospital environments. Nevertheless, their responses can be summarized using a unit scale of noisiness of each noise source provided through factor analysis.

OPEN IN VIEWER

Figure 3.

The noisiness values with their 95% confidence intervals for each hospital according to the questionnaire responses of patients and those accompanying them measured in the lobbies of the hospitals.

In studies of subjective noisiness in hospital lobbies, the patients or their companions (specifically, any patients or companions above 13 years of age) in those lobbies who perceived them to be noisy were interviewed. The 12 noise source groups identified as essential events of noise in the lobbies by a source-specific classification (Table 2) were rated according to a five-point verbal scale ranging from “1” to “5” in the questionnaire given to those surveyed to assess the noise source items in each hospital. ⁴⁰ The questions were posed in the form of “Do you agree that [each noise source and its possible events] in this hospital is a serious problem?,” with the available responses regarding each specific noise event being: (1) “Disagree strongly,” (2) “Disagree somewhat,” (3) “Neutral,” (4) “Agree somewhat,” and (5) “Agree strongly.”

The entire questionnaire, including items regarding personal information (age and the type of property they resided in), took less than 5 min for each subject to complete. The total number of subjects was 584. For each hospital, at least 28 people participated in the survey. The age bracket of each participant and type of property that he or she resided in were queried simultaneously, and the results are shown in Figure 1. As shown there, the largest percentage of participants consisted of participants aged 20–29 years, while the participants were fairly balanced in terms of the types of properties they resided in, with the exception being the small number of participants who resided in industrial areas. The subjective responses of the participants to the questionnaire survey was classified for noisiness according to the definitions established by Berglund et al. ⁴¹

Repetitive feature of noise levels

In past studies regarding the statistical attributes of time-varying traffic noise, potential eigenvalues of temporal variations resulting from such factors as earthquake waves were frequently determined using correlation models. ^{42 –44} This intrigued us in light of the findings reported by Kaczmarek and Preis, ¹¹ who found that the highest annoyance ratings were obtained for situations in which the traffic was distributed evenly, while the most clustered distributions resulted in the lowest annoyance ratings. To evaluate the disturbances caused by the various hospital noises using the normalized autocorrelation function (NACF), the effective duration of noise levels (L_Aeq ,_{15 m}), denoted by τ_e values (h), was calculated to detect each noise’s temporal repeated feature between forward and backward signals along the measured timeline.

The autocorrelation function (ACF) was also proposed as a means of evaluating the sound qualities in a concert hall (Ando, ⁴⁵ Section 2.1, Ando and Kageyama ⁴⁶ ), with the function being expressed as the similarity of the sound signals along the timeline and defined by

Φ p ( τ ) = l i m T → ∞ 1 2 T ∫ − T + T p ′ ( t ) p ′ ( t + τ ) d t

2

where p′(t) = p(t)s(t), in which p(t) is the sound pressure and s(t) is the ear sensitivity. For convenience, s(t) was chosen as the impulse response of the A-weighted network. The value τ represents the time delay (h), and the value of 2T is the integration interval. As such, the NACF ⁴⁵ is expressed by

ϕ ( τ ) = Φ p ( τ ) Φ p ( 0 )

3

where Φ(0) represents the ACF at delay time τ = 0 as the maximum Φ(τ). The τ_e values were defined by the 10-percentile delay (at −10 dB) obtained practically from the decay rate extrapolated in the range from 0 to −10 dB of the logarithmic NACF modulus (see Appendix 1). Namely, the τ_e values (h) were calculated with a sampling rate of 0.25 h against the alpha rhythm of the brainwave signals, 0.01 s.

Global results of the survey of daytime and nighttime noise levels

In the early stage of the measurement process, it was determined that the noise levels in general hospitals have a continuous steady state, with variations during the daytime (9:00–17:00) being assumed to be normally distributed. This explains why the noise events were averaged at 15-min intervals in this study to detect the noise events without an impulse wave. ² The frequency of occurrence and the sound pressure level of multiple impulse noise events can easily be assessed through their cumulative percentile sound pressure level. Such impulse noises include ambulance sirens, human screams, and construction noise. As shown in Figure 4, the cumulative percentages of daytime (9:00–17:00) noise events were normally distributed for both indoor and outdoor areas of hospitals A–I. Sporadic impulse noise events marked by high decibel levels in the nighttime (17:00–24:00) are located on the left side of the distribution; these events were supposed to be non-normal and excluded from our analysis. This procedure was chosen based on the objective of detecting effective descriptors of subjective noisiness in the hospital lobbies, which was in contrast to the aim of the study by Xie et al., ³⁸ which was to identify hospital wards with fewer noises and fewer interruptions in the nighttime given that the nighttime is the most important time for “sleep” in critical care wards.

OPEN IN VIEWER

Figure 4.

Comparative cumulative rates (lines) and distributed percentages (columns) of the levels of L_Aeq at each 15-min interval over the daytime (9:00–17:00) and the nighttime (17:00–24:00) in the hospital lobbies in the preliminary stage of this study (hospitals A–I). The vertical values on the left side denote the frequencies of the noise levels while the percentages on the right side indicate the observed cumulative distribution curves of the noise levels.

Subjective noisiness

Table 2 displays the results of the factor analysis for each noise audible source and the mean noisiness and SD values of the hospitals. Factor 1, which can be interpreted as the “primary factor (noise),” shows the highest loading (eigenvalue = 4.64, percent of total variance explained = 4.64/12) to be primary. As a rule of thumb, an eigenvalue should be larger than 1.0 to be an effective factor in factor analysis. ¹³ The primary noise came from various events of equipment, such as noise induced by medical appliances, the broadcasting of announcements, and facilities. The factor loadings in Table 2 also indicate the correlations between the primary noise factor and the individuals events of noise, with factor loadings of more than 0.6 and less than 0.4 indicating good correlations. ⁴⁷ These were expressed as regular and continual noise events in the hospital lobbies. Factor 2 shows the highest loading for construction noise and can be regarded as “accidental noise.” Such accidental noise events can unexpectedly interrupt your conversation or train of thought and are a common occurrence in urban Taiwanese hospitals. However, factor 2 had a low “percentage of total variance explained” of 11.5 in comparison to that for factor 1 of 38.6. Even though the “construction noise” item had the highest factor loading over the 16 hospital lobbies, the incidence of dissatisfaction events aroused by construction was very low. These two factors showed that 50.1% of the variance in the noisiness could be explained, with the combined eigenvalues for the two factors being greater than 1.0 and each of the factor loadings being greater than 0.50. A reliability analysis yielding Cronbach’s alpha values was conducted to test the reliability and internal consistency of each factor.

The arithmetic means of noisiness scale were calculated using the subjective responses from the subjects in each hospital’s lobby and averaging the five-point scales for the 12 noise sources. Figure 3 shows the noisiness results, with the 95% confidence intervals noted by error bars. These intervals were calculated for the individual responses at each of the 16 hospitals; the variance in noisiness differed among the hospitals, even in those with equal noisiness ratings (e.g. hospitals J and K). The results of analysis of variance (ANOVA) tests showed significant differences in mean noisiness among the individual subjects across the various hospitals (F _{1,15, 0.05} = 3.72, p < 0.001). Figure 5 shows that the ANOVA results also indicated significant differences between the mean noisiness of the 12 noise sources, with the 95% confidence intervals across the hospitals also included (F _{1, 15, 0.05} = 4.85, p < 0.001). In Figure 5, the mean noisiness values of the 12 noise sources across hospitals with two independent samples (hospitals L and M) are plotted. The results for hospital L indicated that the survey respondents were highly dissatisfied with construction noise, since its sound levels were larger than the background noise by 8 dB. The noisiness scale results and their confidence intervals varying substantially among the 16 hospitals, and with the survey respondents for the two independent references (hospitals L and M) also indicated high levels of dissatisfaction with the various noise sources (e.g. facilities, broadcasting). These results indicate that the subjective noisiness scale used in this study for the 12 noise sources across the 16 hospitals was statistically reliable and significantly different in expressing the experienced noisiness ratings.

OPEN IN VIEWER

Figure 5.

The noisiness scale values with their 95% confidence intervals for each noise event across the 16 hospitals according to the questionnaire responses. Two extreme samples (hospitals L and M) are plotted.

Equivalent sound pressure levels

The results of both indoor and outdoor measurements, including their standard deviations, across the hospitals are illustrated in Figure 6. The correlation evaluation results for every independent single factor and the noisiness scales are indicated in Table 3, explained with the correlation coefficient (r) and the statistical reliability. The averaged L_Aeq ,_{15 m} was correlated with noisiness, but other measures were more closely linked to noisiness. For example, calculating τ_e , the NACF of all L_Aeq ,_{15 m} from 9:00 a.m. to 5:00 p.m. in order to detect the noise’s temporal repeated feature between forward and backward propagation signals, only the initial part of the normalized ACF (approximately 0 to −10 dB) showed clear decay for all the data. As indicated in Figure 7, the value of τ_e defined at the 10-percentile delay (–10 dB) was obtained by fitting the straight-line regression for (log(Φ _p (τ)) > −10 dB) of the ACF envelope for all the data. This procedure was similar to that used for measuring the initial reverberation time in room acoustics. Compared to the results of subjective noisiness, no other independent factor correlated with subjective noisiness could be found, as indicated in Table 3. We supposed that the statistical levels could be the best descriptor while De Coensel et al., ⁷ Gupta and Ghatak, ⁴⁹ and Madu et al. ⁵⁰ believed that their statistical evaluations of noise showed that the people who participated in their studies suffered from discomfort due to traffic noise.

OPEN IN VIEWER

Table 3.

Correlation coefficient (r) with p-value results between noise criteria and subjective noisiness results in the hospital lobbies.

Item	Correlation (r)	Fit (r 2)	Eliminate confounder
Noise level (dBA)
Average LAeq ,15 m	0.46	0.22
LAeq ,15 m standard deviation σ LAeq	–0.76***	0.58	–0.79****
L max	–0.13	0.02
L min	0.69**	0.48
Insulation of building against traffic noise	–0.37
Autocorrelation function (ACF) (h)
τe of NACF in LAeq ,15 m	0.76**	0.58	0.79***
Gradient of cumulative distribution function (h/dB)	0.76*	0.59	0.78***
Percentile LAeq values (LN ) (dBA)
L 5	0.33	0.11
L 10	0.56	0.31
L 50	0.69*	0.48
L 90	0.71**	0.51
Compounded indices
L min − L max	0.79***	0.62	0.81***
L 95 − L 5	0.74**	0.55
L 70 − L 30	0.53	0.28
τe /(L 5 − L 95)	0.89***	0.77	0.92***
L 90 − L 10	0.73**	0.53
Noise criteria
NC (noise criterion) 48	0.41	0.17	0.41
RC (room criteria)	0.45	0.20
Average loudness, N (sone) 24	0.64**	0.41
Percentile loudness, N 5 (sone)	0.72**	0.53	0.72**
N 10 (sone)	0.72**	0.52
Sharpness (Aure)30 s	0.45	0.20
Unbiased annoyance (UBA)30 s 51	0.71**	0.51	0.71**
T30 (500 Hz) (s)	0.56*	0.30	0.56*
Multiplication (architecture conditions)	0.64**	0.41

NACF: normalized autocorrelation function.

***p < 0.001, **p < 0.01, *p < 0.05.

OPEN IN VIEWER

Figure 6.

Logarithmic average L_Aeq values for the 16 hospital lobbies (for both indoor and outdoor measurements). The vertical bars indicate the respective daily ranges (standard deviations).

OPEN IN VIEWER

Figure 7.

The noise levels of L_Aeq ,_{15 m} in the hospital lobbies (sampled in hospitals D, H, and L) showing an initial decline in the envelope of the normalized autocorrelation function (NACF), and that this decline can be fit to a straight line regression in a range of 0 to −10 dB of the power of the NACF. The effective duration of the NACF of the noise levels (τ_e , h) is defined as it crosses to −10 dB at that of delay.

***p < 0.001, **p < 0.01, *p < 0.05.In many previous studies, the noise measurements made in hospitals did not determine whether the measured data had a normal distribution or not. As such, their results raise some doubts even with the definition of community noise. ¹² According to Sudarsono et al., ¹⁴ environmental noise is regarded to an increasing degree by people worldwide as relevant to the growth and sustainability of communities due to its adverse effects on quality of life, including the fact that excessive levels of exposure to environmental noise have negative effects on human health. In spite of this increasing recognition, however, most commercial and industrial processes are aimed at maximizing productivity at the lowest cost possible, with far less concern given to the levels of noise they produce. Relatedly, Sudarsono et al. ¹⁴ proposed that the effective management of noise would entail the achievement of the best compromise possible among the conflicting goals of the relevant stakeholders and potential options for noise abatement. In this view, it is important for hospitals to determine the views of their patients and other visitors regarding the noise frames in their lobbies. The subjective noisiness perceptions of patients and other visitors can, when utilized along with the appropriate noise exposure criteria, yield appropriate frameworks for environmental noise issues. Therefore, this study proceeded by examining the real statistical distributions of the noise for the different hospitals, while also not excluding the effects of confounding bias on the values of statistical factors.

Statistical levels

The variation of L_Aeq ,_{15 m} for all the measures throughout the daytime period (9:00–17:00) was found to be more important than the averaged L_{Aeq ,15 m} in the previous study. ⁹ As shown in Table 3, L _min − L _max, L ₉₅ − L ₅, L ₉₀ − L ₁₀, and the effective duration of the NACF (τ_e ) of all the measurements made at 15-min intervals were calculated and used as indices expressing the variation of the noise levels. The correlation coefficients (r) among the noisiness scale and the L _min − L _max, L ₉₅ − L ₅, L ₉₀ − L ₁₀, and τ_e values were 0.79 (p < 0.001), 0.74 (p < 0.01), 0.73 (p < 0.01), and 0.76 (p < 0.01), respectively. However, the correlation coefficient between the noisiness scale and the L ₉₅ − L ₅ in the nighttime was 0.39.

In addition, the variances in these sets of noise sources reflect the noise concentration trend, which is a hint aroused by the findings presented in the study by Kaczmarek and Preis. ¹¹ Therefore, due to the use of equivalent observation time periods for the noise events measured at two different noise levels, the distance between these two levels can reflect the rate of noise occurrence in a certain area. This reason is similar to that for the differences between two L_N values (L_{N 1} − L_{N 2}) in the same measured set, but these two points are not clear in a statistical probability curve. It is clarifying that an expected linear curve shown in Figure 8(b) was observed for the results measured in each hospital between the cumulative rate range of 0.3–0.7 (4.8 h). The gradient of the cumulative distribution function of the daytime noise events measured in the 16 hospital lobbies was close to the mean value of 1.12 h/dB. The gradient indicates the frequency of similar noise sources in a certain space, or, in other words, the continuity of noise. For example, as shown in Figure 8(a), the difference in the mean noise levels of hospitals A and F was only 3.25 dB. However, there was a remarkable difference in their gradients of the cumulative distribution function (1.326 and 0.894 h/dB), as the gradient of hospital A was 1.48 times greater than that of hospital F. As shown in Figure 3, the noisiness scale result in hospital A was 2.86, which was greater than that in hospital F (2.58). As shown in Table 3, the gradients of the cumulative distribution function correlated well with the subjective noisiness results, which were similar to (τ_e , NACF) for all the measurements, with the correlation coefficients (r) of both equaling 0.76.

OPEN IN VIEWER

Figure 8.

Normally expected distribution curves (a) and the probability curves (b) of daytime noise events in the hospitals, measured hourly from 9:00 a.m. to 5:00 p.m. (comparison of hospitals A, B, and F).

Even if the noise levels were not normally distributed, the L ₉₅ − L ₅ values could be easily determined before. This indicates that using the primitive L_N levels of responses to subjective noisiness without confirming their data normality is not conscientious with respect to statistical data integrity. ¹⁰ While the normal distributions of the measurements indicated good correlations among the statistical predictors and subjective noisiness, the probability plots of the noise level L_Aeq ,_{15 m} for each hospital were determined using the Kolmogorov–Smirnov test, ⁵¹ with the results being shown in Table 4. Using this procedure, we could conduct a normality test to prove whether the noise levels for each hospital were non-normal or not. Since a p-value less than 0.05 showed non-normality in our measurements, the test results found that the L_Aeq ,_{15 m} values for hospitals L and M were not unstable in the measures. To compare two nearly identical gradients of the cumulative distribution function (1.326 and 1.623 h/dB) of the noise levels, Figure 9 shows both the cumulative observed and expected curves of the cumulative distributions for hospitals A and L, respectively. Referring to the curve for hospital A, the probability of the cumulative noise level distribution of hospital L clearly indicates kurtosis at 66 dBA. This kurtosis of the observed curve tells us the height and sharpness of the level peak among the shape of the distribution, relative to that of a frequently occurring specific noise event that was recorded in the given location.

OPEN IN VIEWER

Figure 9.

Normally expected (dashed lines) and observed (dots) probability cumulative distribution curves of daytime noise events in hospitals A and L.

OPEN IN VIEWER

Table 4.

The results of normality testing obtained using the Kolmogorov–Smirnov test.

Hospital	A	B	C	D	E	F	G	H
Kolmogorov–Smirnov d-value	0.062	0.072	0.011	0.093	0.011	0.063	0.068	0.034
Chi-Square test, p-value	>0.15	>0.15	>0.15	>0.15	>0.15	>0.15	>0.15	>0.15
Hospital	I	J	K	L	M	N	O	P
Kolmogorov–Smirnov d-value	0.070	0.11	0.052	0.145	0.176	0.102	0.128	0.108
Chi-Square test, p-value	>0.15	>0.15	>0.15	0.044*	0.034*	>0.15	0.095	>0.15

*p < 0.05.

Therefore, the following question was raised: “Which non-normal distribution does the data best fit?” Individual distribution identification was conducted for the noise level L_Aeq ,_{15 m} measured in hospital L. Seven non-normal distributions were identified, and the results of a goodness of fit test are listed in Table 5. In this scenario, the three-parameter Weibull distribution fits the data best (p = 0.48), and it is extremely useful in most cases. However, the Weibull distribution is not an appropriate model for every situation reported by Weibull and Sweden. ⁵² It is important to identify how subjective noisiness does not fit a non-normal distribution of the noise levels during the daytime (9:00–17:00) measured in hospital L. As indicated in Figure 5, the subjective noisiness measurements for hospital L indicated the greatest dissatisfaction with construction noise among the 16 hospitals, while those made at hospital M varied harmoniously with the medium values. Although the construction noise events induced the highest factor loadings of noisiness in the results of a factor analysis (Table 2), five accidental noise events only explained 11.5% of the total variance. Eventually, the construction noise was identified as a confounding factor in the present research, meaning that it risks introducing bias into the subjective annoyance ratings provided by the survey respondents. As indicated in Table 3, if the correlation evaluations between noisiness and each of the noise indicators were restricted to eliminate the data of hospital L as the individual confounder of the distribution, the r values could be clearly raised for all the statistical descriptors but the noise loudness criteria.

OPEN IN VIEWER

Table 5.

The results of individual distribution identification obtained by goodness of fit test for hospital L.

Distribution	AD	p
Lognormal	0.436	0.280
Three-parameter exponential	14.233	<0.003**
Two-parameter exponential	4.627	<0.010**
Weibull	0.328	0.250
Three-parameter Weibull	0.314	0.482
Smallest extreme value	0.360	0.250
Largest extreme value	0.956	0.015*
Gamma	0.433	0.250
Logistic	0.412	0.250
Loglogistic	0.443	0.228

**p < 0.01; *p < 0.05.

It was also reconfirmed that the noise levels, as measured by L_Aeq at 15-min intervals in the hospital lobbies, are a sensitive descriptor of the subjective annoyance responses. From a statistical point of view, the shorter interval (<15 min) of noise level integration makes the distributed data more normal. Therefore, it would be weak in detecting annoyance feelings affected by any confounding effects.