Noise is measured using a logarithmic scale of dB. The threshold for normal human hearing is 0 dBA, a quiet room or a whisper is about 30 dBA, normal conversation is about 55 dBA, a television generates about 60 dBA, heavy traffic at 10-m distance is about 80 dBA and a pneumatic drill is about 100 dBA. A 3-dB change in noise level is considered just discernible; a 5-dB change is clearly discernible; and a 10-dB change louder or softer is perceived as a doubling or halving of volume, respectively. For speech to be easily intelligible, it needs to be 15 dB above background noise levels. Thus the recommended WHO average levels for hospital wards are the equivalent of a very quiet room with transient peaks at night well below conversation level.
Although it has been reported that there is no significant reduction in overnight activity in the ICU [5], the link between sleep deprivation and poor outcome has been well-reported in recent years [6–8], and all five units in our present study routinely decrease overnight activity and lower the unit lighting to encourage natural sleeping patterns. The noise levels certainly drop by about 5 dB in the early hours of the morning, but only to the level of continuous conversation. The beginning and end of the night are characterised by obvious increases in noise levels at handover time (see Figures 1 and 2). On average, there were approximately 25 minutes of every hour during the day when peak levels above 85 dBA occurred. Peak levels above 85 dBA occurred less frequently overnight, but a patient can still expect to be disturbed at least once every 7 to 16 minutes of every hour between 10:00 PM and 7:00 AM (Figure 4). At these dB levels, it is highly likely that this is alarm activity, and, as has been reported elsewhere [9], electronic sounds are more arousing than human voices, so they are very likely to continually disturb patients' sleep. Frequent and persistent arousal has been shown to have negative effects for both healthy volunteers and patients [10, 11].
Hospitals generally appear to be getting noisier over time. A review of published data over the past 50 years [12] suggests an average increase of 15 dB since the 1960s, more than a doubling of the perceived noise. The same study looked at noise in multiple hospital locations in an American teaching hospital and demonstrated the highest levels in the paediatric ICU, although there were no recordings from the adult ICUs.
It is immediately obvious that sound levels in the Thames Valley Intensive Care Units are considerably higher than the WHO guidelines recommend. The patients are subjected to a continuous level of sound which, at best, is only a little below conversation level and during the day equates to a nearby television or dishwasher. At no point during any of the measurement sessions did the LAeq near the patient fall below 50 dBA. Peak levels (measured as LApeak) were always above 60 dBA and at worst were almost 128 dBA. In previous studies conducted in specialist ICUs, average levels were about 10 dB higher in a Turkish cardiac surgical ICU with a similar time profile [13], similar to our results in a two-bed Swedish neurosurgical ICU with a comparable frequency distribution of peaks [14], and 5 to 10 dB higher in an American paediatric ICU with no diurnal variation [12].
Given the physical and environmental differences in the selected units, it was perhaps surprising that the data generated were so alike. It might be expected that the single-room ICUs would be louder than the three-room John Radcliffe Hospital Adult ICU, but this proved not to be the case. The quietest unit was also not the unit with the lowest number of patients during the recording period. This suggests that noise level is associated with more than simple acoustics and occupancy.
LAeq values were between 51.3 and 59.1 dBA at the central station and 54.1 to 59.9dBA at the patient location. The sound level adjacent to the patient's head was almost always greater than that at the central station. This is probably due to the way equipment is positioned. All the units use pendant or rail systems to suspend the equipment adjacent to the patient's head on each side of the bed, allowing unhindered access to the back of the patient's bed. Whilst this is both convenient and conventional, it does put noise sources close to patients' ears. In all the units studied, the ventilator was positioned at one side of the head of the bed and the monitor was placed on the other side, so these two sources of noise often were 50 cm or less from patients' ears and a similar distance from the recording devices. The noise generated by functioning equipment and alarms seems to be considerable, as evidenced by the levels recorded when the equipment or alarms were recorded in isolation. All units provide patient entertainment (television and/or radio), and it is possible that their use contributed to the slightly higher values recorded at the patient location; however, we did not record television and radio use during the assessment period. We also did not record patient intervention activity, which may have increased noise levels in the bed space.
The frequency components recorded during the day and at the quietest period show a different pattern from that reported in an adult ICU [14] and a paediatric ICU [12] in that there was much less noise below 400 Hz. This is probably simply an effect of the weighting used. We used A-weighting, which is less sensitive to lower frequencies to approximate human perception, which reduces the level at both ends of the frequency spectrum, whereas previously reported results were based on unweighted (absolute sound level) measurements. If the A-weighting is removed, the results more closely resemble the earlier adult ICU results [14]. The change previously noted during the quietest part of the day, with a reduction in sound levels predominantly above 400 Hz [14], was also seen in this study. This may be because much of the lower-frequency sound may be caused by hospital plant and other factors that do not show diurnal variation. By contrast, the higher frequencies, where conversation, alarm sounds and the like are found, do decrease at night. It would be interesting to repeat this exercise at a time of year when the day-night light durations are different to compare seasonal diurnal effects.
A comparable study recorded sound levels in an outpatient chemotherapy clinic [15] and found similar, constant, average (55 to 60 dB) and peak (>90 dB every minute) sound pressure levels during the day. Concurrent questionnaires completed by patients, visitors and staff revealed that, whilst staff felt that the noise was disruptive, in particular causing difficulties with communication, neither patients nor visitors were concerned. Although this suggests that levels seen in the ICU may be acceptable, the authors of that study found a correlation between the time an individual spent at the clinic and the level of irritation expressed. Thus the levels measured in our investigation are likely to affect both staff and patients in the ICU, and attempts should be made to lower noise levels.
The frequency spectra of the alarm sounds were recorded in an attempt to distinguish alarm sounds from background noise, but the acoustic 'signature' of the alarms was difficult to distinguish from the broadband background noise. As a result, we could not consistently measure alarm and non-alarm sounds separately. However, these sounds have been reported elsewhere [5], and it is clear that a significant proportion of the background noise is probably generated from modifiable behaviour such as conversation, operating and moving equipment, telephone use and allowing doors and container lids to close freely. A number of studies have reduced LAeq levels in the adult ICU, at least for a limited period of time, by introducing noise awareness initiatives and unit-level behavioural changes [16–18]. Introducing 'quiet times' has also been shown to improve general well-being [19] and sleep patterns when synchronised with natural circadian rhythms [18].
Three previously reported studies [20–22] used continuous polysomnography alongside environmental noise measurements to determine whether noise could be the reason for irregular sleep patterns in ICU patients and reported that environmental noise caused between 11% and 17% of arousals and awakenings. In interviews after ICU discharge, patients regularly reported disturbed sleep, attributing this to noise, light and frequent nursing interventions [23–28]. Sleep disruption in the ICU is also associated with increased requirements for anxiety and depression treatments [28]. Volunteers exposed to a simulated ICU environment show disturbed sleep and biochemical markers of stress [29], and two studies [25, 30] used the ICU Environmental Stressor Scale to assess patient experiences in the ICU. Both studies reported that patients identified alarms as a source of stress. The problem of environmental noise is not limited just to patients; high levels of noise on an ICU have been associated with increased levels of stress for staff [14, 31, 32]. Studies outside the hospital environment have demonstrated that noise has a negative impact on physiology [33, 34], motivation and general health [35].
Mechanical measures to reduce perceived sound levels, such as earplugs or ear defenders, which each reduce perceived noise by 15 to 30 dB, have also been shown to be effective. A recent 136 patient, randomised controlled study in a large Dutch mixed-use ICU showed a dramatic reduction in delirium and an improvement in sleep with this simple intervention [36]. An earlier, smaller US study in a general ICU and a cardiac ICU showed subjectively reported sleep quality was improved with the use of earplugs [37].
Discussions with ICU staff during our data collection period revealed that many we spoke to considered some patient monitor alerts to be disproportionate to their urgency, which led to louder sounds being prolonged while more immediate needs were treated. This inappropriateness in the alarm 'urgency mapping' [38] may quickly lead to desensitisation [39] and a corresponding reduction in alarm response. Alarm fatigue has been cited in a recent report as the leading hazard faced by hospitals in the United States [40]. Visual correlation of the data recorder real-time screens with alarm sounds confirmed that equipment alarms were the likely source of at least some of the peak values. It has been shown that active alarm management can reduce the total number of alarms. A study in the United States [41] introduced a programme by which staff were encouraged to modify machine default limits in line with their patients' individual physiology, thus reducing the opportunity for alarm fatigue to become established. Additionally, the development of smart alarms has been advocated [42–44]. In 2009, Gorges et al. [45] reported that only 23% of the alarms in the ICU were 'effective', specifically suggesting that introducing a 19-second delay would eliminate 67% of the ignored and ineffective alarms.
Research is ongoing to improve the system by which patients whose condition is deteriorating are identified [44, 46–48], and, although not in widespread use in the United Kingdom, there are alarm management systems which can transfer the audible alert from the patient bedside to a centralised control room or to the care provider. There may therefore be technological solutions that could be used alongside awareness programmes to lower sound levels by more than that which can be achieved by behavioural interventions alone.
We could achieve sound levels within the WHO guidelines only in a closed side room with all patient monitoring equipment switched off. Although some studies have found that it is possible to lower noise levels, at least temporarily, none achieved levels below the WHO guideline limit. Our findings suggest that, with the current equipment required for patient care, the WHO guidelines are not achievable in ICUs in the United Kingdom.
Limitations
Our study was limited to one day of recording at each site. One full week at each site would have provided more robust data less susceptible to short-term events, which might have affected the sound levels recorded on any given day. We did not collect information on patient sleep assessment or document activity around the patient bed space (for example, treatment and interventions or visitor and/or patient use of television and/or radio), which may have contributed to the noise levels in the patient's vicinity. A more accurate description of the sources of the noise may have been possible with more frequently sampled data, combined with greater frequency discrimination. This would have enabled us to run more detailed analysis of noise levels, particularly with regard to the number of peak levels and their duration.
