Medical audible alarms: a review

Abstract

Objectives

This paper summarizes much of the research that is applicable to the design of auditory alarms in a medical context. It also summarizes research that demonstrates that false alarm rates are unacceptably high, meaning that the proper application of auditory alarm design principles are compromised.

Target audience

Designers, users, and manufacturers of medical information and monitoring systems that indicate when medical or other parameters are exceeded and that are indicated by an auditory signal or signals.

Scope

The emergence of alarms as a ‘hot topic’; an outline of the issues and design principles, including IEC 60601-1-8; the high incidence of false alarms and its impact on alarm design and alarm fatigue; approaches to reducing alarm fatigue; alarm philosophy explained; urgency in audible alarms; different classes of sound as alarms; heterogeneity in alarm set design; problems with IEC 60601-1-8 and ways of approaching this design problem.

Alarms are important in medical informatics because they are typically used when a physiological, patient, equipment, or some other variable exceeds a previously determined threshold. Significant patient safety-related organizations have listed clinical alarm safety as one of the top priorities,^{1 2} and there is huge interdisciplinary effort going into dealing with the issue. The aims of this review are threefold. The first is to give a broad overview of some of the key issues of concern in the design, implementation, and standardization of audible alarms in a medical context. The second is to highlight the way the key issue in medical alarms—false alarms—hinders the application of the evidence base relating to the design of the alarms themselves. The third is to demonstrate the relevance and shortcomings of an important medical alarms standard in relation to that evidence. The topics selected for discussion reflect the aims through sections designated ‘false alarms’, ‘audible alarm design’ and ‘IEC 60601-1-8’, the standard in question. The fourth section, ‘alarm philosophy’, is included as it is concerned with the way designers and people who implement alarms think about how the alarms will function in the environment for which they were designed. The arguments presented, because they are evidence based, can be applied to all audible medical alarms, whether they are used on large pieces of equipment in hospitals, more mobile equipment that might be used on home visits or ambulances, by the patient in the home, or even on mobile ‘apps’.

When talking about alarms, we sometimes mean a whole alarm system, which would include the entirety of equipment, sensors, monitors, displays, audible alarms and their interactions with clinicians and patients; we sometimes mean the conditions that trigger those alarms, such as an unacceptably low temperature; and sometimes we mean the actual alarms themselves, the signals that annunciate the conditions to which we should attend. In this article, I am considering only the last of these, discrete medical audible alarms, the sounds that signal when some clinical parameter in the patient has been exceeded.

The author has extensive experience in carrying out peer-reviewed, published research on the topic of both audible and visual alarms covering 25 years (50+ journal articles, three books, >100 conference papers), has served on several alarms standards committees as a technical expert and has designed sets of alarms for clients ranging from aviation through rail, nuclear power stations, and medicine. The selection of the topics of this paper was done both on the basis of their apparent importance in the literature, and the author's experience of working in this area as a researcher and a designer.

The review presented is narrative rather than systematic, as it focuses on these four main topics. However, a comprehensive search strategy was carried out that included searches of medical literature databases (eg, PubMed), as well as both broad (Web of Knowledge) and more specific relevant databases (PsychINFO). In addition, the gray literature was searched. Finally, a manual search of key journals in the human factors and medical area (particularly anesthesia) was carried out.

Sanderson et al³ draw our attention to the reality that successful alarm design requires a whole range of expertise. This ranges from medical expertise, through human factors, cognitive engineering, psychoacoustics, to the standards process. Each of these key areas have sub-areas within them that are sometimes themselves vast in scope but impinge heavily on alarm issues. We may not even be aware of the existence of some important issues. For example, we do know that the underlying intelligence of many alarms is not as good as it might be,⁴ but we may not be aware that human decision-making can be flawed and subject to bias,^{5 6} regardless of how good the information being given to the clinician might be. In a different but related domain, the design of alarm sounds themselves might appear on the surface to be a relatively straightforward task. However, evidence relating to the alarms currently supporting IEC 60601-1-8,⁷ a broad standard concerned with medical alarms, suggests that appropriate consideration of the design of the alarm sounds has not been given, which has resulted in less than adequate alarm sounds.^8–13 Proper design of alarm sounds at a cognitive and perceptual level itself requires input from a large range of sources, and can result in improved alarm implementation.^14–20

Some pieces of information are more specific. For example, it has been demonstrated that contour (shape, rather than specific tones) is the defining feature of short melodies.^{21 22} The implication of this is that short sequences with similar contours will be confused with one another. Yet IEC 60601-1-8 contains short, three-tone melodies, some of which share the same contour. These have been shown to be readily confused with one another.^{9 12} Therefore, if tonal alarms are used, then research into music perception and cognition also has relevance.

Aside from requiring input from a vast range of sources, alarms issues are hampered by progress in some areas being more advanced than in others. For example, there is a large database of knowledge on the relationship between sound design and urgency, both for non-verbal alarms and for speech.^{15 17 23–30} Appropriate application of this work could readily lead to a properly urgency-mapped ‘master’ alarm varying only in its urgency according to the varying urgency of the situation(s) it is indicating. However, the underlying intelligence of alarm systems is generally not sophisticated enough to allow this to happen. If this ‘master’ alarm was designed according to the principles of Patterson¹⁹ then it would also be localizable and resistant to masking. It would not be necessary to have a range of different alarms because the mapping of alarm urgency to the underlying alarm algorithm(s) would allow the observer to have awareness of the developing situation, in a sonification-like manner.^31–34 (Sonification is a process whereby monitoring is carried out via continuous auditory feedback, as exemplified by Watson and Sanderson).³⁵ The current status of alarm engineering thus compromises the usefulness of the urgency principles that might be used. For example, while there is regular reporting that the urgency of alarms does not match the urgency of the situation,^36–38 and (quite appropriate) statements to the effect that it is the situation, and not the alarm, that will determine urgency and that situational urgency is context specific,^{24 27 39} it is still true that if the underlying intelligence of alarm systems was good enough, and the alarm urgency algorithm was good enough, there could be appropriate urgency mapping between an alarm and the situation it represents.⁴⁰

False alarms

Many studies cite the high prevalence of false alarms in clinical settings.^41–48 This situation appears not to have changed much over the past 25 years. For example, O'Carroll⁴⁴ reported that of 1455 soundings of alarms, only eight were associated with potentially life-threatening problems. In 2011, Talley et al⁴⁵ reported that false alarm rates in cardiopulmonary monitors range from 85% to 99%, with very few alarms indicating serious clinical events.

The problem of false alarms has led to the phenomenon of ‘alarm fatigue’,⁴⁶ and is important because, aside from the general noise and irritation caused by false alarms (which might be reduced if alarms sounded somewhat different), people tend to match their response rate to alarms by the perceived accuracy of the alarms. Bliss and colleagues^49–51 have demonstrated that if the accuracy rate of an alarm is 90% then people will respond slightly more than 90% of the time, whereas if the accuracy rate is 10% then they will respond only just over 10% of the time.

One of the major causes of false alarms in medical devices is that the alarm algorithms underpinning the coding and interpretation of the signals from patients are poor and not as technologically advanced as they might be. Imhoff and Kuhls⁴ reviewed the types of algorithms that can underpin medical devices and studies that have attempted to evaluate those algorithms in terms of false alarms. Almost always, improved or new algorithms tend to reduce false alarm rates,^{42 52} although this knowledge is not always taken up by manufacturers, which appears to have been slow but for which few data are available (Imhoff and Kuhls, p. 1533).⁴ The best designed alarm in the world will not work unless it is supported by good signal detection and interpretation.

Significant practical effort is now being put into alarm reduction on a case-by-case basis. A methodology that has had considerable success⁴⁶ demonstrates a step-by-step approach to reducing alarms to acceptable levels. These steps include a safety assessment that includes carrying out a task analysis on alarm notification and response processes, alarm setting, and improvement planning, which aims at developing strategies such as alarm tailoring, the implementation of technological solutions (such as centralized alarms) and so on. This methodology has resulted in quite startling effects, reducing the number of alarms from very high levels to numbers such as four a day. Also, introducing a delay has much more of an impact on reducing ignored or ineffective alarms than it does on effective alarms.⁵³

Alarm philosophy

One of the key conceptual issues with alarm design is the issue of an alarm philosophy.⁵⁴ The alarm philosophy is concerned with thinking about the alarm set as a whole, and will include issues such as mode of delivery of alarms, the assigning of alarms to functions, the mapping of alarms to priorities, the design of the individual alarms and so on. The development of an alarm philosophy has been central to many alarm design projects in areas other than medicine⁵⁵ as well as in the area of alerting in electronic prescribing.⁵⁶ Some of the features that an alarm philosophy should possess are as follows. Alarms should be kept to a minimum, and the way in which this is achieved is partly determined by the assignment of alarms to the situations considered worthy of an alarm. Alarms should be assigned to a particular modality or modalities (usually vision, hearing, or touch)^{57 58} on the basis both of the a priori matching of the information to be conveyed by the alarm to the natural predispositions of the senses,⁵⁹ and on the basis of detailed analysis of the work domain and the appropriate capture of attention given that work domain.^{35 60 61} One-to-one mapping of alarms to functions should be an aspiration for high-priority situations, with lower priority situations requiring only priority coding and thus a single priority-specific alarm can be used. Another feature of an alarm philosophy should be that of hazard matching, so that higher priority situations have more urgent alarms. A further key feature of an alarm philosophy is that standardization should occur. The key features necessary in an alarm philosophy can be seen on the UK's Rail Safety and Standards Board website.⁶² Although this has been designed for the train cab, the same general principles apply to the hospital and indeed almost any environment where alarms are used. Generating an appropriate alarm philosophy is an onerous task for any environment, and requires detailed thinking about the activities of and demands on the people working there. Even for a single, relatively unchanging environment (such as a train cab) the demands can be large, but for environments where equipment and equipment set-up changes regularly, as is the case in many medical environments, the task is almost infinite in scope, although there are successful demonstrations here.³⁵ To some extent IEC 60601-1-8 (2006) possesses an underpinning alarm philosophy that has its roots in an earlier approach to thinking about the way auditory alarms are used in anesthesiology.⁶³ The principles set out in this paper can be applied to any environment where medical alarms might be used. It would be important therefore to consider the limitations of that environment when the alarm philosophy is developed. For example, if alarms are being designed for a device that is mostly used by patients in the home, or even on an ‘app’, the fit of any alarms to that environment should be considered. For example, whereas speech alarms might be of little use in a noisy hospital environment, they might be used to great effect in a quieter, home environment. There is little scope for misunderstanding if speech is used.

Audible alarm design

High false alarm rates often tend to drive practicable solutions. Sanderson,³⁹ for example, argues that until alarms can be made to be more reliable, then continuous monitoring via sound, with critical states being indicated by evolving sounds constantly monitored by the user at a low, ‘preattentive’ level⁶⁴ might be a more useful approach to providing auditory displays and alarms in anesthesia. The sonification of respiratory parameters of Watson and Sanderson³⁵ demonstrates this approach. In addition, some⁶⁵ argue for ‘softer’ alarms than the ones typically in use at the moment in order to ameliorate the hazardous side-effects of false alarms. Even in an ideal world where the underlying alarm intelligence could be relied upon, designing alarms that are shrill and irritating is unergonomic as such alarms are aversive and will interrupt communication at the very point where people need to communicate.^{19 66}

One of the key reasons why alarms will be aversive and distracting, rather than attention-getting, is that they may be too loud. Setting alarms to an appropriate level given the noise background is a key component of the design principles of Patterson,¹⁹ and software has been available for 20 years that allows the user to assess alarms in terms of their detectability in the relevant background noise,⁶⁷ although this appears to be rarely used in practice.

From a cognitive and perceptual viewpoint, it is interesting that while alarms seem to cause recognition problems, we mostly go about our daily business while exposed to hundreds of sounds, but know their meaning and can act accordingly. A major contributing factor to this apparent ease of recognition is that the sounds that we hear in our everyday environment are usually the product of the objects or events making those sounds. Therefore, for example, we understand that a car might be just about to crash because we hear the tires skidding, or we know that someone is having difficulty breathing by their breathing patterns. In semiotic terms^{68 69} these are signs, and there is direct mapping between sound and event. Using actual sounds (called audification or amplification) is one design approach that will be useful for some applications.³ At the other extreme there are sounds for which the relationship between sound and meaning has to be learnt, because it is an artificial relationship (such as that between most abstract alarm tones and their meaning), and in between there are sounds that in some metaphorical way represent the object or event that they are signaling. The strength of the relationship between the sound and the object or event it is representing (in the case of alarms, the relationship between the alarm sound and the event it is signaling) seems to determine the ease with which sounds can be learnt.^70–72 Petocz et al⁷⁰ argue that learned associations can be just as meaningful as direct links between sound and event, so sounds that are overlearned can be as readily interpreted as those for which there are direct sound-referent links.⁷³ One example of a learnt association between an abstract sound and an event is a doorbell, and the most obvious example is that of speech. Although abstract, we are able to develop very clear relationships between sounds (words) and the objects and events that they represent through the process of language acquisition. The practical implication of this is that some sounds will a priori be easier to learn than others, but with learning and exposure almost any sound might, in principle at least, function as an alarm sound. The influential field of ecological acoustics, which is concerned with direct perception of invariance and change in complex acoustic stimuli (typical of those to which we listen in the real world) rather than the relatively simple, discrete sounds that have typically been used as the basis of audible alarm design, is arguably a more useful way of approaching the design issue because it considers real-world perception, which of course does not go away just because clinicians are functioning in a clinical environment. There are studies that have looked at the application of ecological acoustics to alarm design.^74–76

Combining alarms

How all the alarms in a single environment are put together is one of the keys to a successful alarm set. Different types of sounds that might be used as alarms have specific benefits and disadvantages. Table 1 presents a classification of types of sounds potentially of use in the design of discrete alarms. The table indicates the key acoustic and cognitive issues associated with each class of sound. The reader is also referred to the UK's Rail Safety and Standards Board sound library, which demonstrates examples of these sounds.⁶² This table differs somewhat from that of Sanderson et al,³ particularly in that table 1 here includes traditional and modern abstract alarms, which abound in the medical world. Blattner et al⁷⁷ classify those alarms as earcons. Also, audification and sonification are not included as this review specifically concerns discrete alarms, whereas sonification and audification are concerned with giving continuous auditory feedback.

Table 1.

Examples of sound classes used as discrete auditory alarms noting key acoustic and cognitive issues

Class of sound	Acoustic issues	Cognitive issues
Traditional abstract	Shrill, aversive, disruptive	Sound ‘alarm-like’
For example, bell	Can be heard through noise	Can become associated with meanings through learning
		Can be hard to learn initially but meaning evolves through use
Modern abstract	Can be tailored to acoustic environment	Sound ‘alarm-like’
For example, ‘ping’	Some are amenable to urgency manipulation	Difficult to learn and can be difficult to discriminate
	Typically more aversive than
	Is necessary (eg, high pitched, monotonous)
Tonal/Patterson	Can be tailored to acoustic environment	Music perception issues
For example, IEC 60601-1-8	Amenable to urgency manipulation	Amenable to coding
		Difficult to learn and can be similar to one another
Auditory icons	Enormous variation in acoustic structure as they are typically everyday sounds	Meaning is related to sound
For example, heartbeat		Do not sound ‘alarm-like’ but can represent alarm events
		Easier to learn than most other alarms
Speech	Sometimes difficult to tailor to noise environment	Usually easy to understand
		Can be indiscreet
		Easy to learn

Certainly some types of alarms are easier to learn than others (eg, speech and auditory icons are easier to learn than abstract alarms); those that are more difficult to learn are probably better in terms of their discreetness; some are better in terms of their acoustic robustness; and some will be more acceptable by users than others. If users do not like a set of alarms, then they will not adopt them, no matter what their qualities might be.

If it is important that alarms are easy to learn, then auditory icons or speech would be the best choice. If acoustic robustness is desirable, and we want alarms to sound ‘alarm-like’, then we might make different choices and use some abstract or traditional alarms. Determining which class of sounds to use for IEC 60601-1-8 is a directly relevant issue here, and this is discussed below.

IEC 60601-1-8

IEC 60601-1-8 concerns medical device alarms and is wide in its scope. It is an international standard aimed at harmonizing medical device alarms, and specifies basic safety and performance requirements for medical equipment alarms. Although compliance with the standard is voluntary, non-compliance by medical device manufacturers is unlikely for fear of litigation.

In its current format, it possesses many elements that represent good practice in terms of alarm implementation, although there are areas where this could be improved. Sanderson et al¹¹ discuss the history of the development of the standard to its current form. Originally a set of alarms based on the design principles of Patterson¹⁹ were designed to support an earlier standard.⁷⁸ Although these alarms were tonal and became thought of as melodies, the way they were specified in recording (which was done for ease of demonstration) did not do full justice to Patterson's principles in terms of overall alarm structure, variation in urgency of alarms over time and so on (for details of this structure see Edworthy).⁶⁶ The individual bursts of sounds had variation in them in terms of timbre (sound quality), temporal pattern and pitch pattern, all of which improve the variability from alarm to alarm, which should aid both differentiation and learnability.⁷⁹ The alarms currently specified in IEC 60601-1-8 each possess the same number of pulses and are uniform in rhythm. This makes distinction between the alarms more difficult than it need be. One of the most clearly established phenomenon in all of psychology is that people's ability to remember items and to differentiate between them depends on both their number and on the ways in which they differ from one another.⁸⁰ By virtue of having the same number of pulses (three for cautionary and five for emergency) and the same temporal pattern throughout, the alarms specified in IEC 60601-1-8 are too uniform and will therefore be difficult to tell apart. Such is the concern with the IEC 60601-1-8 alarms that the originator has issued an apologia¹⁰ and has speculated that the set designed by Patterson et al⁷⁸ may have been more acceptable than originally thought, though this claim remains untested.

IEC 60601-1-8 is underpinned by an alarm philosophy first set out by Kerr,⁶³ which is based on patient harm and which has a small number of organ-based categories. Basing the philosophy on the patient makes the strategy future proof and also prevents proliferation of alarms in principle. It may be that an alarm philosophy based on organs is not the best strategy, but determining the most appropriate basis for an alarm philosophy is the domain of medical practitioners and experts. In any case, proliferation of alarms is never a good idea. The more alarms there are, the longer it will take to learn them, there will be more potential for confusion between alarms, competition from alarms will mean that they get louder and louder, and the potential for irritation and stress for the medical staff and patients also increases. There will also be other categories of sound and alarm-type sounds that might need to co-exist with these alarm sounds (such as pulse oximetry signals, pagers, cellphones, and so on) but the aim should be to keep the number of alarms to a minimum. In its current form, however, the standard has nothing to say about signal detection and alarm triggering,⁴ and there are issues concerning the alarms themselves, which research evidence suggests are less than ideal.

It is not just the design of the alarm sounds themselves that matters, but the differences between them that will also determine the success of the alarm set. On the one hand it would be possible to design a set of alarms that could be learned very easily. This would be achieved by designing an alarm set that were either auditory icons or speech. However, neither may be acceptable to users. Given this, making the alarms as different from one another as possible might be another design approach.

Conclusions

False alarm rates are being reduced through developments in underlying signal extraction and the application of practical methods such as introducing delays between the detection of signals outside clinical parameter limits and the onset of the alarm itself. Reduction in false alarm rates means that effort put into the design of audible alarms will have greater efficacy. Software and principles concerning detection and audibility could also be employed more regularly in the design and specification process. The range of the types of sound that can be used as alarms varies from everyday sounds (auditory icons) to tonal alarms with several hybrid types of sounds between. The choice of alarm type will depend on the context in which an alarm is used and, depending on the type selected, research evidence concerning ecological acoustics (auditory icons) and urgency mapping (tonal alarms) can be applied to make the resultant alarms ergonomic and user friendly. It is important to note that there are advantages and disadvantages associated with particular design choices, which need to be considered in the context of the particular application. Alarms that are shrill and irritating do not constitute good design.

Alarm philosophies, frameworks for thinking about how alarms might be implemented in specific instances, should be employed. Here, decisions as to what to alarm about, what priorities specific situations should have, how alarms should be mapped to those functions, what modality to use (vision, hearing, touch), and what the resultant alarm should sound like (in the case of audible alarms) should all be considered.

Standardization of alarms represents good practice in principle because it prevents proliferation of alarms and makes learning of those alarms more straightforward. However, it is important that any alarms indicated in standards are underpinned by evidence and best practice. It is not clear that this is the case for IEC 60601-1-8.