Principles of Alarm Systems
A well-implemented clinical alarm system can save a patient’s life. During clinical care, the monitor’s alarm system is always vigilant and, when an alarm is triggered, provides a clear signal that a notable event has occurred. The signal can draw attention to the event and facilitate timely treatment. Alarms can cue action that can save lives; however, they are far from perfect. They function as a combination of technology and a human operator, built to capitalize on a machine’s ability to detect a change, and the human’s ability to interpret its meaning. While alarms alert the clinician to an event, they cannot distinguish between clinically relevant and trivial occurrences. They have very limited ability to (1) understand the context of the situation, (2) reason out the cause of a problem, and (3) determine clinical relevance. Because of these limitations, alarm systems are often plagued by high rates of false alarms. False alarms interrupt care during critical junctures and cause stress and frustration to care providers. By understanding the principles of alarm systems, a well-prepared clinician can manage them efficiently and thus maximize the value of the monitors, ease workload, and increase patient safety.
To understand alarm systems, we must understand the principles underlying the determination that an “event” has occurred. It is rarely possible to measure an event of interest directly. In normal circumstances, we measure some aspect of the environment that reflects the underlying state of the system. For example, during open heart surgery, one can see the heart beating directly, but we more often measure the electrical signals at a patient’s skin via an electrocardiogram (ECG). When a situation is “normal” (i.e., when no event such as a change in the patient’s state has occurred), measures taken from the environment tend to cluster around a particular “normal” value. Similarly, when a particular event has occurred, measures of the environment tend to cluster around a different value. In most systems, there is overlap between the values associated with a normal situation and an event situation ( Figure 13-1 ). Many values could indicate either an event or a normal situation. The value or threshold at which an alarm is sounded is called the decision criteria in signal detection theory, or “β” (“beta”). Given any reading from the environment, a monitoring system will make a determination of whether the parameter (patient) is normal or abnormal, and the resulting action by the monitoring system can be classified into one of four categories: a hit (true alarm), which occurs when the system detects an event when an event has occurred; a correct rejection (true normal), in which the monitoring system does not detect an event when no event has occurred; a false alarm, where the monitoring system detects an abnormal situation when the situation is in fact normal; and a miss, in which an abnormal situation arises, but the monitoring system fails to detect it ( Table 13-1 ).
| The Monitoring System Response | |||
|---|---|---|---|
| Sound Alarm | Do Not Sound Alarm | ||
| True state of the world | Situation is normal | False alarm | Correct rejection (specificity) |
| Situation is abnormal | “Hit” (sensitivity) | Miss | |
From the four states, we can derive two important characteristics of a monitoring system: sensitivity and specificity. Sensitivity represents the likelihood that when an event (abnormal situation) occurs, it will be detected (i.e., the “hit rate”) (equation 13-1 ). The specificity signifies the likelihood that when the situation is normal, no event will be indicated (i.e., the correct rejection rate) (equation 13-2 ).
Sensitivity = hits hits + misses
Specificity = correct rejections correct rejections + false alarms
The sensitivity and specificity of a system is determined by:
- ▪
The characteristics of the signal: Every signal has inherent variability, such as normal variations in heart rate or in blood pressure. Some signals have a very clear dividing line between normal and abnormal, such as the presence or absence of ventricular fibrillation (VFib). Others are more indistinct, such as end-tidal CO 2 levels or oxygen saturation levels. The extent to which the normal and abnormal states are distinct affects the overlap between the two curves in Figure 13-1 .
- ▪
The mechanics of the sensor: The mechanics of the sensor determine its ability to detect the underlying state of the system. Each type of sensor has inherent advantages and disadvantages. For example, invasive monitoring often is less susceptible to artifacts, and may provide a more direct measure of the underlying state than a noninvasive measure. However, invasive monitors are associated with higher risk of complications and cause insult to the patient.
- ▪
Mechanics of the artifact rejection: Monitoring devices are often hard-wired with the ability to sense degradation in the quality of the signal and can sometimes filter out noise that does not indicate a true change in patient state. The degree to which these filtering techniques are successful affects the specificity of the alarm system.
- ▪
Programming by the user: Adjustable alarm limits (decision criteria) can shift the proportion of hits to misses and, correct rejections to false alarms. These shifts occur without changing any of the inherent qualities of the alarm system’s ability to detect state changes. Changes in decision criteria should reflect the relative value placed on detecting an event as compared with missing an event or sounding a false alarm.
The process of determining the relationship between the measure and the state of the world is as follows. Alarm systems typically monitor a single value or monitor for signs of a single event. When the measure/parameter reaches a certain decision criteria, β, the alarm is sounded. A system’s effectiveness (hit rate, false-alarm rate, etc.) is determined by two factors. One is the difference between the two populations (i.e., how much overlap there is in the curves in Figure 13-1) , which is a type of signal-to-noise ratio known as the discriminability index, d¢ (“dee prime”). The second determining factor for effectiveness is the decision criterion (i.e., the point at which the alarm goes off, or β ). Setting the decision criteria determines the effectiveness of the alarm system, as much as the inherent difference between the population of “true events” and “normal situations.” As the decision criteria is made more stringent (higher threshold for an alarm), the number of false alarms decreases, but the number of misses increases as well. The discriminability index, d¢, indicates how easy it is to distinguish a true event from a nonevent and is a function of the characteristics of the parameter being measured and the technology used to measure it. The decision criterion β is set by the operator, and determines at what point an alarm will sound. For a given system, the value of β will determine proportion of true alarms to misses, and the proportion of false alarms to true normal reading. The value of d¢ will determine the proportion of true alarms to false alarms for a given β.
Concepts and Components of an Alarm System
A number of key definitions for basic concepts and terms that facilitate an understanding of alarm systems are listed next.
- 1.
Alarm annunciators: The auditory signal (bell, buzzer, chime), visual indicator (flashing display, light, marquee), or tactile signal (vibration) that indicates that an alarm has been triggered. Annunciators can be separate from the clinical monitoring device itself, such as with a pager, mobile phone, or hand held device.
- a.
Visual annunciator: The element of the visual display that indicates that an alarm is sounding. An annunciator can consist of text or graphical icons. Visual annunciators often are color coded in red or yellow, may also blink to indicate severity or attract attention, and may include highlighting on the monitor or even on a device separate from the monitor itself ( Figure 13-2 ).
Figure 13–2
Row A. (from left to right) Icons indicating a disconnected sensor (first), and for silencing and pausing alarms on monitor screens (second and third). Physical buttons for alarm management (fourth through end) may include backlighting or other visual annunciators to indicate that their status. Row B. Two levels of visual annunciation (first, second) for parameter alarms (third), as well as system-status alarms, highlighted (fourth) or nonhighlighted (fifth). Row C, Different styles of visual annunciations including border highlighting, parameter highlighting, and text highlighting. Row D, Visual annunciators may be installed separate from the clinical monitor, such as in hallway marquees on critical care or telemetry units. Similarly, phone or paging systems may be used for auditory, visual, and tactile annunciation of alarms. Row E, System-status indicators from an Ohmeda monochrome ventilator display. System-status alarms may be presented as plain text or highlighted and may flash to indicate priority level.
- b.
Auditory annunciator: A sound produced by the monitoring system to indicate that an alarm is sounding. The sound may include a pattern of or repetition of tones to indicate severity, such as single tone for low priority alarms, and repeated tones for higher priority alarms.
- c.
Haptic/tactile annunciator: A touch-based signal produced to indicate that an alarm is sounding. Most commonly, pagers in “silent” mode annunciate by vibrating.
- a.
- 2.
Alarm states: Terminology for alarm states can be confusing, with the original terms coming from early electronics systems in which an alarm would sound when a current in an alarm circuit ceased to flow, or “went off.” The terminology has remained, as an alarm “going off” indicates that an alarm is sounding. In clinical monitoring, there are terms describing the state of the annunciator, and the state of the alarm system as a whole.
- a.
Enabled/disabled: Enabled/disabled describes an alarm system, not an individual alarm’s activity. Enabling or disabling an alarm system indicates that the system as a whole will or will not detect events, respectively. When an alarm is enabled, it will detect events as they occur.
- b.
Sounding/not sounding: Sounding describes the activity of an individual annunciator. With an alarm system enabled and an event detected, an alarm will “sound.” An enabled system will not sound when no event is detected.
- c.
Alarm suppression: Alarms that are sounding can be acknowledged, and the annunciator silenced or suspended. In the “silenced” state, alarm states are actively detected, but annunciators not actuated. Alarms can be silenced or suspended for a brief period (e.g., 30 seconds to 2 minutes). Alarm systems can silence the auditory annunciators permanently, while leaving the visual annunciators active. Some alarms must be “dismissed” from a monitor (see “latched” alarms in Alarm Types section).
- a.
- 3.
Alarm Types
- a.
Latched: Once it is triggered, a latched alarm will continue to sound until silenced. For example, a burglar alarm sounds if a door is opened, and will sound until reset. In clinical monitoring, crisis-level alarms are often “latched.” When a latched alarm sounds, it must be “dismissed” through interaction with the monitor.
- b.
Unlatched: Unlatched alarms will sound only as long as criteria for sounding are met. An example is the auditory alarm that indicates that you have left lights on in the car—the alarm stops when the lights are turned off. In clinical monitoring, the pulse oximeter will sound an alarm when saturations drop below the alarm threshold, but the alarm stops when saturations return to normal.
- c.
Tiered: An alarm system that assigns a relative priority to the various alarms that may sound. Typically there are three to four levels of alarm annunciation, ranging from informational alerts and messages at the low levels, to crisis alarms at the highest level (see later discussion).
- a.
- 4.
Classification of alarm events: When an alarm sounds, it is typically classified into one of the first four categories below. The four terms are commonly used, but in practice the classification of alarms may not capture the subtlety of the clinical domain. The fifth category, “nuisance alarm,” can be an informal classification of a true or false alarm.
- a.
True alarm: An alarm caused by a change in the underlying state of the system being monitored. Note that because of the highly contextual nature of monitoring, a “true” alarm may have no clinical relevance (see “nuisance alarm”).
- b.
False alarm: An alarm that sounds when the underlying state being monitored is normal. False alarms can be caused by an increase in the “noise” contained within the monitored signal (artifacts) or can be a natural function of the decision criteria (alarm thresholds) set on the monitoring system.
- c.
Miss: A true event occurs but an alarm does not sound.
- d.
Correct normal: When the situation is normal (no event occurs) and an alarm does not sound. The term “correct rejection” is often used in signal detection theory to indicate the rejection of the hypothesis that a signal is present.
- e.
Nuisance alarm: Alarms that sound often or predictably, but provide no useful information to the practitioner. Nuisance alarms can be true alarms or false alarms. They can indicate a true change in the state of the patient or equipment that either has no clinical significance, or that may already be known, such as an apnea alarm that sounds during the process of tracheal intubation. When alarm thresholds are set inappropriately, nuisance alarms may indicate a trivial change that does not represent a threat to the patient. False alarms caused by artifacts such as electrocautery or patient movement can also be nuisance alarms.
- a.
- 5.
Characteristics of Alarm Systems
- a.
Sensitivity: The probability that a true event will be detected. For alarms, it is the probability that an alarm sounds when an event has occurred
- b.
Specificity: The probability that no event will be detected when no event is present. For alarms, it is the probability that an alarm will not sound when an event has not occurred.
- c.
False-alarm rate: The probability that when an alarm sounds, it does not indicate a true event.
- d.
Alarm limits, alarm thresholds: The value or level at which an alarm will sound. See also decision criteria, β.
- e.
Default parameters: Alarm limits that have been preset to typical, acceptable, standard values. Default parameters are at default values upon initiating clinical monitor, and typically return to default values when the patient is discharged from the monitor.
- a.
- 6.
Signal Detection Theory
- a.
Signal-to-noise ratio: The clarity of a signal, as measured by the strength of the desired signal compared with the random disruptions present in the signal. In clinical monitoring, it can relate to the relationship between the underlying physiology being measured and the measured parameter representing it.
- b.
d′ (“dee prime”): The discriminability index or sensitivity index indicating how discernable a true event is from a nonevent. This is a specific type of signal-to-noise ratio determined by the characteristics of the monitored parameter and the technology used to monitor it. The value is calculated using the magnitude of the difference in the parameter’s value between normal and abnormal states (separation), and the variability in each state (spread).
- c.
β (“Beta”): The decision criteria. The threshold at which a monitored parameter is considered abnormal. It is determined by the operator of the system. In clinical monitoring, changes in the β are expressed as changes in alarm limits, which determine the proportion of false alarms to true alarms, and the proportion of misses to correctly identified nonevents.
- d.
Base rate: The prevalence or frequency of a condition or event within a given population, often expressed as the probability of an event occurring. In clinical monitoring, it may be the a priori probability that a particular clinical condition will arise during a given case or epoch.
- a.
Clinical monitoring systems indicate more than the presence or absence of an alarm. They additionally classify its severity. In most systems, a tiered or graded classification of events is used, where alarms are classified into classes such as message, advisory, warning, or crisis. Each class of alarm may have a unique annunciation pattern ( Table 13-2 ). Such tiered responses can help clinicians respond appropriately to the urgency of an event ( Table 13-3 ). Alarms at a particular tier can trigger automatic actions in the monitoring system. All crisis alarm events, for example, may initiate permanent archiving of the alarm log in the central monitoring database, while warnings would be stored only until a patient is discharged. Arrhythmia alarms can trigger automatic printouts of ECG strips.
