Alert Fatigue

Abstract

Alert fatigue is a form of mental fatigue — which MeSH classifies under behavior and behavior mechanisms — in which repeated exposure to safety alerts desensitizes an operator, slowing or suppressing the response to each new warning. It is studied most intensively in health personnel, who face a torrent of clinical-decision-support and monitor alarms, the large majority of which prove false or clinically irrelevant. This article develops it as a cognitive problem with three converging mechanisms: habituation, the vigilance decrement, and a signal detection account in which a warning's low positive predictive value makes overriding rational rather than careless. It traces the cry-wolf effect of false alarms, the dose-response between alert volume and override rate, and the design remedies that outperform exhortation. Three interactive demonstrations let the reader explore predictive value, habituation decay, and alert burden.

Keywords: alert fatigue, habituation, signal detection, positive predictive value, vigilance decrement

Alert fatigue is where three old threads of cognitive psychology — habituation, vigilance, and signal detection — are braided into a single, costly, modern failure. A clinician who dismisses a drug-interaction pop-up without reading it, a nurse who silences a monitor alarm before it finishes sounding, and a radiologist who no longer notices a banner that fires on every study are all displaying the same thing: a learned, automatic down-weighting of a class of signals that once commanded attention. The behaviour looks like carelessness and is routinely blamed as such, but the cognitive account shows it to be the predictable output of normal mechanisms fed a diet of mostly false warnings. Understanding it means asking not why operators are negligent but why a rational, adapting mind stops responding to a channel that has stopped being informative.

What Alert Fatigue Is

Alert fatigue is the reduction in attention and response to safety alerts that follows repeated exposure to them, especially when most of those alerts have proven unimportant. The MeSH descriptor Alert Fatigue, Health Personnel fixes the construct to the clinical setting, where it was named, but the phenomenon is general to any monitored system — aviation, process control, and intrusion detection all describe the same decline. What makes it a cognitive-psychology topic rather than merely an engineering complaint is that the decline is not a property of the alerts; it is a property of the operator's mind adapting to them. The same stream of warnings presented to a fresh observer and to a habituated one produces different behaviour, and the difference is the subject of study.

The adaptation is not a malfunction. A warning system is a communication channel, and a channel that fires mostly false signals carries little information. A mind that reallocates attention away from a low-information channel is doing what an efficient information processor should do; the tragedy of alert fatigue is that the rare true signal rides the same channel as the many false ones and is discarded with them. This is why the problem resists the obvious fix of telling clinicians to pay more attention: the down-weighting is an adaptive response to the statistics of the channel, and it returns as soon as the exhortation fades unless the statistics change. Figure 1 shows the self-reinforcing loop that drives the effect.

Cognitive Mechanisms

Two of the three mechanisms behind alert fatigue are among the oldest documented in experimental psychology. The first is habituation: the decrement in a response that follows repeated or prolonged presentation of the eliciting stimulus, the simplest and most ubiquitous form of learning. The modern consensus description lists its defining properties — the decrement is stimulus-specific, recovers with rest (spontaneous recovery), is faster and deeper with more frequent or weaker stimuli, and is reversed by a novel stimulus (dishabituation) (Rankin et al., 2009). Every one of these properties is visible in alert behaviour: an operator habituates faster to a frequently firing alert, responsiveness recovers after a break, and a change in the alert's form briefly restores attention. Habituation is not forgetting and not fatigue of the muscles; it is a central down-weighting of a stimulus that has proven inconsequential, and it is the single best-fitting label for the core of alert fatigue.

The second mechanism is the vigilance decrement, first demonstrated when radar operators' detection of rare targets was shown to fall measurably within the first half-hour of watch (Mackworth, 1948). The decrement is the decline, over time on task, in the ability to detect infrequent signals, and it is governed by how rare the signal is and how sustained the watch must be — precisely the conditions of monitoring a patient for a rare deterioration across a long shift. Where habituation is driven by the repetition of a specific stimulus, the vigilance decrement is driven by the sustained demand of watching for anything at all; the two compound in the clinical environment, where the operator is both bombarded with specific repeated alerts and required to sustain attention for rare true events over hours. The third mechanism, the signal detection account, explains why the down-weighting is not merely a passive decay but can be the output of a rational decision rule, and it is developed in the next section.

The Signal Detection Account

Signal detection theory separates an observer's sensitivity — the ability to tell signal from noise — from the criterion, the threshold of evidence the observer requires before responding (Green & Swets, 1966). The theory's central insight for alert fatigue is that the optimal criterion depends on two things the operator cannot change by trying harder: the base rate of true events and the costs of the two error types. When true events are rare, a rational observer should set a conservative criterion — requiring strong evidence before acting — because at a low base rate the overwhelming majority of positive signals are false. Shifting the criterion to respond less often is not a lapse; it is the Bayesian-optimal response to a channel whose signals are mostly noise.

This reframes overriding as a decision rather than an inattention. The quantity that matters is the alert's positive predictive value: the probability that an alert is true given that it fired. Positive predictive value collapses when the base rate is low, even for a highly accurate detector, because a small false-positive rate applied to the large pool of non-events still generates more false alerts than the small pool of true events generates true ones. A laboratory study made the behavioural consequence explicit: when the predictive value of a warning was varied, operators' response times to the warning tracked its predictive value directly — a warning known to be usually false was responded to more slowly, and a warning known to be usually true was responded to quickly (Getty et al., 1995). The operator is not ignoring the alert; the operator is calibrating response to the alert's demonstrated reliability, exactly as the theory predicts. The first demonstration below lets the reader set a base rate, sensitivity, and specificity and watch the positive predictive value — and with it the rational response — move.

rational response: the great majority of alerts are false — overriding on sight is rational

Positive predictive value is the fraction of alerts that are true. At a 2.0% base rate with 95% sensitivity and 90% specificity, the system fires 19 true and 98 false alerts per 1,000 cases, so its PPV is just 16.2% — five of every six alerts are false, though the detector is accurate. Raising specificity drains the false-alert pool and lifts PPV far more than any change in sensitivity, because the false alerts come from the large pool of non-events (Green & Swets, 1966).

The Cry-Wolf Effect

The folk name for the behavioural result is the cry-wolf effect, and it has a formal psychology. The analysis of false alarms holds that each unfulfilled warning does not merely waste a response; it updates the operator's estimate of the warning system's credibility, and that lowered credibility transfers to the next warning, including a true one (Breznitz, 1984). The effect is therefore not a static trait of a lazy operator but a dynamic, learned revision: credibility is spent by false alarms and is not fully restored by the occasional true one, because the stream is dominated by the false. This is the mechanism that links the statistics of the channel (Section on signal detection) to the time course of the response (habituation): the cry-wolf effect is what habituation looks like when the stimulus is a warning carrying a claim about the world.

Its practical sting is that the cost of a false alarm is not paid at the time of the false alarm. It is paid later, when a true alarm is slowed or ignored because the system has been discredited. An operator who has learned that a monitor alarms constantly for artefact will be slower to respond when the alarm finally signals a real arrhythmia, and the delay is the accrued debt of every prior false alarm. This is why counting false alarms as a mere nuisance underestimates their harm: each one is a small withdrawal from the credibility that the system will need at the one moment that matters.

Evidence in Health Personnel

The clinical literature supplies the quantitative backbone. In computerized physician order entry systems, the foundational review of drug-safety-alert overriding found that clinicians override the large majority of the alerts they receive — commonly between 49% and 96% across studies — and that a leading reason is the low clinical relevance of most alerts (van der Sijs et al., 2006). The override rate is not evidence that clinicians are reckless; it is evidence that the alerts are mostly not worth acting on, exactly the low-predictive-value situation the signal detection account describes. The policy analysis of the problem argued that decision-support systems should be redesigned to reduce alert fatigue — chiefly by suppressing low-value alerts — while preserving the high-value ones that protect patients and, incidentally, limit liability (Kesselheim et al., 2011).

Later work established the dose-response relationship that marks alert fatigue as a genuine adaptive phenomenon rather than a fixed attitude. Analysing a large volume of alerts in a clinical decision support system, researchers found that the probability a clinician accepted an alert fell as their recent alert burden, workload, and work complexity rose, and fell further with repeated exposure to the same alert — the signature of habituation at scale (Ancker et al., 2017). The same pattern appears in the parallel world of physiologic-monitor alarms. The integrative review of monitor alarm fatigue documented that the enormous majority of monitor alarms in intensive care are false or clinically insignificant, producing a desensitization that the review framed explicitly as a patient-safety hazard (Cvach, 2012); a companion analysis for critical-care nursing reached the same conclusion and connected alarm fatigue to reported adverse events (Sendelbach & Funk, 2013). A systematic review of intensive-care nursing confirmed that alarm fatigue degrades nurses' responses and is associated with both missed alarms and burnout (Lewandowska et al., 2020). A further complication is that some alerts fire wrongly by malfunction rather than by design; an empirically derived taxonomy of decision-support alert malfunctions showed that alerts can break silently and flood or vanish, adding a layer of genuinely erroneous signals on top of the merely low-value ones (Wright et al., 2018).

Remedies

Because alert fatigue is an adaptation to the statistics of the channel, the interventions that work change those statistics rather than exhorting the operator. The most direct lever is specificity: suppressing the low-value alerts that generate most false positives raises the positive predictive value of what remains, and a channel that fires only when action is warranted is one an operator does not learn to ignore. The systematic review of medication-safety alert fatigue concluded that the effective strategies are design strategies — tiering alerts by severity so that only serious ones interrupt, tailoring alerts to the clinician's role so that each sees only what is theirs to act on, and improving the information content and presentation of the alert — rather than interventions aimed at the clinician's effort or awareness (Hussain et al., 2019). Tiering exploits the signal detection logic directly: an interruptive alert implicitly claims a high predictive value, so reserving interruption for high-value alerts keeps the claim honest and preserves the response.

The monitor-alarm literature converges on the same conclusion from the hardware side: widening alarm thresholds to clinically meaningful bands, adding short delays so that transient artefacts self-resolve before alarming, and maintaining electrodes to reduce technical false alarms all cut the false-alarm volume and, with it, the desensitization (Cvach, 2012). The common principle is that the credibility of a warning system is a finite resource that false alarms spend, and every remedy that works does so by spending less of it. Training clinicians to be more vigilant, by contrast, fights the statistics of the channel and loses, because the habituation returns as soon as the next wave of false alarms arrives. The second and third demonstrations below make the two halves of this concrete — how a repeated stimulus habituates and recovers, and how rising alert volume drives the override rate.

at presentation 8 — response strength 12.5 (effectively habituated); rate constant k = 0.26

Habituation is the decline in a response to a repeated stimulus. A more frequent alert habituates both faster and more deeply — raising the frequency steepens the decay curve — which is exactly why a warning that fires constantly stops being noticed. The response recovers with rest and is briefly restored by a novel or changed stimulus (dishabituation), two further defining properties of the process (Rankin et al., 2009).

at 80 alerts/shift — override rate 81.9%, acceptance 18.1%

The probability that a clinician accepts an alert falls as the recent alert burden rises, so the override rate climbs with volume. Even at a light load a substantial share is overridden because many alerts carry little clinical value; as the burden grows, overriding becomes nearly automatic. This dose-response signature — acceptance declining with exposure — is what marks alert fatigue as an adaptive process rather than a fixed disposition (Ancker et al., 2017).

Worked Example

The heart of alert fatigue is that an accurate detector can still produce mostly false alerts, so the arithmetic is worth doing once in full. Consider a drug-interaction alert in an order-entry system. Suppose that of every 1,000 orders, 20 involve a genuinely dangerous interaction — a base rate of 2%. Let the system have a sensitivity of 0.95 (it catches 95% of true dangers) and a specificity of 0.90 (it correctly stays silent on 90% of safe orders). These are good numbers for a detector.

Now count the alerts. The system fires on 0.95 × 20 = 19 of the 20 true dangers (true alerts), and on (1 − 0.90) × 980 = 98 of the 980 safe orders (false alerts). It therefore fires 19 + 98 = 117 times, and its positive predictive value is 19 / 117 = 16.2%. More than five of every six alerts are false — not because the detector is bad, but because true events are rare and even a 10% false-positive rate applied to 980 safe orders swamps the 19 true alerts. An operator who learns that roughly 84% of these alerts are false and begins overriding on sight is responding correctly to the channel's actual predictive value.

Table 1 shows the only lever that fixes this. Raising sensitivity cannot help — it is already near its ceiling and adds at most one true alert. Raising specificity, by contrast, drains the false-alert pool directly.

Moving specificity from 0.90 to 0.99 — leaving sensitivity untouched — lifts the positive predictive value from 16% to 66%, turning a channel that is right one time in six into one that is right two times in three. That single change does more to defeat alert fatigue than any amount of instruction to the operator, because it alters the statistics the operator is adapting to. The arithmetic is the article's argument in miniature: the operator's overriding is rational given the channel, so the fix is the channel.

Discussion

Alert fatigue is best understood as the collision of a well-adapted mind with a badly designed signal environment. Each of its three mechanisms — habituation, the vigilance decrement, and criterion-setting under a low base rate — is an ordinary, generally adaptive feature of cognition, and each is doing its job. The failure is emergent: a system that floods an operator with low-predictive-value alerts recruits these mechanisms to produce exactly the desensitization that then lets a true alert through. Framing the operator as negligent misdiagnoses the problem and prescribes the one remedy — try harder — that the cognitive account predicts will fail.

The evidence is strong and convergent across two independent clinical literatures, decision-support overrides and physiologic-monitor alarms, that the effect is real, large, and tied to alert volume and predictive value. What is harder is the trade-off at the heart of every remedy: suppressing low-value alerts to preserve credibility risks suppressing the occasional true alert that a low-value category still sometimes carries, and the optimal balance depends on costs that differ by setting and are not always known. This is the signal detection trade-off restated at the level of system design, and it has no universal solution — only the discipline of setting the criterion deliberately, with the base rate and the costs in view, rather than letting it drift to wherever a flood of false alarms pushes the operator.

Current Directions

The most active research front is the move from counting alerts to measuring their information value and tailoring them in real time. The systematic review of medication-safety alert fatigue framed the agenda clearly: the field is shifting from blanket alerting toward context-aware, role-tailored, severity-tiered alerts whose interruptiveness is matched to their predictive value, and toward measuring alert burden as a dose that can be titrated (Hussain et al., 2019). The dose-response evidence that override probability falls with recent alert exposure and workload has turned alert burden into a quantity health systems now monitor directly, rather than an inferred nuisance (Ancker et al., 2017).

A second direction is the recognition that some alerts fail by malfunction rather than by design, which demands monitoring of the alerting system's own health. The empirically derived taxonomy of alert malfunctions established that alerts can break in characteristic ways — firing far too often, or silently ceasing to fire — and argued for routine surveillance of alert behaviour as part of clinical-system safety (Wright et al., 2018). Both directions share the article's thesis: because alert fatigue is an adaptation to the signal environment, the leverage is in engineering that environment — its predictive value, its burden, and its reliability — rather than in the operator who is, correctly, adapting to it.

Common Misconceptions

Alert fatigue is just carelessness or laziness.

It is the output of normal cognitive mechanisms — habituation and criterion-setting — adapting to a channel whose signals are mostly false; a conscientious operator shows the same down-weighting because it is the rational response to a low-predictive-value channel (Getty et al., 1995).

A high override rate proves clinicians are ignoring important warnings.

Override rates of 49% to 96% largely reflect that most alerts have low clinical relevance; the override is usually appropriate, and the problem is the alert volume, not the clinician's judgment (van der Sijs et al., 2006).

A more accurate detector will fix the problem.

When true events are rare, even a highly accurate detector produces mostly false alerts, because a small false-positive rate applied to a large pool of non-events swamps the true positives; only raising specificity against the base rate lifts predictive value (Green & Swets, 1966).

Training operators to be more vigilant is the solution.

Vigilance cannot be sustained against a flood of false alarms; the effective interventions tier, suppress, and tailor the alerts themselves, changing the statistics the operator is adapting to (Hussain et al., 2019).

Glossary

Alert fatigue.

The reduction in attention and response to warnings that follows repeated exposure, especially when most warnings have proven unimportant.

Clinical decision support (CDS).

Software that gives clinicians patient-specific guidance, including the drug-interaction and dosing alerts most associated with alert fatigue.

Criterion.

In signal detection theory, the threshold of evidence an observer requires before responding; a conservative criterion yields fewer responses.

Cry-wolf effect.

The erosion of response to a warning system caused by a history of false alarms, which lowers the system's perceived credibility.

Dishabituation.

The temporary restoration of a habituated response produced by a novel or changed stimulus.

Habituation.

The decrement in a response that follows repeated or prolonged presentation of the eliciting stimulus; the simplest form of learning.

Hit rate.

The proportion of true events that a detector correctly signals; equivalent to sensitivity.

Mental fatigue.

The decline in cognitive performance and willingness to respond that follows sustained mental effort; the parent construct of alert fatigue.

Override.

The act of dismissing or bypassing an alert without taking the action it recommends.

Positive predictive value (PPV).

The probability that an alert is true given that it fired; it collapses when the base rate of true events is low.

Sensitivity.

The proportion of true events a detector catches; high sensitivity misses few real dangers but may raise false alarms.

Signal detection theory.

A framework separating an observer's sensitivity from the decision criterion, used to analyse responding under uncertainty.

Specificity.

The proportion of non-events a detector correctly passes without alarming; low specificity is the chief driver of false alerts.

Vigilance decrement.

The decline over time on task in the ability to detect infrequent signals during sustained monitoring.

Key Researchers

Jessica S. Ancker

(Vanderbilt University Medical Center). Biomedical informatics researcher who quantified alert fatigue as a dose-response effect, showing that the probability of accepting an alert falls with recent alert burden, workload, and repeated exposure. Faculty · ORCID

David W. Bates

(Brigham and Women's Hospital / Harvard Medical School). Patient-safety and clinical-informatics researcher whose work on computerized order entry and decision support defined the trade-off between alert sensitivity and the override behaviour that produces alert fatigue. Scholar · ORCID · Wikidata

Shlomo Breznitz

(1936–2026; University of Haifa). Psychologist whose monograph Cry Wolf analysed the psychology of false alarms, showing how repeated unfulfilled warnings erode the credibility of and response to a warning system. Wikipedia · Wikidata

Maria Cvach

(Johns Hopkins Health System). Nurse and patient-safety leader whose integrative review of monitor alarm fatigue framed the physiologic-monitor alarm burden as a distinct safety hazard and informed national alarm-management standards. Faculty

Norman H. Mackworth

(1917–2005; Applied Psychology Unit, Cambridge). Originated the experimental study of sustained attention with the Mackworth Clock, demonstrating the vigilance decrement — the decline in detection of rare signals over time. Wikipedia · Wikidata

John A. Swets

(1928–2016; BBN Technologies). Co-founder of signal detection theory in psychology, whose later work measured how a warning's predictive value governs operator response — the formal basis for understanding why low-value alerts are rationally ignored. Wikipedia · Wikidata

Kai Zheng

(University of California, Irvine). Informatics researcher who showed that medication-safety alert fatigue can be reduced through interaction design and clinical-role tailoring rather than by suppressing alerts outright. Faculty

Frequently Asked Questions

What is alert fatigue?

It is the reduction in attention and response to safety alerts that follows repeated exposure to them, especially when most of those alerts prove false or unimportant; it is most studied in health personnel facing clinical-decision-support and monitor alarms (van der Sijs et al., 2006).

Why is alert fatigue a cognitive-psychology topic rather than just a clinical complaint?

Because its mechanisms are classic cognitive ones: habituation, the decrement in a response to a repeated stimulus (Rankin et al., 2009); the vigilance decrement in sustained monitoring (Mackworth, 1948); and criterion-setting under uncertainty as described by signal detection theory (Green & Swets, 1966).

Why do so many alerts turn out to be false?

Because true events are usually rare, and even an accurate detector applied to a large pool of non-events produces more false positives than true positives, so the positive predictive value of an alert is low unless its specificity is very high (Green & Swets, 1966).

What is the cry-wolf effect?

It is the finding that a history of false alarms lowers the credibility of a warning system, slowing or suppressing the response to subsequent warnings, including true ones; each false alarm spends credibility that the system needs for the rare real event (Breznitz, 1984).

How large is the override problem in practice?

The foundational review found clinicians override the large majority of drug-safety alerts, commonly between 49% and 96% across studies, chiefly because most alerts have low clinical relevance (van der Sijs et al., 2006).

Is overriding an alert the same as a mistake?

No. When most alerts are false, overriding on sight is the response that matches the channel's actual predictive value, and response time tracks predictive value directly, as a controlled study of warnings showed (Getty et al., 1995).

What actually reduces alert fatigue?

Design changes that alter the signal: tiering alerts by severity, tailoring them to the clinician's role, and raising specificity to suppress low-value alerts outperform efforts to make operators more vigilant (Hussain et al., 2019).

Does alert fatigue affect physiologic-monitor alarms too?

Yes. Reviews of monitor alarm fatigue in intensive care find that the great majority of alarms are false or insignificant, producing a desensitization tied to missed alarms, adverse events, and nurse burnout (Cvach, 2012).

References

Ancker, J. S., Edwards, A., Nosal, S., Hauser, D., Mauer, E., & Kaushal, R. (2017). Effects of workload, work complexity, and repeated alerts on alert fatigue in a clinical decision support system. BMC Medical Informatics and Decision Making, 17(1), 36. https://doi.org/10.1186/s12911-017-0430-8

Breznitz, S. (1984). Cry wolf: The psychology of false alarms. Lawrence Erlbaum Associates.

Cvach, M. (2012). Monitor alarm fatigue: An integrative review. Biomedical Instrumentation & Technology, 46(4), 268-277. https://doi.org/10.2345/0899-8205-46.4.268

Getty, D. J., Swets, J. A., Pickett, R. M., & Gonthier, D. (1995). System operator response to warnings of danger: A laboratory investigation of the effects of the predictive value of a warning on human response time. Journal of Experimental Psychology: Applied, 1(1), 19-33. https://doi.org/10.1037/1076-898X.1.1.19

Green, D. M., & Swets, J. A. (1966). Signal detection theory and psychophysics. Wiley.

Hussain, M. I., Reynolds, T. L., & Zheng, K. (2019). Medication safety alert fatigue may be reduced via interaction design and clinical role tailoring: A systematic review. Journal of the American Medical Informatics Association, 26(10), 1141-1149. https://doi.org/10.1093/jamia/ocz095

Kesselheim, A. S., Cresswell, K., Phansalkar, S., Bates, D. W., & Sheikh, A. (2011). Clinical decision support systems could be modified to reduce 'alert fatigue' while still minimizing the risk of litigation. Health Affairs, 30(12), 2310-2317. https://doi.org/10.1377/hlthaff.2010.1111

Lewandowska, K., Weisbrot, M., Cieloszyk, A., Medrzycka-Dabrowska, W., Krupa, S., & Ozga, D. (2020). Impact of alarm fatigue on the work of nurses in an intensive care environment: A systematic review. International Journal of Environmental Research and Public Health, 17(22), 8409. https://doi.org/10.3390/ijerph17228409

Mackworth, N. H. (1948). The breakdown of vigilance during prolonged visual search. Quarterly Journal of Experimental Psychology, 1(1), 6-21. https://doi.org/10.1080/17470214808416738

Rankin, C. H., Abrams, T., Barry, R. J., Bhatnagar, S., Clayton, D. F., Colombo, J., Coppola, G., Geyer, M. A., Glanzman, D. L., Marsland, S., McSweeney, F. K., Wilson, D. A., Wu, C.-F., & Thompson, R. F. (2009). Habituation revisited: An updated and revised description of the behavioral characteristics of habituation. Neurobiology of Learning and Memory, 92(2), 135-138. https://doi.org/10.1016/j.nlm.2008.09.012

Sendelbach, S., & Funk, M. (2013). Alarm fatigue: A patient safety concern. AACN Advanced Critical Care, 24(4), 378-386. https://doi.org/10.1097/NCI.0b013e3182a903f9

van der Sijs, H., Aarts, J., Vulto, A., & Berg, M. (2006). Overriding of drug safety alerts in computerized physician order entry. Journal of the American Medical Informatics Association, 13(2), 138-147. https://doi.org/10.1197/jamia.M1809

Wright, A., Ai, A., Ash, J., Wiesen, J. F., Hickman, T. T., Aaron, S., McEvoy, D., Borkowsky, S., Dissanayake, P. I., Embi, P., Galanter, W., Harper, J., Kassakian, S. Z., Ramoni, R., Schreiber, R., Sirajuddin, A., Bates, D. W., & Sittig, D. F. (2018). Clinical decision support alert malfunctions: Analysis and empirically derived taxonomy. Journal of General Internal Medicine, 33(9), 1526-1535. https://doi.org/10.1007/s11606-018-4415-9