At 10:14 a.m., a physician opens an order.
At 10:14:02, an alert appears.
At 10:14:05, another one appears.
At 10:14:09, a third one.
At that point, the problem is no longer clinical. It is neurological.
When a system keeps interrupting all the time, the brain stops distinguishing between an important alert and an irrelevant one. And that is where alert fatigue begins.
AHRQ’s PSNet puts it plainly: in a Veterans Affairs primary care study, clinicians were receiving more than 100 alerts a day, and alert fatigue occurs precisely when professionals become desensitized to the volume of warnings and end up ignoring both the trivial and the critical ones (psnet.ahrq.gov)
Executive translation: you do not have a “very careful” system. You have a system that is training your doctors not to listen.
The crisis is real: more alerts do not mean more safety
For years, many hospitals operated under an intuitive but dangerous logic:
“If we add more alerts, we prevent more errors.”
The evidence says otherwise.
PSNet summarizes that clinicians often override the vast majority of CPOE/CDS warnings, including critical ones, and cites literature showing that physicians may override alerts up to 95% of the time (psnet.ahrq.gov).
And the most recent studies continue pointing in the same direction. An analysis published in 2025 on alert burden at an academic medical center found 196,225 alerts in 2023 and an overall override rate of 93.5%; for drug allergy alerts, the override rate reached 98.9% (pmc.ncbi.nlm.nih.gov).
Clinical translation: when almost everything is overridden, the alert stops being a safety net and becomes operational noise.
The problem is not the existence of alerts; it is their lack of context
Here is the key point.
Not all alerts are bad. What is bad is triggering them without enough clinical context.
AHRQ, in its project on “meaningful” drug interaction alerts, explains that many commercial systems use simple rules based only on the presence of medication pairs. The result: excessive alerts, low relevance, and therefore alert fatigue. Their proposal goes in the exact opposite direction: contextualizing alerts with patient-
specific data so that they interrupt only when there is truly relevant clinical risk (digital.ahrq.gov).
Translation for IT and Medical Leadership: the future is not about having more rules. It is about having better signals.
When the system screams too much, the physician stops hearing even the critical things
This is the dangerous part.
PSNet warns that the proliferation of alerts, which in theory should improve safety, can end up paradoxically increasing the likelihood of harm, because the professional learns to ignore both the trivial and the truly serious (psnet.ahrq.gov).
And that is not an abstract idea. A systematic review of overrides found average override ranges between 46.2% and 96.2%, with particular concern in certain categories where many overrides proved inappropriate (pubmed.ncbi.nlm.nih.gov).
In addition, another study found that, out of 1,087 alerts evaluated for appropriateness, 67.89% of overrides were inappropriate, and that in a sample of those inappropriate overrides, associated medication errors were identified (pubmed.ncbi.nlm.nih.gov).
The uncomfortable message: a system that interrupts for everything may make you feel “covered,” but in reality it is eroding the physician’s ability to respond when it truly matters.
Alert fatigue = mental fatigue
This issue does not only affect safety. It also affects clinician mental health.
The burden of messages and notifications inside the EHR is part of everyday cognitive strain. A study on electronic inbox work patterns showed that, on workdays, primary care physicians had an average of 100 message views during work hours and 53 after hours, for an average total of 153 views per day; they
also spent nearly 52 minutes a day on the inbox (pmc.ncbi.nlm.nih.gov).
Although “views” does not exactly equal “unique notifications,” it does portray something important: the modern professional already lives immersed in a saturated digital environment. If, on top of that, clinical alerts are not useful, the system stops helping and starts wearing people down.
The thesis: you need “smart alerts,” not “infinite alerts”
The solution is not to turn everything off. Nor is it to leave only one final red alert and pray.
The solution is to change the logic:
- from simple rules to clinical context,
- from constant interruptions to relevant interventions,
- from noise to signal.
The alert safety literature itself points in that direction.
A systematic review concluded that alert fatigue can best be reduced by improving interaction design
and adapting alerts to the clinical role and workflow (pmc.ncbi.nlm.nih.gov).
And AHRQ, once again, emphasizes that individualizing alerts according to patient circumstances can reduce the overall burden and increase the clinical value of the warnings that do appear (digital.ahrq.gov).
CLARA is not designed to “shout louder,” but to understand better
CLARA is presented as an AI assistant integrated into the HIS that processes the complete medical record, identifies key information such as allergies, medical history, and recent results, and offers intelligent clinical recommendations, including alerts about interactions and risks, always as support and never as a
substitute for medical judgment (harmonimd.com)
In addition, HarmoniMD describes CLARA as a tool that analyzes, summarizes, predicts, and alerts in real time within the same HIS/Harmoni-Go interface, reducing search and documentation time (harmonimd.com)
The strategic difference is not minor:
- an old system launches alerts because it detects an isolated condition,
- an intelligent system should evaluate whether that condition matters for that patient, in that context, at that moment.
Put simply: you do not need a system that talks to you all the time. You need one that interrupts you only when it is worth interrupting you.
How to tell whether your hospital has “noise” or “signal”
If you are a Medical Director or IT Director, these questions will help you diagnose the problem quickly:
1. What percentage of alerts are being overridden in your operation today? If it is extremely high, you no longer have a support system; you have a noise system (pmc.ncbi.nlm.nih.gov).
2. Do your alerts use patient context or just general rules? AHRQ shows that contextualizing alerts is precisely the path to making them meaningful (digital.ahrq.gov).
3. Do all alerts interrupt in the same way? The evidence suggests that not all of them should be interruptive or treated with the same level of severity (psnet.ahrq.gov).
4. Does the system reduce or increase the clinician’s mental burden? If the real answer is “it distracts them,” then the cost is no longer only operational: it is human.
5. Does your HIS help prioritize what matters or simply document the chaos? That is the tipping point between old software and clinically useful software.
Give your physicians their peace of mind back by replacing noise with signal
Alert fatigue is not a user whim. It is a predictable consequence of systems that interrupt too much and understand too little.
The evidence is already clear:
- clinicians may receive more than 100 alerts a day in certain environments (psnet.ahrq.gov),
- override rates may exceed 90% (pmc.ncbi.nlm.nih.gov) (psnet.ahrq.gov),
- and when the system loses credibility, patient safety loses strength as well (psnet.ahrq.gov).
That is why the goal is not to “add more alerts.”
The goal is to make every alert matter.
Because a good system does not shout louder at the physician.
It speaks only when it truly matters.
If you want to see how HarmoniMD + CLARA can help you move from massive alerts to smart, contextualized alerts, schedule a demo. We can review with you where the noise is today, which signals are actually worth preserving, and how to give your staff peace of mind back without compromising patient safety
(harmonimd.com).
