When every alarm is noise, the real one gets missed

Most monitoring problems people worry about are visible: a flat battery, a sensor that stops reporting, a gap in the record. Alarm fatigue is harder to see, because everything looks as if it's working. Readings arrive, alarms fire, someone acknowledges them. The trouble is that acknowledging has become a reflex.

This article covers fixed monitoring: fridges, freezers, cold rooms and stores in hospital laboratories and pharmacy, NHS and independent, pharmaceutical distribution and the food cold chain. It explains what alarm fatigue is, why it matters so much, what usually causes it, what good looks like, and the questions to ask any monitoring system, including the one you have now.

What alarm fatigue is

Alarm fatigue is a learned response. When most alarms turn out to be a door opened for a delivery, a defrost cycle or a sensor that dropped off the network for ten minutes, people learn that alarms usually don't need action. They acknowledge without looking, mute the sounder, or stop checking their phone at night.

Nobody decides to ignore alarms. It happens gradually, one nuisance alarm at a time, and it happens to careful, well-trained teams. The cause is almost always in how the system is set up.

Why it is the biggest risk

A monitoring system has one job: to make sure that when something goes wrong, a person knows in time to act. Alarm fatigue breaks that job while leaving every other part looking healthy. It shows up in four ways.

• Real alarms are missed or dismissed. A compressor failure at 2am looks, on a phone screen, much like the door alarm that went off forty times that week.
• Staff make the noise stop. Under pressure, people widen alarm limits, lengthen delays with no data behind them, move a probe away from where the mapping says it should be, or switch an alarm off. Each change is made to reduce noise and each one creates a blind spot.
• The record stops meaning anything. Hundreds of incidents closed with "door opened" or "no action needed" bury the ones that needed a root cause, and they make it harder to show an inspector or auditor that every excursion was assessed.
• Inspections find it. The MHRA's published GMP deficiency examples include alarm limits set wider than the storage range. An MHRA Inspectorate blog on transport risk shows a vehicle trace with alarm limits set so wide that the "lack of alarm triggering led to incorrect assurance that stock was within acceptable temperature range". Neither finding says why the limits were set that way, but widening limits to cut the noise produces exactly that blind spot.

There is no figure for how many real excursions are missed this way, because a missed alarm leaves no trace.

What the rules expect

No UK regulator sets an alarm delay in minutes. They do all expect alarms, or the monitoring behind them, to be set deliberately, answered by a person and justified in your own procedures.

• Medicines storage and distribution. EU GDP 3.3, which Regulation 43 of the Human Medicines Regulations 2012 makes a licence condition for wholesale dealers (in Great Britain, as the guidelines stood at the end of 2020), asks for alarm systems that alert you to excursions, with alarm levels set appropriately and alarms tested regularly. The European Commission's GDP questions and answers add that alarm settings should guarantee a timely alert, without giving values. See MHRA GDP temperature control.
• Hospital pharmacy and clinical areas. NHS SPS guidance on alerts suggests setting the alarm just inside the limit (7.5 °C for an 8 °C maximum) and allows a delay for small, acceptable excursions, kept "as short a period as practically possible" so a fridge cycling in and out of range is still caught. Alarms should not be permanently silenced, and a silenced alarm should sound again if the fridge is still out of range. For vaccine fridges, see the Green Book: the daily read, record and reset still stands alongside any alarm.
• Vaccine fridges in primary care. CQC's GP mythbuster 17 asks practices to consider whether an alarm could turn itself off once the fridge is back in range, without anyone investigating.
• Medical laboratories. ISO 15189:2022 expects the environmental conditions that affect results, temperature among them, to be specified, monitored and recorded (6.3.1), storage to keep samples and reagents intact (6.3.3), and reagents to be stored to the manufacturer's specification with conditions monitored (6.6.2). It doesn't prescribe temperature alarms. What UKAS assessors do look at is how you allow for measurement uncertainty: UKAS's guidance on decision rules, LAB 48, uses a product's maximum storage temperature as one of its own examples, and with a guard band treats a reading as passing only when it sits below the limit by at least the expanded uncertainty.
• Blood. The Red Book (JPAC, 6.7), written for the UK blood services and commonly applied by hospital transfusion laboratories, expects a log of alarm events describing the corrective action taken.
• Placing probes. WHO Technical Supplement 6 (TRS 961 Annex 9, section 2.5.2) says monitors should not be placed where door openings "generate an abnormally high number of alarms", and that if the problem can't be fixed, the area shouldn't be used for temperature-sensitive stock at all. Our page on door probes and nuisance alarms sets out where the guidance differs.
• Food. BRCGS Food Safety and Storage and Distribution expect suitable alarms on temperature-controlled storage. UK food law is silent on alarm settings, so the delays belong in your HACCP plan, with a reason.

So every set point and every delay is yours to justify, and an inspector or assessor can ask you to.

Where the noise comes from

In our experience, the same causes turn up again and again, in every sector.

Alarms with no delay

A fridge, freezer or cold room warms briefly every time the door opens and every time it defrosts. With no delay, every one of those is an alarm. One 12,000 m³ freezer we monitor at a food distribution depot has its door opened 1,200 to 1,600 times a day. A door alarm with no delay on that freezer would never stop.

One alarm per sensor, not one incident per problem

A walk-in cold room or a store may have several sensors. When the room warms, every sensor goes out of range, and a system that treats each sensor separately raises an alarm for each one. One real-world problem becomes many alarms, many messages and many records, each asking for its own root cause.

Readings that flap around the limit

A reading that sits right on a limit drifts above it, dips below, then rises again. If the alarm clears each time the reading dips back, one long incident turns into a string of short ones, each opened, notified, acknowledged and closed. The team sees noise. The record shows a dozen minor incidents instead of one real one, and how long the problem really lasted is lost.

Offline, door and power alarms treated like temperature alarms

A sensor that misses one reading has usually not failed. A door open for two minutes during a delivery is normal. A power blip of a few seconds is not a power cut. When these alarms fire with no delay, they often outnumber temperature alarms, and they are the first ones people learn to ignore.

Alarms sent to everyone

When every alarm goes to every person, nobody owns it. Each person assumes someone else is dealing with it, and out of hours the people who can act are buried in messages they can't.

Probes in the wrong place

A probe at the door threshold measures the door, not the stock. A probe in moving air alarms on every opening when the product inside hasn't changed temperature at all.

Limits that don't match the product

Limits set tighter than the product needs alarm on harmless changes. Limits set wider than the product's range stay quiet when it matters. Both usually come from a default that nobody revisited.

What good looks like

Good alarm management doesn't mean fewer alarms for the sake of it. It means every alarm is worth answering.

• One real-world problem is one incident. A problem with a unit or a room should reach people as one incident, with one owner and one root cause, however many sensors saw it. Each sensor's breach is an event within that incident, not a new one.
• An incident lasts as long as the problem. A reading that crosses a limit and dips back, again and again, belongs to one incident, and the record should show it that way.
• Every alarm has a delay, and every delay has a reason. Set each delay from your own mapping and qualification data, especially how long the unit takes to recover after a normal door opening or restock. As a starting point we use: a door alarm only when the door has been open longer than normal access, typically 5 to 10 minutes; a sensor-offline alarm only after several missed readings, typically 30 to 60 minutes, shorter for blood and other high-risk storage and never longer than the unit's holdover allows; a power alarm delayed past brief blips, typically about 15 minutes and always shorter than the unit's holdover time; an air-probe alarm on a medicines or vaccine fridge of no more than 15 minutes, and 10 minutes on a blood fridge. The one alarm we would never delay is the blood product or core alarm. The full recommendation is on our monitoring and alarms page.
• A delay changes when someone is told, not what is recorded. The data still shows any excursion, and under GDP and GMP every excursion is still a deviation to assess.
• Alarm limits sit inside the storage limit by at least the sensor's measurement uncertainty. A reading exactly at the limit has about an even chance of being over it, which is why UKAS's LAB 48 uses a guard band. Take the expanded uncertainty from the sensor's calibration certificate and set the alarm inside the limit by at least that much, so it fires before the true temperature can breach. A UKAS calibration certificate states the uncertainty, so the figure should already be in your file. Never set an alarm outside the product's range.
• Each alarm type has its own rules. Temperature, door, offline and power alarms need different delays and often different people.
• Each alarm goes to someone who can act on it, with cover out of hours and an escalation if they don't respond.
• Alarm counts are reviewed. A unit that alarms repeatedly is telling you something: a door seal, a loading habit, a probe position or a delay that needs data behind it. Fix the cause before you lengthen the delay.

The questions to ask any monitoring system

Whether you are choosing a system or reviewing the one you have, these questions show quickly how it will behave on a busy day.

1. When 20 sensors in one cold store go out of range, do you get one incident or 20?
2. What happens when a reading flaps around the limit? Does the incident close and reopen each time, and what does the record show afterwards?
3. Can you set a delay for each alarm type and each unit, and record the mapping or qualification data that justifies it?
4. Do door, offline and power alarms each have their own delay, separate from temperature?
5. Can an alarm clear itself without anyone looking at it?
6. Who receives each alarm in working hours and out of hours, and what happens if nobody acknowledges it?
7. Can you show how many alarms each unit raised last month, and how many were real excursions?
8. Who can change a limit or a delay, and where is that change recorded?

If the answers are vague, ask to see it happen on a live system.

Where to start

Start with last month's alarms. Count them by unit, mark which were real excursions and which were doors, defrosts or dropouts, and look for the units that alarm again and again. Most sites find a handful of units cause most of the noise.

Our guide, Cutting alarm fatigue without missing a real excursion, takes you through that audit step by step: how to classify alarms, find flapping and duplicates, check delays against your mapping data, review routing and escalation, and write delay justifications you can explain at inspection. It's free to download below. If you'd rather talk it through, book a 20-minute check: a specialist goes through your alarm set-up with you and you get a short list of what to change first.

85% fewer temperature incidents for a home healthcare provider, after it changed sensor positions, thresholds, alerting and root cause work together. Read the case study.