Fall Detection Sensor vs Personal Alarm in a Care Home

Why the unwitnessed fall decides the comparison

The comparison between a sensor and an alarm is not about technology preference. It is about the fall nobody sees. In a one year prospective study of 110 people aged over 90 living in their own homes or in care homes, 82% of falls (217 of 265 reports) happened when the person was alone, and 54% of fall reports described the person as being found on the floor. Of those who fell at least once during the year, 80% were unable to get up after at least one fall, and 30% had lain on the floor for an hour or more.

The same study is blunt about the limit of a button. Call alarms were widely available to those participants, but the authors report that they were not used in most of the falls that led to lying on the floor for a long time, and they list the reasons the older people gave: a sense that the alarm was not relevant to the situation, concern about independence, and practical difficulty in using it. That is the honest framing for everything that follows. The question is not whether a person can call for help. It is what happens when they cannot.

A pendant with a manual call button is a phone, not a detector

The Hong Kong service usually called the 平安鐘 is a personal emergency response service. The operator, the Senior Citizen Home Safety Association, describes itself as a social enterprise and non profit charity that has served Hong Kong since 1996, and its Care-on-Call service connects a subscriber to a 24 hour response centre over a mobile network. The wireless personal emergency link comes with a remote SOS button, and the service list covers emergency assistance, outpatient appointment booking, medicine taking reminders and care calls; location search is listed on the newer mobile products.

It is a summoning device rather than a detector, and that distinction survives every improvement in the hardware. The fall is identified by the person, not by the device. The operator's own account of one pilot beneficiary, in its release of March 2026, makes the point without argument: she fell at about 5am on the way to the bathroom, was unconscious for around two hours, could not make herself heard by her husband in the bedroom, and eventually climbed onto a chair to call the police. The detector was installed after that fall.

The running cost is knowable in advance, which is the strength of this approach. The operator's comparison page lists a standard monthly fee of HK$178 for the wireless personal emergency link on a 12 month contract, with subsidised plans available (page read 14 October 2026).

Wearable automatic fall detection reads the movement

A wearable detector uses motion sensors, commonly an accelerometer with a gyroscope and a magnetometer, to classify a movement pattern as a fall. No press is required, which is exactly the gap the pendant leaves open. The published evidence on how well that works in practice is less flattering than the marketing, and worth reading in full.

In a four month study of 18 community dwelling older adults wearing one commercial detector, participants recorded 84 alarms indicating a fall, of which 83 were false. There was a single true positive, a fall backwards with a head impact. Three further falls were reported by participants while they were wearing the device and the device did not identify them. The manufacturer's laboratory figures were a sensitivity of 94.1% to 94.4% and a specificity of 92.1% to 94.6% on an independent data set; the study authors concluded that the device was inaccurate in real world settings given the very low sensitivity they observed, and recommended real world testing before deployment.

Two operational details from the same study matter to a care home more than the headline numbers. First, eight falls were reported while the device was not being worn, and half of those happened while it was sitting in the charger, often in the early morning or at night, which is the period when a fall is least likely to be witnessed. Second, adherence fell in both groups about halfway through the trial, and the study found no significant change in adherence in the five days after a false alarm (p=0.67), so the failure was gradual disuse rather than a direct reaction to noise.

Ambient radar lets the room read the posture

Non contact sensing moves the detector off the person and into the room. Hong Kong's largest operator now runs this in a pilot scheme launched in March 2026: a wall mounted detector using non thermal imaging sensing covers a living room, bedroom or bathroom, the release states that the detection process does not involve photography or recording, and on sensing a fall it connects automatically to the 24 hour service centre. Staff speak to the user through the device's two way communication, and if the user needs rescue or the staff cannot confirm the user's safety, they follow established procedures including calling an ambulance. The same system flags a user who has not left a designated area for an extended period, described in the release as more than 24 hours. The pilot provided 24 months of free service to about 300 high risk households funded by the Labour and Welfare Bureau and the HKEX Foundation, with about 200 more households covered by charitable funds, roughly 500 in total.

Accuracy figures for this approach have to be read with their test conditions attached. A 2026 evaluation of millimetre wave fall detection, published in Scientific Reports on 25 February 2026, reported an overall accuracy of 97.9% in a 12 by 12 metre indoor test area, with 10 healthy adult participants across 10 multi person scenarios, validated against video ground truth. The same paper reports that the false negative rate rose as the number of people in the space increased, because of occlusion, and that earlier published results sit in the 92.2% to 99.5% range but come mostly from much smaller areas with at most three people present.

The privacy property is the reason this category is being put in bathrooms and bedrooms. Millimetre wave radar does not capture an identifiable image, which is why the operator can describe the fall detector as suitable for those rooms. Our own design notes on privacy preserving vision in aged care cover the alternative of processing posture on the device rather than sending images away.

Camera vision works where a camera may lawfully point

A camera based system evaluates posture in the zones it covers and raises an alert when the posture matches what a fall looks like. Technically it belongs in the same family as radar: the detector watches the room, not the person's wrist. Its coverage limit, however, is legal as much as it is technical.

The Privacy Commissioner for Personal Data sets the boundary in its statement of 17 December 2025 on a CCTV case and its guidance: Data Protection Principle 1(1) of the Personal Data (Privacy) Ordinance requires personal data to be collected only for a lawful purpose directly related to a function or activity of the data user, and only to the extent that it is necessary and not excessive, while Principle 1(2) requires the means of collection to be lawful and fair. The Commissioner states that organisations should avoid using CCTV to collect personal data under unfair circumstances, and that cameras should not be installed to collect images in places where individuals would have a reasonable expectation of privacy, giving changing rooms and restrooms as examples.

Read that against the first section above and the tension is plain. The rooms where an older person is most likely to be alone and on the floor are also the rooms where a camera is least likely to be acceptable. A camera deployment therefore needs a stated purpose, a sign that tells people what is being recorded, a retention rule for the footage, and a decision about who may view it. None of those are optional extras once images of an identified person are being collected.

How the four approaches compare, attribute by attribute

The table sets the four approaches side by side on the attributes that decide a deployment. Every figure carries the scope and date of the source it came from.

Read down the second column and the real divide appears: two of the four approaches require the person to act, and two do not. Read down the fourth column and the second divide appears: the approaches that need no action are also the ones that produce events when nothing has happened, which is a staffing question rather than a technology question.

What a false alarm costs in staff time and attention

False alarms are usually discussed as an accuracy statistic, and they are better understood as an operational cost. A systematic review of alarm fatigue in intensive care, covering 69 publications, reports that of all patient monitoring alarms, 72% to 99% are either technically false or clinically irrelevant. That range comes from vital sign monitoring in intensive care, so the rate does not transfer to a care home floor; what transfers is the mechanism the review describes, in which an overwhelming volume of alarms produces sensory overload, desensitisation, and staff who delay, dismiss or re-set alarms.

A scoping review of alarm fatigue in healthcare, covering 32 publications, reaches the same conclusion from a different direction: repeated exposure to frequent or non actionable alarms produces desensitisation and reduced responsiveness, and the consequences include delayed responses and staff switching alarms off. In a care home, the equivalent of switching an alarm off is a sensor that the night shift stops trusting.

The practical test is therefore not the accuracy figure on a specification sheet. It is the number of false alarms per resident per month in this building, with this furniture and this resident mix, and the staffing arithmetic that follows. A pilot can produce that number before a purchase decision does.

Combining the two, and what to write into the plan

The button and the sensor answer different questions, so the common configuration is both. The automatic layer covers the unwitnessed fall, which is the event the pendant cannot cover. The button covers everything else: pain, needing the toilet, a dizzy spell, a request that is not a fall at all. Neither replaces the other, and a home that installs sensors and leaves the call system to decay has swapped one gap for another.

The written plan should state, for every covered zone, four things: what triggers an alert, who receives it, the expected time to respond, and what happens when the alert is false. Then state the gaps. A bathroom without a sensor, a corridor without a camera and a bed area outside coverage are the zones where the risk concentrates, and writing them down is what stops a purchase from being mistaken for coverage. Our notes on choosing a fall detection system and on wearable fall detection go through the evaluation questions in more detail, and the building level document that should record the coverage gaps is described in care home risk assessment.

One limit belongs in the plan in plain words. Detection does not prevent a fall. It changes how long the person lies on the floor, and therefore what happens next. Where the environment is what causes the fall, the fix is the environment, which is the subject of night time falls in care homes.