White Paper on Assessment of How Artificial Intelligence Deployed at Long Term Care Facilities Improves Resident’s Safety and Nurse Productivity
Authors: dr. Ran Tao and dr. Harro Stokman
Date: May 19, 2021
The work of care givers is intensive. Of the residential care home nurses who provide direct care, a staggering 37% report feeling burned out . This number has only increased with the recent outbreak of Covid19.
Various technical monitoring systems have been introduced to care facilities to try to help care staff with their work. For instance, all sorts of motion sensors have become commonplace in nursing homes such as magnets to capture door openings, or passive infrared motion sensors. However, these systems are unable to articulate the difference between a resident needing urgent help and a resident just moving his legs a bit in his sleep. As people toss and turn during the night, these types of sensors frequently result in a high number of false alarms. Bed sensors to detect bed exits and chair pressure sensors are also common in care homes to help detect and prevent falls . Again, these sensors are easily triggered by a slight movement, leading again to a significant number of false alarms.
Because of the many false alarms, care workers are frequently distracted from their tasks leading to increased stress levels and a reduction in care quality delivered. In addition, the unnecessary checks by the care givers responding to false alarms disturb the residents, negatively impacting their sleep patterns and consequently their quality of live and even their health conditions. When exposed to an excessive number of alarms, the care workers become tired of the alarms, so-called alarm fatigue [4, 5, 6] which proves dangerous for residents. Desensitization to alarms leads to delayed responses to alarms and even ignorance of true alarms . The Emergency Care Research Institute listed overloads of alarms, alerts and notification as one of the top ten health technology hazards for 2020 [11, 12].
1.2 Combining Artificial Intelligence with Optical Sensors to Help Care Givers and Residents
At Kepler Vision Technologies, we employ the Kepler Night Nurse software instead of relying on motion sensors. This software looks into live video streams of optical sensors such as fisheye cameras using artificial intelligence. The application automatically recognizes if and when a resident needs care, but without creating false alarms. For example, the Kepler Night Nurse recognizes when a resident has fallen or has difficulty leaving the bed. If a resident is not sleeping in the bed at night, or a resident has left the room, the Kepler Night Nurse recognizes that as well. Only in these situations, does Kepler Night Nurse send an alarm in the form of a text message to the nurse on duty.
In this article, we firstly quantify the recognition performance of the Kepler Night Nurse. Secondly, we compare it to other solutions in the market. Finally, we present the benefits that the Kepler Night Nurse brings to both care workers and residents.
2. Clinical Results Quantifying False Alarms and Reliability
To make an objective comparison, we first mention one study of interest , where an accelerometry-based fall detector records a false alarm rate of 1 false alarm per 40 usage hours. In another study , a fall detector that utilizes radar signals generates 2 false alarms per day. A system using Microsoft Kinect reports 1 false alarm per month with a modest reliability (98%, 70% and 71% for standing, sitting, and lying down falls respectively) . However, due to the limitation of its depth sensor, the performance of the system drops greatly when an objects blocks the view of the depth sensor or when the falls take place far from the sensor (greater than 4 meters) . In our customers’ care facilities, the size of the rooms is typically 4 by 5 meters. In addition, our customers demand a reliability for fall detection close to 100%. Both factors challenge the use of depth sensing devices.
To compare these technologies to the Kepler Night Nurse, we report on two studies conducted in the Netherlands, quantifying the false alarm rate and the reliability. In the literature, the term ‘recall’ is much more often encountered than ‘reliability’. However, we use ‘reliability’ because our customers relate this term more easily to their day-to-day practice.
In the first extensive evaluation program with a Dutch care organization, the Kepler Night Nurse was deployed to look after the well-being of elderly residents in 12 different rooms. Each room has a ceiling-mounted 6MP fisheye camera. Here, the evaluation period was three weeks, from Jan 10th 2020 onwards. The focus was on fall detection functionality. For comparison reasons, an alternative motion-sensing system that was used by the care organization back then ran in parallel. The results are listed in Table 1. As the experiments show, the Kepler Night Nurse detected the one real fall incident and generated 9 false alarms. The motion sensing system missed the real fall but instead generated 2195 false alarms. Using Kepler Night Nurse reduced false alarms by more than 99%.
Table 1. Quantitative comparison between the Kepler Night Nurse and the other system used by the customer. The Kepler Night Nurse reduces the number of false alarms up to 99%.
Real fall detected?
Number of false alarms
Kepler Night Nurse
The other system used by the customer
The Kepler Night Nurse is now deployed at multiple care centers belonging to the same care organization. Table 2 summarizes the recognition performance. The reliability observed is close to 100% but, at the moment of writing, we are waiting for our customer to further quantify this.
Table 2. The recognition performance of the Kepler Night Nurse software with ceiling-mounted 6 MP fisheye cameras.
False alarm rate
May 9th 2020 until March 19th 2021
32 ~ 53
1 false alarm per room per 34 days
Close to 100%
In the second evaluation project with another Dutch care organization, the Kepler Night Nurse is used not only to recognize falls, but also responsible for looking after the daily activities of the elderly residents in 14 rooms. Each room has a 4MP dome camera with 2.8mm lens installed at a corner of the room. Here, the evaluation period is nine months, from June 20th, 2020 to March 19th 2021. There is no alternative sensing system in place to compare to. The results on fall detection are listed in Table 3. As is shown, the Kepler Night Nurse generates 1 false alarm of fall detection per room per 96 days. In addition, the feedback provided by the care workers says they always received an alarm for real falls, that is 100% reliability.
Table 3. The recognition performance of the Kepler Night Nurse with 4MP dome cameras.
False alarm rate
June 20th 2020 until March 19th 2021
1 false alarm per room per 96 days
In summary, the Kepler Night Nurse generates far fewer false alarms than existing solutions without compromising on reliability.
With the Kepler Night Nurse, nurses no longer need to enter the resident’s room unnecessarily to check on their physical well-being. During the day, and especially at night, the Kepler Night Nurse provides a considerable reduction in work. Thanks to its low false alarm rate, the Kepler Night Nurse frees the care workers up to focus on providing actual care, also improving job satisfaction.
Residents are no longer disrupted unnecessarily and receive better quality of care as a result. Furthermore, the Kepler Night Nurse also increases the safety of the residents as the Kepler Night Nurse reduces the chance of delayed alarm response and ignorance of true alarms. This will be further quantified in the next section.
3.Clinical Results Quantifying the Impact on Patient’s Safety
Getting up from the floor after a fall is a challenge for the elderly. According to the study , even when not injured, 47% cannot get up without assistance. Remaining on the floor after a fall for a prolonged period increases morbidity and mortality rates. An earlier study  found that half of those who remained on the floor after a fall for an hour or longer pass away within 6 months. Another study  showed that 60% of those who were on the floor after a fall for more than an hour were admitted to hospital in the following year. The study in  shows that in 5 out of 6 fall cases the help did not arrive within 5 minutes.
To quantify the effects of the Kepler Night Nurse, we measure, once the Kepler Night Nurse solution is in place, how long it takes for nurses to arrive at the scene and help residents to get up. The detailed results are shown in Table 4 and are compared to the literature in Table 5.
Table 4: How long does it take for a nurse to arrive in case of a fall? In 2 out of 14 cases, it takes more than 5 minutes. Nurses sometimes have other priorities to attend.
When was alarm sent?
When did nurse arrive?
Within 5 minutes?
4 minutes and 45 seconds
3 minutes and 12 seconds
Table 5 shows that in 5 out of 6 cases, using the Kepler Night Nurse assistance arrives within 5 minutes after a fall, compared to 1 in 6 cases where customers do not deploy the Kepler Night Nurse. The Kepler Night Nurse therefore significantly reduces the chances of being on the floor for a prolonged period of time and hence increases the safety of the residents.
Table 5. In case of a fall, the Kepler Night Nurse always sends an alarm within 60 seconds. This greatly improves the chances of staff arriving in time, as the table shows. Hence, the software increases the safety of the residents.
Assistance arrives within 5 minutes after fall
With Kepler Night Nurse
5 out of 6 falls
Without Kepler Night Nurse
1 out of 6 falls 
4. Clinical Results Quantifying the Impact on Saving Care Workers’ Time
Besides functioning as an alarm system, Kepler Night Nurse also generates reports of daily activities (see Figure 1). Activity reports replace the shift reports that care workers write for their colleagues to provide them with a patient status update. Automating reporting saves care workers’ time and makes the reports objective and more detailed.
After consulting with nurses in the field, we have calculated that the automatic reporting from the Kepler Night Nurse saves three minutes per resident per shift. As a typical nurse tends 20 residents in a care home, and as there are three shifts per day, this saves a total of three hours of work time every day per 20 residents.
In this study it was clinically shown that the Kepler Night Nurse acts as an accurate alarm system with a very low number of false alarms without sacrificing reliability. In summary:
- The false alarm rate for 6MP fisheye camera is 1 false alarm rate per 34 days.
- The false alarm rate for 4MP dome camera is 1 false alarm rate per 96 days.
In both cases, the reliability is almost 100%.
Without the software in place, in 1 out of 6 cases assistance to a fall arrives within five minutes. With the Kepler Night Nurse, a message is sent to the nurses within 60 second. This greatly reduces the time spent on the floor. The report shows that assistance arrives within 5 minutes in 5 out of 6 cases. Finally, the software saves a total of three hours of work time every day per 20 residents.
As such, it brings several benefits to the care workers and residents:
- Care workers using the Kepler Night Nurse no longer need to enter a resident’s room unnecessarily to check on them. During the day, but especially at night, the Kepler Night Nurse provides a considerable reduction in work. Thanks to its low false alarm rate, the Kepler Night Nurse frees the care workers up to focus on providing actual care.
- Residents are no longer disrupted unnecessarily and receive better quality of care. It increases their safety because it reduces the chance of delayed alarm response and ignorance of true alarms.
Take a look at the Kepler Vision Technologies homepage!
[1.] Kangas, Maarit, et al. “Sensitivity and false alarm rate of a fall sensor in long-term fall detection in the elderly.” Gerontology 61.1 (2015): 61-68.
[2.] Su, Bo Yu, et al. “Doppler radar fall activity detection using the wavelet transform.” IEEE Transactions on Biomedical Engineering 62.3 (2014): 865-875.
[3.] Stone, Erik E., and Marjorie Skubic. “Fall detection in homes of older adults using the Microsoft Kinect.” IEEE journal of biomedical and health informatics 19.1 (2014): 290-301.
[5.] Cvach, Maria. “Monitor alarm fatigue: an integrative review.” Biomedical instrumentation & technology 46.4 (2012): 268-277.
[6.] Nguyen, James, et al. “Combating Alarm Fatigue: The Quest for More Accurate and Safer Clinical Monitoring Equipment.” Vignettes in Patient Safety-Volume 4. IntechOpen, 2019.
[7.] Fleming, Jane, and Carol Brayne. “Inability to get up after falling, subsequent time on floor, and summoning help: prospective cohort study in people over 90.” Bmj 337 (2008).
[8.] Tinetti, Mary E., Wen-Liang Liu, and Elizabeth B. Claus. “Predictors and prognosis of inability to get up after falls among elderly persons.” Jama 269.1 (1993): 65-70.
[9.] Wild, Deidre, U. S. Nayak, and B. Isaacs. “How dangerous are falls in old people at home?.” Br Med J (Clin Res Ed) 282.6260 (1981): 266-268.
[10.] McHugh, Matthew D., et al. “Nurses’ widespread job dissatisfaction, burnout, and frustration with health benefits signal problems for patient care.” Health Affairs 30.2 (2011): 202-210.