Real‐World Accuracy and Use of a Wearable Fall Detection Device by Older Adults
Shomir Chaudhuri, Daan Oudejans, Hilaire J. Thompson, George Demiris · Journal of the American Geriatrics Society · 2015
A significant danger with falling is the inability to get up afterward, which is reported to occur in as many as 30% of falls.1 Individuals have been shown to have better chances of survival the sooner they are discovered after a fall.2, 3 Thus, several devices have been designed to detect a fall. A pilot study was conducted to investigate the real-world use and accuracy of a wearable fall detection (FD) device with community-dwelling older adults. The device had the ability to detect falls automatically using a combination of accelerometer, magnetometer, and gyroscope. It also had audio feedback and global positioning system capabilities. The company reported testing their device in a laboratory setting with subjects performing prescribed falls, activities of daily living, and stumbles. The system was subsequently tested on an independent dataset and yielded results of sensitivity range from 94.1% to 94.4% and specificity from 92.1% to 94.6%. Of 18 participants, eight completed the 4-month study; of the 10 who partially completed the study, nine voluntarily left the study, and one was unable to complete because of an injurious fall. Participants had the device for an average of 80.7 days (range 8–124); 84 alarms indicating a fall were recorded, of which 83 were false alarms. The largest percentage of false alarms (42.2%) was during normal device use. Another 16.9% of false alarms occurred when the participant dropped the device. Device misuse and putting the device down each constituted 10.8% of false alarms; 19.3% of false alarms occurred for unknown reasons. Table 1 shows the binary classification measurements for the study. Those who completed the study had 58 false alarms over 812 days (7.1%), and those who partially completed had 25 false alarms over 263 days (9.5%) (P = .31). Only one true positive was recorded, when a participant fell backward and hit her head. Participants reported three additional falls while wearing the device, although the device did not identify them as such. In one situation, a participant reported "a light fall into a person's lap." The other two falls occurred from a seated position. Eight falls were reported that occurred while participants were not wearing their devices. Half of these occurred with the device in the charger, in the early morning or at night. Device adherence was a binary measurement; if the participant removed the device from its charger for at least 20 minutes, they were recorded as having used the device on that day. Those who partially completed the study had significantly less adherence (P = .003), although those who completed had a drop in adherence similar to that of partial completers approximately halfway through the trial. To examine the influence of false alarms on adherence, a paired t-test was used to compare adherence 5 days before and after a false alarm (P = .67). Use of the device 5 days before a fall was also compared with use after the fall (P = .63). These findings suggest that the device is inaccurate in real-world settings given the low sensitivity observed. The manufacturer reported testing their device using 59 volunteers. Based on this testing, the company reported sensitivity ranging from 94.1% to 94.4% and specificity ranging from 92.1% to 94.6%. Although the specificities match fairly closely between the laboratory and real-world settings, the difference in sensitivities is stark. Although it is difficult to compare the two studies given the difference in sample size and fall data, such a comparison would appear to match previous evidence suggesting that real-world falls are more difficult to detect accurately.4-6 This finding points to necessary improvements in the accuracy of the device and a need for real-world testing before deployment. The similar decrease in adherence between the two groups near the halfway point of device usage might indicate that participants grew weary of using the device or forgot to use it as the study continued regardless of their willingness to participate in the study. There is a need for more research to better understand what motivates older adults to use these devices so as to encourage greater use.2, 3, 7 Even with limitations of a single device and a short observation period (4 months), this study demonstrates the critical need for real-world testing of fall detection devices by older adults, as well as the need to gather data regarding the actual usage of these devices by their intended audience. Clinicians working with older adults need to assess for the availability (and accuracy) of real-world testing of any fall detection devices before recommending them to patients. This research is supported by Grant T15LM007442 from the National Library of Medicine. Conflict of Interest: The editor in chief has reviewed the conflict of interest checklist provided by the authors and has determined that the authors have no financial or any other kind of personal conflicts with this paper. Author Contributions: Chaudhuri: study design, recruitment, data collection, preparation of manuscript. Oudejans: data analysis and interpretation. Thompson, Demiris: study design, assisting primary author throughout the study, data interpretation, preparation of final manuscript. Sponsor's Role: None.