A robust wearable biosensor for real-time fall detection
Background: Fall is one of the leading causes of morbidity and mortality. It has serious consequences, particularly in senior population, including decline in physical and cognitive abilities, and worsen chronic illnesses, behavioral and socioeconomic issues. Real-time fall detection and automatic notification to a healthcare/homecare provider may facilitate rapid medical response, mitigate the effects of the fall, and reduce the medical costs.
HealthPatch® is a disposable biosensor worn on the chest that detects falls in real-time using posture angle, acceleration of chest and vital signs. The performance of fall detection in a large clinical study conducted in both controlled and free-living conditions is presented.
Methods: Volunteers (N=106) were recruited in two specific groups for a clinical study. 30 healthy young volunteers (age: 19−30 years, BMI: 18.9−32.5, gender F/M: 15/15) performed 33 falls each (11 types, repeat of 3 times) in a laboratory setting wearing 3 patch sensors at the recommended chest locations. The sensitivity and positive predictive value of fall detection was assessed for each of the patch locations. 76 senior volunteers (age: 59−86 years, BMI: 13.5−59.5, gender F/M: 49/27) with range of medical histories wore one HealthPatch sensor for 50 consecutive days following their normal daily practices in free-living settings. They replaced the sensor every 3 days and rotated the patch placement around the 3 recommended chest locations. Senior participants also carried out various activities of daily living (ADL) for about 30 minutes on day-1 and day-4 in a laboratory setting that determined the specificity of fall detection. The number of recorded fall alerts over 50-day period was used to assess the false positive rate of falls per day.
Results: The sensitivity of fall detection was determined to be 95.8%, 94.7% and 95.2% for the 3 patch locations respectively using the real fall data acquired in young participants (N=30) in a laboratory setting. The overall sensitivity and positive predictive value of fall detection were 95.2% (87.6%−98.3%) and 94.7% (86.9%−98.0%) respectively in 2682 captured fall events among 3 patch locations collectively. The specificity of fall detection was found to be 100% using the ADL data acquired in senior participants (N=76) in the controlled setting. On the other hand, the false positive rate of fall detection was found to be 0.0027 falls/day in free-living conditions over 3603 days utilizing data from 1333 patches collectively.
Conclusion: The HealthPatch wearable sensor has been demonstrated to be very effective for real-time fall detection in both the controlled and free-living conditions. In addition to the remote and unobtrusive monitoring of vital signs, the HealthPatch helps to improve the safety of the individual by real-time automated detection and notification of fall events, which can potentially mitigate serious health consequences of fall.