Comparison of acceleration signals of simulated and real-world backward falls
Introduction
Over the past several decades, falls among older adults have gained increasing attention worldwide, with many studies focusing on fall prevention [1]. Facing the demographic shift and the observed high incidence rates of falls that often leads to injuries, fractures and even death, this common problem has become a very important public healthcare issue. Although much research has advanced the field, the understanding of falls is still limited.
Most of the knowledge on falls to date has been obtained in large-scale epidemiological studies. These have been performed using questionnaire-based fall reports which rely on oral reports by the subjects themselves or by proxies and are biased in many ways [1], [2]. False-reporting or not reporting of falls can be related to the cognitive status of the subjects, the shame of reporting, the fear of consequences, or simply difficulties in defining a fall. Zecevic et al. showed that fallers and caregivers may have different definitions of falling [2]. For example, falling to the ground without an injury might not be interpreted as a fall by every person. There is evidence that 75–80% of all falls without injury are not reported at all [3].
Objectively measured data of falls in a real-life environment are missing and, therefore, many aspects of fall events remain unclear. Body fixed sensor technology, optoelectronic approaches, exo-sensors, and even video material might enhance our understanding of falls and thereby lead to more effective methods for fall prevention and fall detection. However, due to access problems to the target population and other difficulties (e.g. cost, adherence), the number of recorded, documented and published real-world fall data of older people is minimal. Most studies report that approximately one out of three older person falls unintentionally each year. Although this incidence rate is quite high in terms of a disease, in relationship to the feasible observation periods of existing sensor devices the events are relatively rare. The number of measurements needed strongly depend on the particular hypothesis. As an example, to capture 100 real-world falls it would be necessary to record approximately 100,000 days of physical activity (300 person years). Given a sensor observation period of 7 days this means about 15,600 observation intervals each with recharging and data download. In the absence of sufficient real-life fall recordings, researchers focus on data derived from fall simulation studies to develop biomechanical models of falls [4]. Most of the work is done in the field of fall detection, mainly using devices based on accelerometer technology [5], [6], [7], [8], [9], [10], [11], [12], [13], [14], [15], [16], [17], [18], [19]. There is a large consensus that sensor-based fall detection could prove to be very useful for patients. Primarily, it could automatically provide help and shorten rescue time if a faller is unable to send an alarm. Approximately 3% of all fallers lie for more than 20 min without external support [20]. In a cohort of people aged 90 years or older, 80% of the fallers were unable to get up by themselves and 30% remained on the floor for an hour or more [3]. Analysis of the pre-impact sensor signals could help to understand fall mechanisms, such as potential protective movements during the fall. Thus, specific exercise regiments could be developed. Furthermore, information on the impact intensity could lead to new preventive approaches like designing or choosing adequate absorbing floor surfaces. These approaches so far are derived from simulation models [21]. Unfortunately, most of the simulation approaches have failed so far to come up with satisfactory results when they are applied to real-world situations. For example, although fall detection algorithms achieved high sensitivity and high specificity in most simulation studies, the devices had a poor performance in real-life, resulting in high rates of either false positive or false negative alarms and therefore poor user compliance of the patients [22].
To our understanding, the observed gap between experimental and real-life performance of fall detection devices might be due to random and/or systematic differences between simulated and real-world falls for many reasons. For example, the experimental simulation implies consented information to the volunteer. Most experimental designs allow self-initiation of the fall leading to anticipation that may change postural control and response mechanisms, whereas an actual fall generally is unanticipated [23]. Most of the simulation approaches focused on the impact phase of the fall and neglect the phase before the fall. Only a few studies considered other phases of the fall event [16], [24], [25], [26]. The second ethical dilemma is linked to this impact phase. While many (injurious) falls occur on hard materials, such as tiles, ethical constraints require soft landing materials or protective clothing, both absorbing the impact, to avoid injuries to the participants. Thus, acceleration values of a simulated impact phase will never reflect a real-world fall. Basically, real-world falls in contrast to simulated falls are uncontrolled and not planned; recording a real-world fall means recording the intuitive reaction of a person. To perform meaningful simulations, real-world fall events and possible differences between real-world and simulated falls have to be considered first. Subsequently, it might be possible to create an experimental protocol for more realistic fall simulations.
Although the recording of real-world falls is difficult, we were able to collect a number of real-world fall acceleration data in a recent European project (www.sensaction-aal.eu) within a high risk population. Based on these data, we tested two hypotheses in the corresponding experiments 1 and 2: (1) there are systematic differences in acceleration values between real-world and simulated, self-initiated backward falls; (2) a sudden release from a backwards lean during the simulation and the instruction to avoid falling will reduce the differences between real-world and simulated falls.
Section snippets
Subjects and design
Acceleration signals of 5 falls of 4 patients (mean age 68.8 years, SD 4.5, all women) suffering from progressive supranuclear palsy (PSP) were used to describe real-world backward falls. The data were taken from a cross-sectional study to describe clinical aspects of PSP patients [27] and from an intervention study to investigate the feasibility of audio-feedback to improve balance, which was offered to the participants after the cross-sectional study 3 times per week [28]. PSP is an atypical
Results
Fig. 2 shows the acceleration pattern of a real-world fall (a), a simulated fall of experiment 1 (b), and a simulated fall of experiment 2 (c). In experiment 1, when the participants tried to fall like an older person, there was significantly more variation within the acceleration signal during the fall phase in the real-world falls compared to the fall simulation. This observation is demonstrated by the fact that all median values of variance and maximum jerk along all axes were higher in
Discussion
This study, to the best of our knowledge, compared for the first time real-world falls of older persons with low level of functional performance with simulated falls of younger persons, while falls were measured with accelerometers. The real-world falls were well documented falls in the backwards direction; these falls were compared to simulated backwards falls. The acceleration signals of the fall phase (pre-impact) were used to describe the backwards fall. As shown in this study, the fall
Conflict of interest
RC van Lummel is the owner of McRoberts BV, the provider of the DynaPort® MiniMod.
Acknowledgements
Menno Zuidema from McRoberts BV helped to prepare acceleration signals for further analysis. The study was funded by the Robert Bosch Foundation and by the European Commission (FP6 project SENSACTION-AAL, IST-045622). The sponsors had no influence on the design and conduction of the study, or on the writing of the manuscript and the decision to submit the manuscript for publication.
References (37)
- et al.
A proposal for the classification and evaluation of fall detectors
IRBM
(2008) - et al.
Evaluation of a threshold-based tri-axial accelerometer fall detection algorithm
Gait Posture
(2007) - et al.
Sensitivity and specificity of fall detection in people aged 40 years and over
Gait Posture
(2009) - et al.
Comparison of low-complexity fall detection algorithms for body attached accelerometers
Gait Posture
(2008) - et al.
Low stiffness floors can attenuate fall-related femoral impact forces by up to 50% without substantially impairing balance in older women
Accid Anal Prev
(2009) - et al.
The identification of vertical velocity profiles using an interial sensor to investigate pre-impact detections of falls
Med Eng Phys
(2008) - et al.
Effect of audio-biofeedback on balance and gait in PSP-patients—a 6 week pilot study
Parkinsonism Relat Disord
(2010) - et al.
Prevalence of progressive supranuclear palsy and multiple system atrophy: a cross-sectional study
Lancet
(1999) - et al.
Hip impact velocities and body configuration for experimental falls from standing height
J Biomech
(1996) - et al.
Early biomechanical markers of postural instability in Parkinson's disease
Gait Posture
(2009)
PROFANE-Group. Systematic review of definitions and methods of measuring falls in randomised controlled fall prevention trials
Age Ageing
Defining a fall and reasons for falling: comparisons among the views of seniors, health care providers, and the research literature
Gerontologist
Inability to get up after falling, subsequent time on floor, and summoning help: prospective cohort study in people over 90
BMJ
Falls in the elderly: detection and assessment
Conf Proc IEEE Eng Med Biol Soc
The design of a practical and reliable fall detector for community and institutional telecare
J Telemed Telecare
Monitoring behaviour in home using a smart fall sensor and position sensors
Conf Proc IEEE Microtechnol Med Biol
A pilot study of long-term monitoring of human movements in the home using accelerometry
J Telemed Telecare
Speedy: a fall detector in a wrist watch
Cited by (132)
A comprehensive comparison of accuracy and practicality of different types of algorithms for pre-impact fall detection using both young and old adults
2022, Measurement: Journal of the International Measurement ConfederationComparing fall detection methods in people with multiple sclerosis: A prospective observational cohort study
2021, Multiple Sclerosis and Related DisordersCitation Excerpt :A strength of this study is that we measured real world falls under free-living conditions in a unique cohort of people who had high likelihood of falling. Most prior fall detection studies have been done with simulated falls in healthy adults captured in a laboratory setting (Aguiar et al., 2014; Bagala et al., 2012; Kangas et al., 2012; Klenk et al., 2011). In most people, falls are rare and therefore challenging to capture.
A cross-dataset deep learning-based classifier for people fall detection and identification
2020, Computer Methods and Programs in BiomedicineCitation Excerpt :At this point, we would like to remark that the appropriateness of validating wearable fall detection systems with mimicked falls mostly executed by young healthy adults is still open for discussion out of the scope of this study. In this regard, a work by Klenk [32] has found not negligible divergences between the inertial signals collected during simulated and real-life conditions. On the other hand, after analyzing the mobility patterns of actual falls experience by elderly people, Jämsä et al. [33] state that real-life and intentional falls present similar characteristics.
VISION-BASED FALL DETECTION USING CONVOLUTION NEURAL NETWORK AND LONG SHORT-TERM MEMORY
2024, ICIC Express LettersHuman stability assessment and fall detection based on dynamic descriptors
2023, IET Image ProcessingToward Real-Time, Robust Wearable Sensor Fall Detection Using Deep Learning Methods: A Feasibility Study
2023, Applied Sciences (Switzerland)