Increasing fall risk awareness using wearables: A fall risk awareness protocol

Accepted Manuscript Increasing Fall Risk Awareness Using Wearables A Fall Risk Awareness Protocol Asbjørn Danielsen, Hans Olofsen, Bernt Arild Bremdal PII: DOI: Reference:

S1532-0464(16)30098-3 http://dx.doi.org/10.1016/j.jbi.2016.08.016 YJBIN 2624

To appear in:

Journal of Biomedical Informatics

Received Date: Revised Date: Accepted Date:

25 February 2016 12 August 2016 14 August 2016

Please cite this article as: Danielsen, A., Olofsen, H., Bremdal, B.A., Increasing Fall Risk Awareness Using Wearables A Fall Risk Awareness Protocol, Journal of Biomedical Informatics (2016), doi: http://dx.doi.org/ 10.1016/j.jbi.2016.08.016

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Increasing Fall Risk Awareness Using Wearables

A Fall Risk Awareness Protocol

Asbjørn Danielsen, Hans Olofsen, Bernt Arild Bremdal, UiT – The Arctic University of Norway, Norway

[email protected], [email protected], [email protected] Corresponding author: [email protected]

Abstract. Each year about a third of elderly aged 65 or older experience a fall. Many of these falls may have been avoided if fall risk assessment and prevention tools where available in a daily living situation. We identify what kind of information is relevant for doing fall risk assessment and prevention using wearable sensors in a daily living environment by investigating current research, distinguishing between prospective and context-aware fall risk assessment and prevention. Based on our findings, we propose a fall risk awareness protocol as a fall prevention tool integrating both wearables and ambient sensing technology into a single platform. Keywords: daily living, fall risk assessment, fall prevention, fall risk awareness protocol, fall prevention tool, wearable sensors, sensor fusion

1

Introduction Each year about one third of elderly aged 65 or older experience a fall [61, 88], and each year older adults are

hospitalized for fall-related injuries up to five times more often than other causes [33]. Between 10% and 20% of falls by elderly cause serious injuries such as fractures or head traumas while non-fatal fall injuries are associated with considerable morbidity including decreased functioning and loss of independence [11]. The falls also present a significant cost in healthcare [64]. Many of these falls may be avoided if fall risk assessment and prevention tools where available as an integral part of daily living. Fall risk assessment is a process in which the probability of a future fall is estimated, usually within a timeframe of 6-12 months. Fall risk prevention, on the other hand, addresses the important question; how should one prevent falls from happening in the first place. This question has been investigated in a number of studies by addressing intrinsic factors like medications and general health status [67, 68], extrinsic factors like hazards found in the living environment [68], and evaluation of balance and mobility using functional tests [9, 28, 63]. In a study by Oliver & Healy [24] they reported on nurses who recognized whether a patient suffer from a prominent risk of falling simply

by watching them move about and reviewing their medical history. Based on these assessments changes in living environment of the elderly are applied, training programs initiated, and medications adjusted, all to reduce the risk of falling. These approaches are all prospective of nature, thus we label these approaches prospective fall risk assessments and prospective fall prevention. In a clinical setting, fall risk assessments are undertaken using functional tests and questionnaires. Functional tests, like Time-Up and Go (TUG) [63], the Berg Balance Scale [9], Performance-Oriented Mobility Assessment (POMA) [28] etc. are objective and used in clinical settings in terms of prospective fall risk analysis. They do not however take into account the responsiveness and discriminative ability of a relatively healthy population. Neither of these approaches is sufficient in terms of accuracy of fall risk assessment [12, 18]. Further, the clinical fall risk assessment tools do not take into account hazards found in the normal daily living environment of the elderly such as carpets, pets, doorsteps etc., which may explain why the clinical tests fail to distinguish fallers from non-fallers with a satisfying recognition percentage [13]. Finally, the Hawthorne effect, the tendency of people to perform better than usual when they know they are being observed [46], may influence data collected in a clinical setting and further invalidate the findings. In a systematic review by Oliver et al. [29], the predictive validity of STRATIFY [79] for identifying hospital patients who will fall was evaluated. They concluded that even the best tools could not identify high-risk individuals for fall prevention with a sufficient certainty. Gates et al. [12] reviewed 29 different functional screening tests for prediction of fall risk, and concluded that most tools discriminate poorly between fallers and non-fallers, and that the evidence is not sufficient in terms of screening instrument for predicting falls. These findings are supported by Callis [76] which document 20 significant fall risk factors which are not used in different fall risk assessment models like STRATIFY [79], HRMII [77], and MFS [78]. While the prospective approaches focus on long-term fall risk assessment and prevention, the number of approaches taking the current situation into account is very limited, though pre-impact fall detection systems to reduce fall-related injuries exist [69, 70, 71]. Falls happen at specific times and in specific contexts. In order to prevent falls from happening, the current context such as time of day, current health status, location, activity, and other relevant information of the current context need to be included in the fall risk assessment as well. Consequently, fall risk assessment has a context-aware property influenced by the current situation, which has not been adequately addressed by existing fall risk assessments. A similar separation is legitimate concerning fall prevention. Prospective fall risk assessments have a number of shortcomings [12, 13, 18, 29], but many of the problems may be solved by close monitoring and counseling by health personnel. Such counseling is both expensive and timeconsuming, and considering challenges most countries face in term of the increase in elderly population, other solutions should be explored. Collecting data related to activities in the daily living environment of the elderly makes it possible to evaluate the performance of activities, how the performance evolves over time, and how gait

characteristics and other health status related properties evolve and correlate with the activities. This approach makes it possible to do prospective fall risk assessment as an integrated and non-intrusive part of daily living using data from wearables and ambient sensing technologies. Since data is continuously collected and determination of context is possible, the approach makes it possible to do context-aware fall risk analysis and prevention as well. We label this context-aware fall risk assessment, and context-aware fall prevention. We investigate fall risk assessment and fall prevention classifying them as either prospective (long-term) or context-aware (current context and immediate). We specifically investigate wearable approaches identifying information relevant to perform fall risk assessment and fall prevention in a daily living environment using wearables. Preventing falls from happening is in many respects the ultimate goal, both individually and in a social economic context. Fall prevention tool using wearable technology are however difficult to find. Based on this observation we propose an approach on fall prevention and how it may be implemented using a fall risk awareness protocol. The rest of the paper is organized as follows. Section 2 presents the literature search done as part of the work presented here, and other related work. In section 3 we present fall risk assessment strategies using wearables and organize the approaches into prospective and context-aware. We further present a survey on wearable sensor types and sensor placement used in fall risk studies. In section 4 fall risk prevention is presented as prospective and context-aware approach strategies. Section 5 presents the fall risk awareness protocol and its implementation along with a discussion of possible effects of the protocol. Section 6 concludes the paper and presents future work.

2

Related Work Leading up to this paper, a systematic literature search was performed. The literature search was performed using

the key-words “wearable”, “fall risk” and “ambient assisted living” combined with publishing year from 2008. The search was conducted on CiteSeerX, Google Scholar, IEEE Xplore Digital Library, ACM Digital Library and ScienceDirect. The purpose was investigating state-of-the-art approaches on wearable sensory solutions focusing on fall risk in an Ambient Assisted Living (AAL) context. The results were filtered during three stages and used as a starting point for the development of the proposed fall risk awareness protocol. Figure 1 gives an overview of the literature search process.

Figure 1. Literature search process

The number of reviews and surveys focusing on wearable sensors for fall risk assessment are limited. Impracticalities like size and wearability of such sensors along with limited battery lifetime, limits the use of such devices in larger long-term studies. Howcroft et al. [13] reviewed 40 studies using inertial sensors focusing on geriatric fall risk. The study differentiated on sensor placement, which variables where assessed in the studies, and which models of risk assessment were applied. In the study the use of inertial sensors was found promising in terms of fall risk assessment. Additionally, they concluded that “future studies should identify fallers using prospective techniques and focus on determining the most promising sensor sites, in conjunction with determination of optimally predictive variables”. Shany et al. [27] reviewed challenges associated with the use of wearable sensors for quantification of fall risk in older people. They focused primarily on how wearable sensors may be used in a clinical situation as an objective supplement to fall risk assessment. They further recognized the value of unsupervised and unstructured sensory measurements as important due to the presumption that knowing the fall risk of as many people as possible will enable healthcare systems to promote awareness, allocate resources by streamlining referrals, address problems sooner by providing timely and suitable intervention, and ultimately prevent falls from happening in the first place.

They concluded that such measurements might become good indicators of fall risk if larger studies and high-quality trials are performed. The number of reviews or surveys that investigate wearable technology in terms of context-aware fall prevention is very limited. Most studies focus on fall risk assessment rather than fall prevention. Delahoz et al. [80] investigate fall detection and fall prevention using wearables, but the main focus of the review is still on fall detection. Out of 24 systems reviewed, 2 systems using wearable technology for fall prevention where found. Hamm et al. [81] study fall prevention intervention strategies, separating them into pre-fall, post-fall, fall injury, and cross fall prevention intervention systems. The study distinguish on what type of information is collected and how this information may contribute in different settings. The number of actual wearable solutions aimed at preventing falls by themselves by issuing some kind of feedback to the individual wearing the system, is still very limited. Data collection using wearables in the daily living environment offers a unique insight into how different activities are performed and how these activities relate to fall risk. Such data collections have to be unsupervised and uncontrolled, i.e. the persons using the wearables are not instructed on which activities or actions to perform and how to perform them. The number of studies investigating this approach is however very limited. As far as we have been able to determine, only two studies since 2012 have focused on fall risk assessment using wearable sensors, recording daily life activities from elderly (age > 65) and with reasonably size (participants > 50). In [20] van Schooten et al. investigated how quality of daily gait characteristics influenced fall risk assessment using a one-week data recording, and concluded that daily life accelerometry contributed substantially to identification of individuals having an increased fall risk. They also found that they were able to predict a future fall within 6 months with good accuracy. Weiss et al. [17] did a similar study based on three-days recording rather than one-week, concluding that accelerometer-derived analysis is useful in the context of predicting falls.

3

Fall Risk Assessment The potential causes why older adults fall are multidimensional and include predictive factors associated with the

natural aging process, various disease processes, polypharmacy along with elements related to the current situation such as the individuals’ current health status, context and activity being performed. The prospective elements are valuable in terms of a long-term fall risk assessments, but do not take the actual situation at hand into context.

3.1

Prospective Fall Risk Assessment

Research shows that someone that has fallen is likely to fall again due to the ‘post fall syndrome’ [26]. The ‘post fall syndrome’ refer to elderly people who has fallen and developed severe anxiety which affected their ability to stand and walk unsupported [92]. Later research has demonstrated that elderly people can develop a similar fear of falling even when they have not fallen [93, 94]. In this context, the personal fall history and recognition of prior fall indicators is most important since they significantly influence the probability of a future fall. A number of studies

have focused on this, and Table 1 summarizes the most important findings and gait parameters distinguishing fallers from non-fallers.

Gait Parameter

Finding

Gait Speed

Non-fallers have significantly higher gait speed than fallers [13, 17, 20, 22, 66].

Step Duration

Step duration is significantly longer in fallers than in non-fallers [17].

Gait Variability

Mediolateral variability is significantly lower with fallers while vertical variability is significantly higher for fallers as opposed to non-fallers [3, 13, 17, 20, 23].

Activity Level

Heesch et al. [31] showed that at least daily moderate to vigorous-intensity physical activity is required for the primary prevention of falls to the ground. [20, 66] found that fallers have significantly lower total locomotion than non-fallers

Multi Scale Entropy

Riva et al. [1] show that fallers have significantly higher

(MSE)

MSE (as an indicator of complexity in gait kinematics) than non-fallers.

Local Dynamic Stability

Local Dynamic Stability is significantly lower for fallers

(Gait Stability)

than non-fallers. [3, 14, 66]

Harmonic Ratio (HR)

Harmonic Ratio of the upper and lower trunk was consistently lower in fallers than non-fallers [1, 13, 15].

Table 1. Gait parameters separating fallers from non-fallers

The prospective fall risk is further dependent on a number of other parameters. [13] and [20] found short stride length to be a significant indicator for prospective falls. Further on, stride time variability has been recognized as a significant factor with respect to differentiate prospective fallers from non-fallers; the larger variation, the more likely it is that a fall will happen within a year [25, 13]. Near falls are more frequent than falls and are closely related to falls [42]. Near falls or missteps are initiated by an initial or sudden loss of balance, but do not escalate to a fall due to the individual’s ability to regain balance. Continuous unstructured and unsupervised monitoring using wearables may help to correctly recognize near-fall events because older adults tend to not recognize the events themselves or recognize the significance of the near-fall events.

Activities of Daily Living (ADL) and Instrumental Activities of Daily Living (IADL) are standardized measures of biological and psychosocial functions that are used to describe how dependent and well-functioning an elderly or chronically ill are. While ADL describes basal activities like eating, toileting, getting in/out of bed, etc. IADL describes more complex activities like preparing meals, go shopping, handling money, etc. These measures are often expressed as 5 limitation stages, from 0 to IV, where 0 express that the individual have no problems in execution of activity, I = mild problems, II = moderate problems, III = severe problems, and IV = all activities are difficult. Brown et al. [16] did a statistical study that concluded that instrumental activity of daily living (IADL) limitation stages can be used as a fall assessment tool: elderly with IADL limitation II or III are significantly more likely to fall compared to those with limitation 0. Weiss et al. [4] found that individuals without mobility disability but with IADL disability had difficulties with turns, had lower yaw amplitude during turns, were slower and had less consistent gait. Even though the findings in [4] are not significant, the findings could be used as indicators in fall risk assessment due to the positive correlation with IADL limitation stages. In addition to the parameters that may be recorded using wearable sensors, we know that the probability of falls increase by age [37]. It is further evident that elderly with cognitive impairment has a higher probability for falls [75], living alone increase the risk of falling [36, 38], as do obesity [47]. We also know that general health status and polypharmacy significantly increases the risk of falling [19, 21, 32]. Finally, research show that women fall more than men and are more seriously injured, and further that gender actually differentiate the importance of a number of other risk factors [73].

3.2

Context-aware Fall Risk Assessment

Recognizing the context in terms of where an activity is executed, the activity itself, as well as the individuals’ current health status holds great value. It is evident that factors like sleep quality and excessive daytime sleeping [6], dizziness [7], vertigo, and balance disorders [8] influence fall risk. Such risks may be identified using questionnaires in a prospective fall risk assessment, but may also be recognized using wearable sensors in the context aware analysis. We have not been able to find any studies focusing on these factors using wearable sensors. The long-term fall risk assessment is what most approaches have been investigating, though some work has been done in the area of context aware fall risk assessment, e.g. [34] and [35]. In [34] Koshmak et al. present an approach in which data collected from wearable sensors are combined with data collected in a smart home environment using a Bayesian network. The smart home environment provided contextual information, obtained from environmental sensors, and contributed to assessment of fall risk probability. The evaluation of the developed system was performed through simulation. de Backere et al. [35] reported on a multi-sensor platform in which wearable sensors was one of many types that may be included in the system. They performed a proof of concept by simulating an apartment in their own lab using more or less off-the-shelf sensors, both for ambient sensing and inertial sensors.

It is further evident that some activities lead to a higher risk of falling than others. Robinovitch et al. [10] found that more than 60% of all falls by elderly were related to incorrect weight shifting, confirming findings in an earlier study by Rubenstein [48] which discussed causes why elderly people fall and which individual risk factors that dominated falls. Hence all activities involving weight changes, especially major weight changes like rising up from chair, sitting down, turning, etc. should be considered as inducing a higher risk of falling in a context-aware fall risk assessment. Near falls are related to fall risks, are more frequent than falls and may precede falls. If the ability to regain balance declines, the risk of an imminent fall rises [42], and vice-versa. Consequently, recognizing near falls are valuable both in prospective and context-aware fall risk assessments. We have not been able to find any studies registering near-falls in an elderly population, though several studies [71, 49, 72] describe algorithms and wearable sensors that may be used to register near-falls. Biofeedback monitoring may further increase the accuracy as shown by Horta et al. [30]. Tinetti et al. [37] found changes in gait characteristics as a significant indicator of imminent falls. Recognizing such changes is thus important, and experiments like [5] show how data from a low-cost insole using 12 force-sensitive resistors, calculating ground reaction force and ankle moment, has a very high correlation with clinical data. Such sensory systems are considered to be an accurate method to identify and analyze various step phases [2], making it possible to recognize characteristics that influence fall risk. The context of a person may be associated with activities like sitting, sleeping, walking, climbing stairs, etc., and recognized using Artificial Intelligence (AI) algorithms as exemplified by Ernes et al. [62]. Recognizing more complex activities, like cleaning the house, preparing meal, eating, doing the dishes, etc. require sensory information that involve ambient sensing. Context-aware fall risk assessment need not only consider the current situation. It also need to evaluate how performance of activities evolve in a long-term perspective, identifying trends and providing support for prospective fall risk assessment as well.

3.3

Sensor placement and sensor properties

Inertial sensors constitute the most applicable sensor type to collect relevant data associated with gait characteristics. Accelerometers and gyroscopes are used in this context in a number of studies, but placement differs. Table 2 gives an overview of placement and analysis focus using inertial sensors.

Location Lower back

Gait study focus [1]: Harmonic Ratio (HR), index of harmonicity (IH), Multiscale Entropy (MSE),

Recurrence quantification analysis [4]: Balance Shifting in turns, Yaw Amplitude during turns, Gait Speed, and Gait Consistency [15]: Harmonic Ratio [17]: Gait Speed, Step Duration, Gait Variability [20]: Gait Speed, Activity Level [23]: Gait Variability [43]: Kinetic Energy, Gait Speed, Step Duration, Gait Variability

Upper back

[3]: Local Dynamic Stability, Gait Variability [15]: Harmonic Ratio

Insole

[5]: Ground Reaction Force, Ankle Moment, Balance [25]: Gait Variability [39]: Abnormal Gait Pattern

Smartphone

[39]: Abnormal Gait Pattern

Hip

[14]: Local Dynamic Stability Table 2. Inertial sensor location and focus

In most studies, the sensor is located on the lower back. This is due to the general acceptance by the individual wearing the sensor along with a notion that the closer the sensor is to the center of mass; the more relevant data is collected. In addition to inertial sensors, a number of other sensors are being used in both prospective fall risk assessment and in fall risk prevention contexts. Table 3 give an overview on other wearable sensors and how they may be applied in a fall risk assessment and prevention context.

Location

Sensor

Sensing focus

Study focus

Chest

ECG

Heart Rate Variability

[83] Identifying future fallers prospectively [83] Fall prediction from sitting to standing position

Chest

ECG

ST interval length

[83] Fall recognition

R peak amplitude

Lower Leg

EMG

Co-Contraction Indices

[84] Pre-impact fall detection

Integrated EMG

Head

EEG

Brainwave

[82] Fall prevention, recognizing onset of Freezing of Gait (FOG) in Parkinson’s patients

Table 3. Non-inertial sensor locations and focus

In addition to single-sensory approaches towards fall risk assessments, combining several of these sensors may increase accuracy. In [30] Horta et al. demonstrated that ElectroCardioGram (ECG), ElectroMyoGraphy (EMG), Galvanic Skin Response (GSR) or Electro Dermal Activity (EDA), respiration along with other signals could be used to identify patterns that increased the risk of an imminent fall. Annese et al. [85] used a combination of Electroencephalogram (EEG) recognizing Movement Related Potentials, and EMG recognizing Co-Contraction Indices to recognize abnormal gait.

4

Fall Prevention Preventing falls from happening include fall risk assessment, calculating the risk, and acting correspondingly to

the risk. Like fall risk assessment, the prevention strategy is dependent on the prospective risk and the risk associated with the current context.

4.1

Prospective fall prevention

In the prospective fall risk assessment, environmental hazards are evaluated along with health history and clinical tests to determine a long-term fall risk assessment. In this context wearables can be used as an assistive tool to gather objective information related to different tests that provides health personnel a better foundation for fall risk assessment. In terms of actions for avoiding falls from happening, this is normally supervised by qualified health personnel and includes exercise programs focusing on balance and strength, interventions to improve home safety, etc. [40]. Self-administrated clinical tests may to some extent help to update prospective fall risk assessments, and contribute to prospective fall prevention [41].

4.2

Context-aware fall prevention

Context-aware fall prevention using wearables should be based on both prospective and context-aware fall risk assessments as a continuous, context-aware process with the main purpose to provide feedback about the imminent risk of falling. The feedback should be directed to everyone with formal or informal responsibilities with respect to care of the person being monitored, dependent on the context. Caporusso et al. [44] presented a pervasive fall prevention solution, incorporating both prospective and contextaware fall risk assessment. They used an individually adjusted threshold-based approach towards risk analysis, and non-invasive multimodal feedback. Sensors and feedback were implemented in a wristband that included a tri-axial accelerometer, an RGB LED, a transducer and bi-directional communication capabilities. Visual feedback was given to the end-user in form of RGB LED “traffic-light” and tactile feedback in the form of vibration. One main purpose of the wristband was to increase the individuals’ awareness of their risk of falling, thereby preventing situations that may lead to falls. The experiment in [44] was set up as a 10-hour session using both a wrist-worn and a waist-worn

accelerometer. The subjects were told to behave as normal. 20 persons attended the experiment, ranging from 25 to 88 years. The recordings were later analyzed and 7 distinct activities were recognized; washing, eating, sleeping, resting, sitting, walking and dressing. The average recognition rate of an activity was 62,06%. It was further evident that the persons attending the experiment recognized and responded to more alarms from the wrist-worn device (85,83%) than from the waist-worn (29,08%). The experiment showed a very high acceptance to using a wristband (95%), but did not conclude whether awareness of fall risk decreased the actual number of falls due to the short timeframe of the experiment. Previous studies do however show correlation between awareness of falls and the actual number of falls [45].

5

Fall Risk Awareness Collecting data related to activities in a daily living environment makes it possible to recognize activities, evaluate

the execution of activities, how execution evolves over time, and how gait characteristics, balance, and other health status related properties evolve and correlate. This approach makes it possible to do context-aware fall risk assessment while providing valuable information for objective long-term, prospective fall risk assessments. It further provides a basis to implement a context-aware fall risk awareness protocol. In a larger six-month study performed in 2009 Dykes et al. [45] found a positive correlation between awareness of fall risk and the actual number of falls, both in hospital settings and intervention settings. The awareness of fall risk is not only an individual process, but moreover a process involving everyone with formal or informal responsibilities with respect to care of the person being monitored. Educating the elderly about the fall risk itself have been shown to be effective in terms of reducing the number of falls [59], and we know that educating both relatives [60] and health personnel [45] on fall risk increases awareness and reduce the number of falls. Some approaches have been done to attempt to raise fall risk awareness based on current situation, though not taking the complete context into consideration. Majumder et al. [39] recognized abnormal gait using a smartphones accelerometer and gyroscope combined with a sensor shoe insole. They used abnormal gait as a fall risk indicator and used an alert message as feedback. They reported an accuracy of abnormal gait recognition of 97,2% using a decision tree algorithm. Horta et al. [30] demonstrated that ElectroCardioGram (ECG), ElectroMyoGraphy (EMG), Galvanic Skin Response (GSR) or Electro Dermal Activity (EDA), respiration along with other signals could be used to identify patterns that increased the risk of an imminent fall. They did 50 recordings of each activity (walk, run, sit, jump, fall), and used a threshold-based approach to determine fall risk using data from accelerometer, EMG and GSR, and ECG. Activity recognition rate varied from 79% to 93%, with an average recognition rate of 86,8%. In [35] de Backere et al. presented an approach towards context-aware fall prevention using ambient sensors. It was a social and context-aware platform for fall risk assessment with feedback using an ontology-based approach with multiple ambient sensors to determine context and risk. The approach included an alarm central, and distinguished between formal and informal caregivers. Feedback was implemented using smartphones, tablets, email

and wristband. It should be noted that the work presented by de Backere et al. [35] is a proof-of-concept technological platform for doing context-aware fall detection and risk assessment. As far as we have been able to determine, no actual installations or experiments using this technology have been reported. We have identified a number of properties that significantly influence fall risk, and separated fall risk assessment and fall prevention into prospective and context aware. The positive correlation between awareness of fall risk by health personnel, relatives, and the elderly, and the actual number of falls seem to be very clear as well. We propose a new fall risk awareness protocol based on context aware fall risk analysis.

5.1

Fall Risk Awareness Protocol

The need to coordinate feedback with respect to fall risk awareness to all actors is a primary requirement. Still, feedback has to be easy to understand in order to be accepted by the elderly being monitored. In addition, the feedback need to be timed correctly to be constructive, and it need to be designed in such a way that the elderly respond to the feedback by decreasing the imminent fall risk. We therefore propose a simplistic and configurable feedback system to the elderly as the basis for our fall awareness protocol. The feedback part of the system used by the elderly is integrated into a single device including an accelerometer, a gyroscope, an alarm button, a vibration sensor for tactile feedback, an RGB LED emitter for visual feedback, a sound unit for audible feedback, and a two-way low energy communication unit. It can visually signal the elderly using a combination of blinking (the higher frequency the higher fall risk) and color traffic-light system (morphing from green to yellow to red dependent on risk assessment). Messaging may be reinforced using vibrations, and pre-recorded messages or alarm sounds may be activated using the audible feedback. Each level of the feedback functionality is configurable.

5.1.1

Turning sensors data into a Fall Risk Probability Score

Figure 2. Calculating Fall Risk using the Fall Risk Probability Engine

Figure 2 shows three distinctly colored regions indicating where processing takes place. The Indoor processing is done locally in the individual’s home environment. Processing in the Outdoor region is done on a mobile platform using smartphones. The Wearables deliver data either to the smartphone in an outdoor situation, or to the indoor processing unit in case the wearables recognize such a device. The Fall Risk Probability Engine (FRPE) is responsible for analyzing data and generating a fall risk probability score. The FRPE is a Complex Event Processing (CEP) engine which use a distributed approach towards processing and recognition of events, focusing on capturing complex event patterns while distributing the load. Cugola et al. [91] give a good overview of the concept behind a CEP engine and present this approach along with other Information Flow Processing designs. Combining ambient sensor data with sensor data from wearables is important to be able to recognize context and activity. Figure 2 gives an overview on how this may be achieved independently of sensor platform and infrastructure. We distinguish between Wearables, Indoor ambient sensing and Outdoor ambient sensing as Sensory Data Origins (SDO). Each SDO produces snapshots of data at some predefined frequency. The snapshot data is fed into artificial intelligence (AI) algorithms that recognize Activity and Location in an Indoor or Outdoor setting, and

calculate and recognize relevant information from the Wearables SDO. Wearables may include inertial sensors that may be used for recognizing activities found in wrist mounted devices [51] and smartphones [50], or insole pressure measurements like WIISEL [52] which are used to recognize gait characteristics. The result from Indoor and Outdoor recognition is an identification of an activity like walking, sitting, standing, etc. and a location. Further on, indoor ambient sensing may add information to the context by simple indoor localization solutions based on Passive Infrared (PIR) sensors registering movement [53], or more advanced solutions [54, 55, 56]. Lowe et al. [57] give a good overview on different technologies used for monitoring behavior in a home environment. The quality of sensory measurements and the algorithms used to identify activities and events are crucially important when calculating the Fall Risk Probability Score (FRPS). The sensors, both wearables and ambient, produce a continuous series of snapshots. These snapshots will in the Indoor and Outdoor situations be continuously and independently processed to recognize Activity and Location, while the wearables will be processed into Activity, Gait Characteristics [37], Vital Signs like pulse or ECG [83,85] and EMG [30, 84, 85], etc. The independent and distributed processing makes it possibly to maintain an acceptable performance of the system by focusing the processing to the current context of the elderly. E.g. if an elderly is in the kitchen, sensory snapshots from ambient sensors in other locations do not need to be processed since the data are irrelevant in terms of current fall risk. It is further important that sensory observations are identified within a uniform notion of time to maintain quality of the observations by reducing the number of false negatives and false positives. E.g. two different ambient sensor register presence of a person in two different rooms at identical points in time, person A in room X, and person B in room Y. One individual, A, is recognized as walking (probability of 90%) and the other one, B, is lying down on the floor (probability of 80%). Simultaneously, the activity recognition based on wearables indicates with a 90% probability that the elderly is walking. The system will, based on these observations, during Fall Risk Analysis, deduce that the person walking is the elderly due to the strong correlation between the observations done in room X and the observations from the wristband. This correlation adds accuracy and sensitivity to recognition by positively identifying the actual situation or event using different SDOs. Data from the SDOs recognize Activity, but may also recognize events like “sitting down”, “raising up”, during the Wearable Data Merge. These events will be subjected to similar processing to verify correlation between the wearables and the ambient sensors in the Fall Risk Analysis processes. Recognizing a fall or a near-fall [49, 71] have special requirements. A fall in our context is an involuntary event resulting in a person coming to rest on the floor or other lower level. Such a fall may be a sudden incident, but may also be an event that take quite some time. The fall might have the effect that the individual in question is unable to raise up again in which case the system should signal for assistance. A near-fall is a different situation in which the individual loose balance, but regains it and thereby avoids the fall. The need to identify these situations are important, but the need to do so immediately is not important in terms of FRPS calculation, though they add information to the Fall Risk Analysis.

The Fall Risk Analysis compare and process information to find correlations to recognize situations and events which influence the current fall risk. The events recognized and an indication of the current calculated fall risk is then stored in the Indoor or Outdoor Trend History for further processing, while the data is passed to the Fall Risk Calculation processes to calculate FRPS. The FRPS is an indicator on how the fall risk is expected to be developing based on the current context. To calculate the FRPS we not only use the current fall risk as found in the Fall Risk Analysis, but include trend developments using newly added data and evaluate these with similar historical trends and similar patterns to create it. Information on gender, age, weight, diagnosis, polypharmacy, etc. are also included. The FRPS is in short two numbers. The first being the current fall risk calculation in the form of a number between 0 and 100, the other number is a number between -10 and 10 which indicate whether the risk of falling is going down, is unchanged or increasing.

5.1.2

The Feedback

The FRPS is continuously updated, and will create a trend showing current development of the context-aware fall risk. It is the FRPS that is used for feedback purposes. The feedback is used to inform the elderly, and others, on how the current activity in the current context influence fall risk. The actual design of the feedback, i.e. which kind of feedback to give to the elderly will be dependent on the situation at hand and how the elderly prefer feedback. It will further need to be timed so that the feedback is delivered timely and become constructive. Feedback is also dependent on cognitive abilities of the elderly, time of day, localization, etc. The contextual data in Figure 3 refers to data collected using inertial sensors, ambient sensors determining location or activity, and other data relevant to perform context aware fall risk assessment. In addition to indoor contextual data, similar techniques, including GPS signaling, may be used in an outdoor situation to add context.

Figure 3. Fall Risk Awareness Protocol Overview

Relatives and other informal caregivers need information related to fall risk as well. Similar, but more demanding requirements exist for health personnel. We propose to develop a service which is continuously updated with current Fall Risk Probability Score (FRPS), and which is able to communicate to other systems using standard protocols including messaging by email, SMS, and application specific protocols for immediate messaging and alarms. The service should present fall risk assessment developments with respect to actual contexts as well as indicate expected future trends based on the FRPS. The level and type of feedback is dependent on actor, risk-level, and whether the fall risk is decreasing or increasing. Figure 3 gives an overview of the proposed fall risk awareness protocol and illustrates how contextual data may be integrated into a fall risk assessment service making it possible to increase fall risk awareness. Contextual data is continuously transferred from the elderly and into the Fall Risk Probability Engine as part of the Fall Risk Assessment Service. The data is analyzed to detect behavior and calculate imminent fall risk, sending information to the wristband of the elderly indicating current expected development of the fall risk if current execution of activity continues. How the wristband reacts to the messaging will be dependent on how feedback configuration is set up. The data is being made available to Health Personnel and Relatives having appropriate consent from the elderly, including both current fall risk assessment of the elderly in question as well as historical

data trends showing fall risk development and expected future trend. If fall risk increases above a prefixed threshold, an alarm and email is issued to Health Personnel, along with a log entry informing Health Personnel on the current fall risk and requesting action to be taken by Health Personnel. Health Personnel should then do a Fall Risk Awareness Follow-Up by contacting the elderly, by phone or otherwise. The threshold need to be configured and tuned per installation/individual due to individually differences in physical and cognitive abilities, differences in general health status, etc. If the Fall Risk Assessment Service determines a high probability of an imminent fall, an alarm is sent to the Alarm Central, to Relatives, and to Health Personnel. The Alarm Central confirms the alarm situation by sending Alarm Notifications to Health Personnel and Relatives, requesting a Follow-Up by both.

5.2

Discussion

Unstructured and unsupervised collections of contextual data as an integrated and non-intrusive part of daily living offer a unique opportunity to collect and analyze execution of activities. The data collected will be sent to the Fall Risk Assessment Service for local and distributed processing and calculation of the immediate fall risk. The results in terms of the FRPS will be transferred back to health personnel and relatives as well as the elderly. As part of prospective fall risk assessment, exercise programs focusing on balance and strength may be introduced to the elderly. Such programs have a significant impact on reduction of falls and fall risk [65], and the proposed model for monitoring makes it possible to evaluate how balance and strength is developed during the exercises and over time. All data gathered is objective and can be used by health personnel as a supplemental foundation for fall risk assessment, avoiding problems like the Hawthorne effect [46]. Using contextual data implies that the elderly to some extent expose his or her privacy to the system. This is potentially challenging. In this paper, we have not focused on privacy, ethical and social aspects of our proposed protocol. Both ethical and privacy aspects should be addressed when implementing a protocol like the one we propose. The privacy of the elderly should be taken care of, but they should be prone to making concessions since the protocol has their benefit as its primary goal. The implementation of the ethical role of the relatives in the protocol might be the biggest challenge. By giving the relatives the possibility to follow up, they are given a responsibility that involves both handling personal information of the elder as well as actively performing the followup. We do not find a discussion of this to be within the scope of this paper. However, the protocol is flexible enough to incorporate special requirements such as the physical location of specific parts of the fall risk assessment service and the exact roles of the relatives and others. The distributed approach towards Complex Event Processing used in the FRPE support such adaptations. A multimodal approach towards activity recognition increase the appreciation of context, and thereby increase the accuracy of the fall risk assessment. Further, it is possible to recognize basic Activities of Daily Living (ADL)

along with a number of IADLs using AI algorithms, e.g. Fleury et al. [74] recognize ADLs using a Support-Vector Machine-based multimodal approach in a Smart Home, while Doung et al. [86] use a hidden semi-Markov Model. The data collected may be sufficient to successfully recognize behavior of an elderly. Perceiving what is normal for an individual makes it possible to detect behavior that deviates from what is currently considered normal. This can be used as a warning indicator to health personnel and relatives. The challenge is defining “normal” in this context. One way of perceiving this is to establish a set of features that measure the “normality” of an individual over a period of time with a defined resolution. E.g. a three-month multi-sensory collection registering the amount of movement in residence when an individual is present, toilet visiting frequency, movement during night, medication intake using a wireless medication dispenser, usage of electrical devices like TV, coffee machine, etc., all on a daily or other timely basis. When deviations on several of these features crystalizes, this may be interpreted as a signal that the individual is behaving differently than before and used as an indicator for further investigation. Dykes et al [45] showed that raising the awareness of fall risk in elderly, health personnel and relatives [60], result in fewer falls. However, it remains an open question how the feedback to the elderly and relatives should be designed. In our proposed protocol, the feedback to all actors is configurable. Feedback timing is in this context very important, especially towards the elderly. When loosing balance, we automatically try to counter-act this and prevent the fall from happening. In most cases such actions succeed, resulting in a near-fall incident rather than a fall. Introducing feedback in such a situation may alter the focus on reclaiming balance, thus increasing the risk of a fall. We know that immediate feedback is a most effective way of learning [95], but the value in terms of raising the fall risk awareness using feedback in such situations may prove to be dangerous. It is further important to evaluate how the kind of feedback issued by the system influence the cognitively impaired, how often should the status be updated, and how do this kind of feedback influence behavior in general? These questions need to be investigated to verify the proposed Fall Risk Awareness Protocol along with its implementation. We know that loneliness is common for older people and is associated with adverse health consequences, both mentally as well as physically [58]. A fall risk awareness protocol might motivate elderly to exhibit high-risk behavior due to the increased contact received by both health personnel and relatives. The context-aware fall risk is directly dependent on the prospective fall risk. Thus, information which influence the prospective fall risk like age [37], gender [73], clinical picture [8, 38], polypharmacy [32], history of previous falls [26], weight [47], living alone or not [36], etc., need to be included. Inertial data collected using accelerometer on wrist, insoles, smartphones and other wearables will add information about movement and execution which make it possible to detect dizziness [7] or analyze developments which influence fall risk, such as near-falls [42]. Ambient sensing will help in defining context. Performing analysis and doing calculations in terms of immediate and current fall risk might require computational power both in close proximity to the elderly as well as elsewhere. Approaches on how these requirements can be met should be investigated.

The Fall Risk Assessment Service and fall risk awareness protocol presented in Figure 3 is a conceptual sketch on how a service may be developed and integrated into a fully operational solution. The communication protocols which the protocol and the service will rely on are still to be investigated. Similar investigations have to be executed in terms of both ambient and inertial sensor technologies. A number of fall risk prevention studies have been executed in hospital settings, e.g. [45, 89, 90]. These studies focus on preventive approaches using one or a combination of the fall risk intervention approaches outlined in WHO’s Fall Prevention Model [87] to prevent falls from happening. The approach presented in this paper offers a different setting – the home environment. While a number of fall detection systems have been investigated using a number of different technical approaches, the actual number of systems focusing on preventing falls from happening while being aware of the current situation and how the individual is performing are very limited. We have only been able to identify two such approaches. Caporusso et al. [44] used a wrist-mounted accelerometer for activity recognition and used the activity along with other data in a Risk Assessment Protocol to produce a fall risk assessment that was passed on to the elderly. The process was automated, but manually adjusted to each individual by health personnel. In [35] de Backere et al. presented an approach towards context-aware fall prevention using ambient sensors to create a social and context-aware platform for fall risk assessment with feedback using an ontology-based approach. Both papers offer a path on combining data from different sensory origins, mapping this into a context. The approaches differ on the notion on feedback. In [35] the feedback from the system is given to the caretaker, not the elderly. In Caporusso et al. [44], the subjects got feedback from the wrist-band signaling the probability of having an accident based on current context. However, it is not clear how this accident probability was calculated and whether or not this actually reduced the fall risk.

6

Conclusion and Future Work The approach presented in this paper offers a different approach towards giving feedback and calculating fall risk.

We do not signal the actual fall risk, but rather the expected change of fall risk if activity persists in the current context. This information is more valuable than the current fall risk because it holds information of predictable character of the current context. Secondly, we use prospective fall risk assessments, intrinsic factors influencing fall risk, health information, near-fall registrations, diagnosis, etc. to calculate the expected change of fall risk. This is a much more complete background than [35] and [44] propose. The added information on both context and activity give a more detailed and presumably better foundation for our calculations. Finally, we offer a Fall Risk Awareness Protocol that not only encompasses electronic feedback, but also includes feedback to health personnel and relatives, and how they should react. Fall risk assessment, both prospective and context-aware, is an evaluation of past and present with respect to three areas; the individuals fall history, the individual’s health and behavioral patterns, and the environment in which the individual live.

We have focused on wearable approaches for fall risk assessment and fall prevention and identified information relevant to implement fall risk assessment and fall prevention by investigating current research. Fall risk assessment and fall prevention have been separated into prospective and context-aware elements. Further, we have proposed a Fall Risk Awareness Protocol describing how awareness of the elderly can be altered using wearable sensors in combination with other data representing context. We have discussed some of the challenges that the proposed Fall Risk Awareness Protocol can help solve, and highlighted a number of questions that should be investigated with respect to the protocol before implementation. We have not focused on privacy or ethical aspects of our proposed protocol, but rather made the protocol adaptable in this respect. Both ethical and privacy aspects should be addressed when implementing a protocol as we propose. However, we have not found this to be within the scope of this paper. Further on, technical implementation with respect to communication protocol, security, latency, transmission and processing speed, and technical infrastructure in general are issues that have not been addressed by this paper, but are fundamental to an actual implementation of the proposed protocol. We will in future work address the questions raised and develop a prototype platform for evaluating the Fall Risk Awareness Protocol. The Fall Risk Awareness Protocol presented here and this paper is part of the NIAAR (NonIntrusive Anomalous Activity Recognition) project funded by the Norwegian Research Council.

7

References 1.

Riva F., Toebes M.J.P., Pijnappels M., Stagni R., van Dieën J.H.: Estimating fall risk with inertial sensors using gait stability measures that do not require step detection. Gait Posture 38, 170-174 (2013)

2.

Razak A.H.A., Zayegh A., Begg R.K., Wahab Y.: Foot Plantar Pressure Measurement System: A Review. Sensors (Basel) 12, 9884-9912 (2012)

3.

Toebes M.J.P., Hoozemans M.J.M, Furrer R., Dekker J, van Dieën J.H.: Local dynamic stability and variability of gait are associated with fall history in elderly subjects. Gait Posture 36, 527-531 (2012)

4.

Weiss A., Mirelman A., Buchman A.S., Bennett D.A., Hausdorff J.M.: Using a Body-Fixed Sensor to Identify Subclinical Gait Difficulties in Older Adults with IADL Disability: Maximizing the Output of the Timed Up and Go. PLoS ONE 8:e68885 (2013)

5.

Howell A.M., Kobayashi T., Hayes H.A., Foreman K.B., Bamberg S.J.M.: Kinetic Gait Analysis Using a Low-Cost Insole. IEEE Trans Biomed Eng 60, 3284-3290 (2014)

6.

Stone K.L., Blackwell T.L, Ancoli-Israel S., Cauley J.A., Redline S., Marshall L.M., Ensrud K.E.: Sleep Disturbances and Risk of Falls in Older Community-Dwelling Men: The Outcomes of Sleep Disorders in Older Men (MrOS Sleep) Study. J Am Geriatr Soc 62, 299-305 (2014)

7.

Olsson M., Midlöv P., Kristensson J., Ekdahl C., Berglund J., and Jakobsson U.: Prevalence and predictors of falls and dizziness in people younger and older than 80 years of age--a longitudinal cohort study. Arch Gerontol Geriatr 56, 160-168 (2013)

8.

Graafmans W.C., Ooms M.E., Hofstee M.A., Bezemer P.D., Bouter L.M., Lips P.: Falls in the elderly: a prospective study of risk factors and risk profiles. Am. J. Epidemiol. 143 1129-1135 (1996)

9.

Thorbahn L.D., Newton R.A.: Use of the Berg Balance Test to predict falls in elderly persons, Phys Ther 76, 576-83 (1996)

10.

Robinovitch S.N., Feldman F., Yang Y., Schonnop R., Leung P.M., Sarraf T., Sims-Gould J., Loughin M.: Video capture of the circumstances of falls in elderly people residing in long-term care: an observational study, Lancet 381, 47-54 (2013)

11.

Sterling D.A., O’Connor J.A., Bonadies J.: Geriatric falls: injury severity is high and disproportionate to mechanism, J Trauma 50, 116-119 (2001)

12.

Gates S., Smith L.A., Fisher J.D., Lamb S.E.: Systematic review of accuracy of screening instruments for predicting fall risk among independently living older adults, J Rehabil Res Dev. 45, 1105-1116 (2008)

13.

Howcroft J., Kofman J., Lemaire E.D.: Review of fall risk assessment in geriatric populations using inertial sensors, J Neuroeng Rehabil. 10:91 (2013)

14.

Lockhart T.E., Liu J.: Differentiating fall-prone and healthy adults using local dynamic stability, Ergonomics. 51, 1860-1872 (2008)

15.

Doi T., Hirata S., Ono R., Tsutsumimoto K., Misu S., Ando H.: The harmonic ratio of trunk acceleration predicts falling among older people: results of a 1-year prospective study, J Neuroeng Rehabil. 10:7 (2013)

16.

Brown J., Kurichi J.E., Xie D., Pan Q., Stineman M.G.: Instrumental Activities of Daily Living Staging as a Possible Clinical Tool for Falls Risk Assessment in Physical Medicine and Rehabilitation, PM R. 6, 316-323 (2014)

17.

Weiss A., Brozgol M., Dorfman M., Herman T., Shema S., Giladi N., Hausdorff J.M.: Does the evaluation of gait quality during daily life provide insight into fall risk? A novel approach using 3-day accelerometer recordings, Neurorehabil Neural Repair. 27, 742-752 (2013)

18.

Lee J., Geller A.I., Strasser D.C.: Analytical review: focus on fall screening assessments. PM R. 5, 609621 (2013)

19.

Quandt S.A., Stafford J.M., Bell R.A., Smith S.L., Snively B.M., Arcury T.A.: Predictors of falls in a multiethnic population of rural adults with diabetes, J Gerontol A Biol Sci Med Sci. 61, 394-398 (2006)

20.

van Schooten K.S., Pijnappels M., Rispens S.M., Elders P.J.M., Lips P., van Dieën J.H.: Ambulatory Fall-Risk Assessment: Amount and Quality of Daily-Life Gait Predict Falls in Older Adults, J Gerontol A Biol Sci Med Sci. 70, 608-615 (2015)

21.

Fuller G.F.: Falls in the elderly, Am Fam Physician 61, 2159-2168, 2173-2174 (2000)

22.

Studenski S., Perera S., Patel K., Rosano C, Faulkner K, Inzitari M, Brach J, Chandler J, Cawthon P, Connor EB, Nevitt M, Visser M, Kritchevsky S, Badinelli S, Harris T, New-man A.B., Cauley J, Ferrucci L, Guralnik J.: Gait Speed and Survival in Older Adults, JAMA 305, 50-58 (2011)

23.

Moe-Nilssen R., Helbostad J.: Interstride trunk acceleration variability but not step width variability can differentiate between fit and frail older adults, Gait Posture 21, 164-170 (2005)

24.

Oliver D., Healy F.: Fall risk prediction tools for hospital inpatients: do they work?, Nurs Times 105, 1821 (2009)

25.

Hausdorff J.M., Rios D.A., Edelberg H.K.: Gait variability and fall risk in community-living older adults: A 1-year prospective study, Arch Phys Med Rehabil. 82, 1050-1056 (2001)

26.

Tinetti M.E., Mendes De Leon C.F., Doucette J.T., Baker D.I.: Fear of Falling and Fall-Related Efficacy in Relationship to Functioning Among Community-Living Elders, J Gerontol. 49, M140-M147 (1994)

27.

Shany T., Redmond S.J., Marschollek M., Lovell N.H.: Assessing fall risk using wearable sensors: a practical discussion, Z Gerontol Geriatr. 45, 694-706 (2012)

28.

Tinetti M.E.: Performance-Oriented assessment of mobility problems in elderly patients, J Am Geriatr Soc 34, 119-126 (1986)

29.

Oliver D., Papaioannou A., Giangregorio L., Thabane L., Reizgys K., Foster G.: A systematic review and meta-analysis of studies using the STRATIFY tool for prediction of falls in hospital patients: how well does it work, Age Ageing 37, 621-627 (2008)

30.

Horta E.T., Lopes I.C., Rodrigues J.J.P.C., Misra S.: Real time falls prevention and detection with biofeedback monitoring solution for mobile environments. In: Proceedings of the 2013 IEEE 15th International Conference on e-Health Networking, Applications & Services, pp. 594-600, IEEE, New York (2013)

31.

Heesch K.C., Byles J.E., Brown W.J.: Prospective association between physical activity and falls in community-dwelling older women, J Epidemiol Community Health 62, 421-426 (2008)

32.

Chu L.W., Chi I., Chiu A.Y.Y.: Incidence and predictors of falls in the Chinese elderly, Ann Acad Med Singapore 34, 60-72 (2005)

33.

Alexander B.H., Rivara F.P., Wolf M.E.: The cost and frequency of hospitalization for fall-related injuries in older adults, Am J Public Health 83, 1020-1023 (1992)

34.

Koshmak G., Linden M., Loutfi A.: Dynamic Bayesian Networks for Context-Aware Fall Risk Assessment, Sensors (Basel) 14, 9330-9348 (2014)

35.

de Backere F., Ongenae F., van den Abeele F., Nelis J., Bonte P., Clement E., Philpott M., Hoebeke J., Verstichel S., Ackaert A., de Turck F.: Towards a social and context-aware multi-sensor fall detection and risk assessment platform, Comput Biol Med., http://dx.doi.org/10.1016/j.compbiomed.2014.12.002 (2014)

36.

Kharicha K., Iliffe S., Harari D., Swift C., Gillmann G., Stuck A.E.: Health risk appraisal in older people 1: are older people living alone an “at-risk” group? Br J Gen Pract. 57, 271-276 (2007)

37.

Tinetti M.E., Speechley M., Ginter S.F.: Risk factors for falls among elderly persons living in the community, N. Engl. J. Med. 319, 1701–1707 (1988)

38.

Elliott S., Painter J., Hudson S.: Living alone and fall risk factors in community-dwelling middle age and older adults, J. Community Health 34, 301–310 (2009)

39.

Majumder A.J.A., Zerin I., Ahamed S.I., Smith R.O.: A Multi-Sensor Approach for Fall Risk Prediction and Prevention in the Elderly, SIGAPP Appl. Comput. Rev. 14, 41-52 (2014)

40.

Gillespie L.D., Robertson M.C., Gillespie W.J., Sherrington C., Gates S., Clemson L.M., Lamb S.E.: Interventions for preventing falls in older people living in the community, The Cochrane Library 9 (2012)

41.

Mellone S., Tacconi C., Schwickert L., Klenk J., Becker C., Chiari L.: Smartphone-based solutions for fall detection and prevention: the FARSEEING approach, Z Gerontol Geriatr. 45, 722-727 (2012)

42.

Srygley J.M., Herman T., Giladi N., Hausdorff J.M.: Self-report of missteps in older adults: a valid proxy for falls risk?, Arch Phys Med Rehabil. 90, 786-792 (2009)

43.

Marschollek M., Rehwald A., Wolf K-H., Gietzelt M., Nemitz G., Meyer Zu Schwabedissen H., Haux R.: Sensor-based fall risk assessment - an expert ‘to go’, Methods Inf Med 50, 420–426 (2011)

44.

Caporusso N., Lasorsa I., Rinaldi O., La Pietra L.: A pervasive solution for risk awareness in the context of fall prevention, In Proceedings of the 3rd International Conference on Pervasive Computing Technologies for Healthcare, pp. 1-8, IEEE, New York, (2009)

45.

Dykes P.C., Carroll D.L., Hurley A., Lipsitz S., Benoit A., Chang F., Meltzerm S., Tsurikova R., Zuyov L., Middleton B.: Fall Prevention in Acute Care Hospitals: A Randomized Trial, JAMA. 304, 1912-1918 (2010)

46.

Landsberger H.A.,: Hawthorne revisited - Management and the worker: its critics, and developments in human relations in industry, Cornell University in Ithaca, NewYork, (1958)

47.

Mitchell R.J., Lord S.R., Harvey L.A., Close J.C.: Associations between obesity and over-weight and fall risk, health status and quality of life in older people, Aust N Z J Public Health. 38, 13-18 (2014)

48.

Rubenstein L.Z.: Falls in older people: epidemiology, risk factors and strategies for prevention, Age Ageing, 35-S2, ii37-ii41 (2006)

49.

Doty T.J., Kellihan B., Jung T-P., Zao J.K., Litvan I. The Wearable Multimodal Monitoring System: A Platform to Study Falls and Near-Falls in the Real-World, Lecture Notes in Comput. Sci., 9194, 412-422 (2015)

50.

Martín H., Bernardos A.M., Iglesias J., Casar J.R.: Activity logging using lightweight classification techniques in mobile devices, Pers Ubiquit Comp 17, 675-695 (2013)

51.

Papadopoulos A., Vivaldi N., Crump C., Silvers C.T.,: Differentiating Walking from other Activities of Daily Living in Older Adults Using Wrist-based Accelerometers, Curr Aging Sci 8, 266-275 (2015)

52.

Talavera G., Garcia J., Rösevall J., Rusu C., Carenas C., Breuil F., Reixach E., Arndt H., Burkard S.: Fully-Wireless Sensor Insole as Non-invasive Tool for Collecting Gait Data and Analyzing Fall Risk, Lecture Notes in Comput. Sci., 9456, 15-25 (2015)

53.

O'Brien A., McDaid K., Loane J., Doyle J., O'Mullane B.: Visualisation of Movement of Older Adults within their Homes based on PIR Sensor Data, In: Proceedings of the 6th International Conference on Pervasive Computing Technologies for Healthcare (PervasiveHealth), 252-259 (2012)

54.

Garcia-Valverde T., Botia J.A., Callaghan V., Dooley J.A., Hagras H., Garcia-Sola A.: A Fuzzy LogicBased System for Indoor Localization Using WiFi in Ambient Intelligent Environments, IEEE T Fuzzy Syst 21, 702-718 (2013)

55.

Rialle V, Duchene F., Noury N., Bajolle L., Demongeot J.: Health “smart” home: information technology for patients at home, Telemed J e-Health 8, 395–409 (2002)

56.

Shenghong Li, Collings I.B., Hedley M.: New Efficient Indoor Cooperative Localization Algorithm With Empirical Ranging Error Model, IEEE J Sel Area Comm 33, 1407-1417 (2015)

57.

Lowe S.A., ÓLaighin G.: Monitoring human health behaviour in one's living environment: A technological review, Med Eng Phys 36, 147–168 (2014)

58.

O ́Luanaigh C., Lawlor B.A.: Loneliness and the health of older people, Int J Geriatr Psych 23, 1213– 1221 (2008)

59.

Lee D-C. A., Pritchard E., McDermott F., Haines T.P.: Falls prevention education for older adults during and after hospitalization: A systematic review and meta-analysis, Health Educ J 73, 530-544 (2014)

60.

Ryu Y.M., Roche J.P., Brunton M: . Patient and family education for fall prevention: involving patients and families in a fall prevention program on a neuroscience unit, J Nurs Care Qual, 243-249 (2009)

61.

Tromp A.M., Pluijm S.M., Smit J.H., Deega D.J.H., Boutera L.M., Lips P.: Fall-risk screening test: A prospective study on predictors for falls in community-dwelling elderly, J Clin Epidemiol 54, 837-844 (2001)

62.

Ermes M., Pärkka J., Mantyjarvi J., Korhonen I.: Detection of daily activities and sports with wearable sensors in controlled and uncontrolled conditions, IEEE Trans Inf Technol Biomed. 12, 20-26 (2008)

63.

Podsiadlo D., Richardson S.: The Timed ‘‘Up and GO’’ A test of basic functional mobility for frail elderly persons: J Am Geriatr Soc. 39, 142-148 (1991)

64.

Stevens J.A., Corso P.S., Finkelstein E.A., Miller T.R.: The costs of fatal and non‐fatal falls among older adults, Inj Prev, 290-295 (2006)

65.

Li F., Eckstrom E., Harmer P, Fitzgerald K., Voit J., Cameron K.A.: Exercise and Fall Prevention: Narrowing the Research-to-Practice Gap and Enhancing Integration of Clinical and Community Practice, J Am Geriatr Soc 64, 425–431 (2016)

66.

Verghese1 J., Holtzer R., Lipton R.B., Wang CF.: Quantitative Gait Markers and Incident Fall Risk in Older Adults, J Gerontol A Biol Sci Med Sci 64A, 896-901 (2009)

67.

Nickens H.: Intrinsic Factors in Falling Among the Elderly, JAMA 145, 1089-1093 (1985)

68.

Bueno-Cavanillas A., Padilla-Ruiz F., Jiménez-Moleón J.J., Peinado-Alonso C.A., Gálvez-Vargas R.: Risk factors in falls among the elderly according to extrinsic and intrinsic precipitating causes, Eur J Epidemiol 16, 849-859 (2000)

69.

Nyana M.N., Taya F,E.H., Murugasuc E.: A wearable system for pre-impact fall detection, Journal of Biomechanics 41, 3475–3481 (2008)

70.

Shan S., Yuan T.: A wearable pre-impact fall detector using feature selection and Support Vector Machine, in Proceedings of the 2010 IEEE 10th International Conference on Signal Processing (ICSP), 1686-1689 (2010)

71.

Lee J.K., Robinovitch S.N., Park E.J.: Inertial Sensing-Based Pre-Impact Detection of Falls Involving Near-Fall Scenarios, IEEE T Neur Sys Reh 23, 258-266 (2015)

72.

Weiss A., Shimkin I., Giladi N., Hausdorff J.M.: Automated detection of near falls: Algorithm development and preliminary results, BMC Research Notes 3, art. no. 62 (2010)

73.

Chang V.C., Minh T.D.: Risk Factors for Falls Among Seniors: Implications of Gender, Am. J. Epidemiol. 181, 521-531 (2015)

74.

Fleury A., Vacher M., Noury N.: SVM-based multimodal classification of activities of daily living in health smart homes: Sensors, algorithms, and first experimental results, IEEE T Inf Technol B 14, 274283 (2010)

75.

Muir S.W., Gopaul K., Motero Odasso M.M.: The role of cognitive impairment in fall risk among older adults: a systematic review and meta-analysis, Age Ageing 41, 299-308 (2012)

76.

Callis N.: Falls prevention: Idenfication of predictive fall risk factors, App Nurs Res 29, 53-58 (2016)

77.

Hendrich A., Bender P., Nyhuis A.: Validation of the Hendrich II Fall Risk Model: A large concurrent case/control study of hospitalized patients, App Nurs Res 16, 9–21 (2003)

78.

Morse J., Morse M., Tylko S.: Development of a scale to identify the fall-prone patient, Can J Aging 8, 366–377 (1989)

79.

Oliver D., Brittion M., Martin F.C., Hopper A.H.: Development and evaluation of evidence based risk assessment tool (STRATIFY) to predict which elderly inpatients will fall: Case control and cohort studies, Brit Med J 315, 1049–1053 (1997)

80.

Delahoz Y.S., Labrador M.A.: Survey on Fall Detection and Fall Prevention Using Wearable and External Sensors. Sensors (Basel, Switzerland) 14, 19806–19842 (2014)

81.

Hamm J., Money A.G., Atwal A, Paraskevopoulos I.: Fall prevention intervention technologies: A conceptual framework and survey of the state of the art, J Biomed Inform 59, 319-345 (2016)

82.

Handojoseno A.M.A., Shine J.M., Nguyen T.N., Tran Y, Lewis S.J.G. , Nguyen H.T.: The detection of Freezing of Gait in Parkinson's disease patients using EEG signals based on Wavelet decomposition, 2012 Annual International Conference of the IEEE Engineering in Medicine and Biology Society, San Diego, CA, 2012, pp. 69-72.

83.

Melillo, P., Castaldo, R., Sannino, G., Orrico, A., De Pietro, G., Pecchia, L.: Wearable technology and ECG processing for fall risk assessment, prevention and detection, in Proceedings of the 37th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC), 7740-7743 (2015)

84.

Leone A., Rescio G., Caroppo A., SicilianoP.: A Wearable EMG-based System Pre-fall Detector, Procedia Engineering 120, 2015, 455-458 (2015)

85.

Annese, V.F., De Venuto, D.: Gait analysis for fall prediction using EMG triggered movement related potentials, in Proceedings of the 10th IEEE International Conference on Design and Technology of Integrated Systems in Nanoscale Era (DTIS), 1-6 (2015)

86.

Duong T.V., Bui H.H., Phung D.Q., Venkatesh S.: Activity recognition and abnormality detection with the switching hidden semi-Markov model, in Proceedings of the 2005 IEEE Computer Society Conference on Computer Vision and Pattern Recognition (CVPR'05), 838-845 (2005)

87.

World Health Organization: WHO global report on falls prevention in older age. Available at http://www.who.int/ageing/publications/Falls_prevention7March.pdf?ua=1. Fetched, January 6th, 2016 (2007)

88.

World Health Organization: WHO global report on falls prevention in older age. Available at. http://who.int/ageing/projects/1.Epidemiology%20of%20falls%20in%20older%20age.pdf,

Fetched,

January 6th, 2016 (2007) 89.

Hunderfund A.N., Sweeney C.M., Mandrekar J.N., Johnson L.M., Britton J.W.: Effect of a multidisciplinary fall risk assessment on falls among neurology inpatients. Mayo ClinProc 86,19–24 (2011)

90.

Weinberg J., Proske D., Szerszen A., Lefkovic K., Cline C., El-Sayegh S., Jarrett M., Weiserbs K.F.: An inpatient fall prevention initiative in a tertiary care hospital. Jt Comm J Qual Patient Saf 37, 317–325 (2011)

91.

Cugola G, Margara A.: Processing Flows of Information: From Data Stream to Complex Event Processing. ACM Comput. Surv. 44, 3, Article 15, 15:1-15:62 (2012)

92.

Murphy J, Isaacs B.: The post-fall syndrome: A study of 36 elderly patients. Gerontol. 28, 265-70 (1982)

93.

Brian E.M., Hollidale P.J., Topper A.K.: Fear of Falling and Postural Performance in the Elderly. J Gerontol 46, M123-M131 (1991)

94.

Lawrence R.H., Tennstedt S.L., Kasten L.E., Shih J., Howland J., Jette A.M.: Intensity and correlates of fear of falling and hurting oneself in the next year: baseline findings from a Roybal Center fear of falling intervention. J Aging Health August 10, 267-286 (1998)

95.

Foerde K, Shohamy D.: Feedback Timing Modulates Brain Systems for Learning in Humans. J Neurosci 31, 13157-13167 (2011)

• • • •

A survey of wearables used in fall risk assessment and prevention Separating fall risk into prospective and context-aware A Fall Risk Awareness Protocol is proposed is built upon ta Fall Risk Probability Engine Combining data from ambient sensing with inertial sensing to create a context-aware fall risk assessment service

Elderly

Relatives

Fall Risk Awareness Follow-up Feedback Current Fall Risk Assessment Fall Risk Development

SMS/email/Alarm: Significantly Increased Fall Risk

Contextual Data

Fall Risk Assessment Service Alarms Current Fall Risk Assessment Fall Risk Development Fall Risk Follow-up log

SMS/email/Alarm Increased Fall Risk

Health Personnel

Alarm Notification

Alarm Central

Alarm Notification

Fall Risk Awareness Follow-up

RGB LED Vibration Sound