Fall incidents unraveled: a series of 26 video-based real-life fall events in three frail older persons

Fall incidents and their consequences are a significant threat to older people’s health. In the community setting falls occur frequently: 28 to 35% of older people fall at least once every year, 25% of these fallers fall more than once [1, 2]. In residential care facilities, the average fall incidence is three times higher than in the community [3]. Falls often lead to various injuries ranging in severity from bruises and lacerations to brain trauma and fractures [4, 5]. Additionally, falls have serious psychological consequences (e.g. fear of falling) and result in a significant economic burden [6, 7].

Depending on the study population, 20 to 43% of community-dwelling older persons who fall are not able to get up unaided [8–10]. In sheltered housing and institutional settings 66 to 100% of residents need help to get up [8]. Lying on the ground too long increases the risk of negative outcomes, such as dehydration, hypothermia, pressure ulcers, bronchopneumonia and death [8, 9]. Although there is evidence that unifactorial (e.g. exercise program) and multifactorial (e.g. falls risk assessment and management of identified fall risk factors) preventive approaches can reduce falls in older persons, it is unlikely for any strategy to prevent every single fall incident [11, 12].

In case fall prevention fails, other interventions are needed to reduce the clinical risks associated with lying on the ground too long. A potentially appealing option is a camera-based fall detection system. Our research project ‘FallCam: camera system for fall detection in older persons’ was set up to build a prototype camera system, designed to detect falls [13, 14]. To detect falls and validate the camera system, it is essential to unravel the way in which older people fall. However, this process is difficult to describe. Noury et al. [15] proposed a classification system that divided a fall into four phases; the prefall, critical, postfall and recovery phase. However, the description of falls is mostly based on retrospective methods (e.g. interview), and the accuracy of this approach remains uncertain [16]. Moreover, older persons, especially those with cognitive impairment, often do not recall falls [17]. Obtaining information from real-life images may therefore be critical for developing a fall detection system. In addition, information on activities associated with falls, balance recovery or protective responses such as stepping or grasping and impact to key body sites is important for future fall and injury prevention [16, 18, 19].

To the authors’ knowledge, only one study provided a description of the cause and activity at the time of falling of video-based real-life falls [20]. However, video cameras were installed in common areas of long-term care facilities (e.g. dining rooms, lounges, hallways) and not in the participant’s person living space, as the current study did, which presents a different environmental and situational context.

The aim of this study was to provide a thorough analysis and description of video-based real-life fall events and to examine the phases of real-life falls using the classification system suggested by Noury et al. [15].

This study provides new insights in the details of the fall process, based on a thorough analysis of 26 real-life fall events of three older persons with a high fall risk. The classification of Noury et al. [15] was used to describe the characteristics of the fall process during each fall phase.

Data from the prefall phase showed that most falls happened during walking or transitioning, consistent with previous studies [16, 20, 29, 30]. Additionally, we found that in more than half of the falls a walking aid was used. This differs from Robinovitch et al. [20] who reported the use of walking aids in 21% of falls. They only included falls in common areas which could indicate more mobile and independent participants, as the participants in our study needed assistance to leave their room and were mobility aid users. Another possible explanation for this difference may be because of the small sample in this study (n=3, compared to n=130). Older people often have to use a walking aid because of impaired balance or an increased risk of falling. However, it has also been suggested that walking aids may be a risk factor for falls, because an increased attention for performing the dual-task is required [31]. While our data could support this suggestion, they could also be interpreted in support of the reasoning that frail older people who need walking aids tend to fall more.

Although most studies have not clearly defined the time period, the majority of falls happen during daytime hours [16, 20, 29, 32], with patterns of physical activity contributing to a higher fall risk [33]. Consistent with these findings, participants in our study fell mainly during the day at times when they were most active.

Our results showed that participants fell backward in 62% of falls unlike other studies where participants mostly fell forwards [29] or sideways [34]. This difference might be due to the retrospective self-reportage with a risk of recall bias in those studies. Recall bias can influence outcomes, e.g. time on the ground may be over- or underestimated by the faller and the health care team [8, 16]. Another possible explanation could be that the latter consisted of younger, community-living populations, with healthier participants. One study [33] with frail older persons in nursing homes showed a high level of backward falls similar to our study. Again, the small sample of participants with multiple falls made it difficult to draw any conclusion, as same fallers may tend to fall similarly in terms of fall direction. In addition, fall direction may be defined differently in literature, as the distinction between ‘initial fall direction’ and ‘landing configuration’ is not always clear [28].

Overall, the participants used their call alarm in 54% of the falls, after an average of 70 seconds. A prospective cohort study of 110 older persons residing in their own home, sheltered housing or in residential care [8] showed that 80% of individuals, who fell and had a call alarm, did not use their alarm. A distinction should be made between being unable to call because of practical reasons (e.g. not wearing wireless personal call alarm at the moment of the fall) or because of cognitive problems. Campbell [35] previously reported that activating the alarm may be prevented by impaired planning, coordination, and execution. In addition, Fleming et al. [8] detected a strong association between severe cognitive impairment and lying on the ground for a long time. Although we were unable to assess this association, participant C who was cognitively impaired needed much more time to use her personal call alarm - and as a consequence was lying on the ground for a longer time - as compared to participant B who was not cognitively impaired (Table 2). The difference in the mean time of lying on the ground with (11 minutes; n=14) compared to without a call alarm (28 minutes; n=5) was large, although not significant, which may be due to the small sample size.

Healthcare workers took an average of eight minutes to respond, with a range between two and 33 minutes. Although 33 minutes is a long time, it is probably not uncommon in health care settings, often characterized by low staffing levels. Healthcare workers frequently receive more than one alarm simultaneously without knowing which call is more urgent, or even which is because of a fall.

Time on the ground in the study of Fleming et al. [8] was substantially longer than in our study; 30% of fallers lay on the ground for over an hour. This difference might be explained because information about falls in Fleming’s study was gathered post-fall during a follow-up visit or phone call, leading to possible recall bias; and because of older persons living in different settings. Furthermore, lying on the ground for over an hour was less common in residential care facilities (8%) compared with participants living in sheltered housing (27%) or their own homes (13%). Despite these findings, even in residential care facilities, time on the ground is an important issue given its serious consequences [8].

All participants needed help to get up after a fall. These findings are in contrast again with studies from younger, community living persons where only 20 to 43% were unable to get up after a fall [8–10]. Fleming et al. [8] reported that all participants living in institutionalized care needed help to get up. This challenging matter underlines the importance of fall prevention and regular surveillance of older persons with a high risk of falling.

Overall, our study results were consistent with the classification system by Noury et al. [15]. The real falls corresponded well with the prefall and recovery phase. For the critical phase, our data showed a prolonged critical phase in all falls. This might be due to the fact that the critical phase can be decelerated as deceleration and a prolonged reaction-time is a common feature of older persons [36]. With regard to the end of the critical phase, i.e. the phase where the body normally hits the ground or an obstacle, attention must be paid to the protective measure of rolling because the ‘impact shock’ [15] may be reduced by this movement. Based on our findings, we would suggest distinguishing two separate phases in the critical phase. First, a prolonged loss of balance phase including attempts to balance recovery with protective responses such as stepping and second, the actual descending phase after failure to recover balance in which the vertical velocity increases linearly with time because of gravitational acceleration, resulting in the impact of the body on the ground. In addition, there is a different type of fall, where the fall consists of two parts, e.g. the resident starts to fall but is able to hold on to an object or another person is able to slow down the event and there is a delay in the individual touching the ground. In our study, the postfall phase was not characterized by inactive participants. Being active or inactive depended on the individual. Further research for developing and testing fall detection systems should take these findings into account.

This study was the first to analyze fall incidents in older people’s person living space by means of video images in a real-life environment. One other study measured falls with video cameras in common areas of long-term care facilities (e.g. dining rooms, lounges, hallways) [20], representing a different environmental and situational context. Although a comparison between both studies is difficult due to the small sample in our study, falls in common areas mostly happened during walking forward while falls in private areas happened mostly during transitioning. Mobility aids were used twice as often in private area falls compared to falls in common areas. The majority of studies [8, 10, 29] used self-reported and/or retrospective methods for collecting falls data, which makes the ascertainment of circumstances less reliable. Furthermore, several previous studies used fall simulations to describe the fall process and evaluate fall detection systems. These simulations were mostly uncomplicated falls performed by healthy, young volunteers and not by older persons with an actual risk of falling [37]. Several other studies [38–40] collected fall-related data from real-life falls, using accelerometers and/or infra-red sensors. However, the precise fall process was unknown because image recording was not available. As stated by Bagalà et al. [41], testing fall detection systems in real-life conditions is essential to produce more effective automated alarm systems with fewer false alarms and a higher acceptance. Indeed, our study results give more insight into the complexities of the fall process that should be considered in designing and testing of fall detection algorithms (e.g. most common sequence of events, such as activities leading to falls, and subsequent causes of imbalance).

Although not the aim of our study, another implication is its importance for educational purposes. While analyzing, it became clear that our data were particularly relevant for clinical practice to increase awareness of behavioral and environmental factors causing falls. For example, participant B fell three times in similar circumstances, taking clothes out of a bottom drawer of a closet, suggesting that a rearrangement of the closet might prevent future falls. Another example was the low awareness of staff of possible fall causes (e.g. guiding the participant to bed and forgetting to put the walking aid in the vicinity). The researchers felt ethically obliged to inform staff and management of the facility. At the end of the study, a feedback session was held to inform and educate staff and management of the facility. Video data with modifiable circumstances were presented to discuss falls and fall prevention, after additional approval had been obtained from the participants.

Our study has some limitations. Although 26 falls were recorded, only three older, female residents participated in the study, and of those, one resident only fell once. Despite efforts to enhance participation (e.g. recruitment via staff because of the confidential relationship with residents), the participation rate was low, possibly due to the profound nature of the study, and probably resulted in selection bias. For example, two of the residents had a high MMSE score not representative for the general population in ALR and RCR where most people have impaired cognitive function. Hence, our findings may not be generalizable. However, fall characteristics were explored and a better understanding of the fall process was obtained. Another limitation was that falls outside the rooms of the participants could not be detected, as the FallCam system was restricted to this area. Finally, ethical issues should be mentioned. The use of cameras to detect falls can raise ethical concerns (in particular with regard to privacy). The select group of participants in this study accepted the installation and use of the camera system, but there was some initial hesitation during the recruitment period. Similarly, a British community survey [42] reported that privacy concerns hampered the acceptance of automatic fall detection units. Therefore, further qualitative research is needed to understand older people’s perceptions of the acceptability of this type of surveillance technology before more future investment in the technical development.

During the feedback session at the end of the study, group discussions were held to learn more concerning the perceptions of staff. Overall, staff viewed the technology as positive and thought that a full operational camera system might help their work. The main reasons why the camera system could contribute to the care of older persons were found to be: sense of security for residents, the ability to provide rapid assistance in case of a fall and preventing lying on the floor for a long time after falling. However, technology can only ever be a tool to improve care, not a substitute for the need for skilled and caring people providing care in person.

The development of fall detector systems that can accurately identify falls and alarm accordingly has potential to reduce harmful “long lies” on the floor, but the reliability of such technology depends on detection algorithms reflecting the reality of fall patterns in the population affected. This study contributes to a better understanding of the complexities of the fall process in older persons with a high risk of fall. Further research with a larger and more representative sample is needed to confirm our findings.