This discussion focuses primarily on monitoring everyday lower-limb prosthesis use, representing the majority of the identified studies. However, with upper-limb studies starting to appear in the last decade, the focus of these are also discussed, particularly in relation to how the field may develop in the next decade in light of the trends in lower-limb research. It is worth noting that many of the findings from the lower-limb papers are also relevant to the upper-limb, for example, comments around sample size. The review’s motivation of community-based activity monitoring in low-resource settings is also addressed in each section, informed by the long-term experience of prosthetics service provision by our co-authors from Cambodia, Uganda and Jordan.
Appraisal of studies by classification
Developing and validating actimeters, algorithms or scores for activity classification
There were few papers that developed ways to accurately monitor lower-limb prosthesis use with more detail than a simple step-count, and few that collected information on the types of activities being performed. This review shows that step detection methods have been well-established and are consistent across actimeters, though less accurate at low walking speeds [30, 108]. Step-count can be a useful indicator of exercise, but it does not provide information on the types of activities people with lower-limb absence can participate in and those that still have a barrier to access. Understanding the types of activity performed and whether someone is using transport (from accelerometer data) or leaving the house (from GPS data [38, 94, 95]) can provide an indication of community participation or isolation, which is often a significant issue for people with limb-absence, and can reveal information on physical functioning and gait quality [109, 110]. Body postures, such as sitting versus standing, also affect residual limb shape and volume. Understanding these changes better could improve socket fitting processes and the measurement of outcomes [111]. The ability to detect donning and doffing is also useful for understanding whether a prosthesis is meeting the user’s needs and may give more specific indications, such as physical and/or thermal comfort [39, 45, 112]. Changes in daily prosthesis wear time or the types of activity undertaken over time might provide an early warning of changes in socket fit and tissue health.
There are inertial measurement units that are pre-programmed to classify activities, such as the activPAL (PAL Technologies, Glasgow, UK) and the ActiGraph (ActiGraph LLC, Pensacola, USA). Studies have used these activity classification methods in a laboratory setting [13, 106, 107], but only a couple have used these methods to monitor activity in the community [45, 90]. These activity classification algorithms have mostly been designed for use with sensors worn on the thigh (ActivPAL), the wrist and the waist (studies have used the Actigraph at varying locations but the wrist and waist have the most validation). For long-term monitoring of prosthesis use, the authors suggest that embedding a sensor in the cosmesis of the prosthesis may enable longer monitoring periods, as it would remove the discomfort of having a sensor taped or strapped to the skin and the user would not need to remember to wear it, improving wear compliance. All prostheses can have a sensor attached to them, but not all have the capacity for the sensor to be embedded, given limited space available in the cosmesis. Certain types of sensors are bulky or have a wrist strap, which may make them unsuitable for embedding in the cosmesis or the wrong size to attach to locations other than the wrist. The lateral side of the shaft of a lower-limb prosthesis, just above the ankle, often has space to embed a sensor, and the authors have successfully embedded activPAL and Axivity sensors at this location on various styles of transtibial and transfemoral prosthesis. Attaching the sensor to a stable interface, such as the shank or socket rim minimises noise in the signal. It would be useful to develop algorithms that can detect the type of activity from a sensor located below the knee, so they can be used with below-knee prostheses. The ActiGraph has only a few studies that used the sensor on the ankle to classify activity, and these had low classification accuracy [113]. Sanders et al. [60] have presented some initial activity classification data from a sensor on the prosthetic ankle, whilst Redfield et al. [114] have also developed an algorithm for classifying activities from the prosthetic ankle, but it has not been tested in a daily-life setting [60, 90].
Considering upper-limb prostheses, prosthesis wear time is a key outcome, as if the user does not find the prosthesis to be of value, then it will not be worn. Detection of prosthesis wear/non-wear from accelerometry signals is complex and there is currently no validated upper-limb prosthesis non-wear algorithm. ActiGraph sensors offer two algorithms for the detection of sensor non-wear (“Troiano 2007” and “Choi 2011”), however both measures were developed based on data from hip-worn sensors and these would likely overestimate wear in the case of prostheses [115]. Chadwell et al. [46, 92] published a non-wear algorithm designed to detect prosthesis wear/non-wear using data collected from wrist-worn sensors and compared the calculated wear time to self-reported wear periods, however, this algorithm would require further validation before it is widely accepted. One of the main issues with detecting non-wear of a prosthesis is that the prosthesis can be removed and carried. Additional sensors such as a pressure, lux, or temperature transducer within the socket could offer a potential gold standard for prosthesis wear [105]. Until validated methods of automatically detecting aspects of prosthesis wear and use are available, self-report activity diaries offer an important complimentary measure.
It is important to note that the upper-limb papers reported data on the movements of the arms, or the number of different grasps performed over a specified period. Neither of these measures on their own provide a complete picture and an understanding of both when the prosthesis is worn and how much it is used. For an upper-limb prosthesis there are many aspects of use to consider, including: is the arm used? Are the arm movements similar to those of an anatomical arm or do they reflect compensatory movements? Are the active capabilities of the hand, such as grasping, being used and if so, to what extent? Although the field is in its infancy, many of these issues are beginning to be explored by different groups and hence there is great potential to combine techniques. For example, by combining accelerometry for the detection of arm movements with recordings of grip choice and frequency of use, comparisons could be made with studies of upper-limb activity in anatomically intact populations [116]. Comparing measures such as ‘system-on time’ against prosthesis wear time also make it possible to understand the value of advanced systems such as sensory feedback [47, 58]. It is worth noting that advanced multi-articulating upper-limb prostheses often log data on, for example, the grasps selected, or time powered up, and making these data available in a common and accessible format would help to move the field forward.
Accelerometers were the main type of sensor used in the studies to monitor physical activity. None of the studies monitored for more than a month without the participant returning to the clinic regularly to have the data downloaded, and there have been no longitudinal studies, such as studies that compare the level of activity of a first-time prosthesis user with their level of activity on their second or third prosthesis. Commercial accelerometer-based sensors are useful for research purposes, and the data recording capacity and battery life are increasing. However, they are still limited to a maximum of 3–6 months of recording time before the data needs to be downloaded and the battery recharged or replaced. These limitations, along with the expense of sensors, make them currently impractical for standard clinical use, particularly in low-resource settings. Cloud storage and inductive charging in areas with regular access to internet and reliable electricity may make long-term recording feasible.
Comparing prosthetic components and intervention strategies
A substantial portion (1/3) of the lower-limb studies focused on comparing prosthetic components. This is unsurprising, given the importance of increasing comfort and function of prostheses, along with improving access and affordability.
Most of these studies that compared prosthetic components did not find clinically significant results. This is not necessarily because there is no difference between the products, but is more likely because of the limited sample sizes, the wide variability between individuals that make it difficult to match controls, the limited time-frames for comparing results, and the insensitivity of the compared outputs (most only looked at step-count, not activity or gait quality). Small sample-size is a common issue in the field of prosthetics, as it is difficult to recruit large participant cohorts from the small limb-absent population [117]. One way to help address this issue is to develop a commonly-agreed framework for reporting participant characteristics, clinical outcomes and the engineering characteristics of the components tested, so that comparisons can be made across studies and the foundations laid for big data approaches [118]. Strengthening the partnerships and collaborations between academic institutions, the prosthetics industry, clinics, hospitals and societies of people with limb absence is important for ensuring research is informed by an understanding of the users’ needs, and that the research outputs inform changes in clinical practice. Having strong links within the prosthetics community, and empowering end-users to contribute to the research should also assist with recruitment.
Many of the studies ran the different interventions on the same participants so that each participant was their own control, to overcome the challenges of finding a well-matched control. However, this has its own limitations, as the order of testing the interventions can affect the outcomes, due to training effects, and the time of week or season might affect activity levels and types (i.e. due to weather, working patterns and religious practices such as Ramadan).
There were few studies that assessed lifestyle interventions for people with lower-limb absence, but the positive results, particularly in interventions to promote physical activity, demonstrate the importance of an inter-disciplinary approach to providing rehabilitation and community support, rather than simply providing the prosthetic device [67, 69, 80, 81, 86].
The role of inter-disciplinary rehabilitation begins before prosthetic fitting and continues after the prosthesis has been provided. Sensors that monitor non-use could also be useful for assessing issues with prostheses, to help identify where better training and/or support is needed, and to help prioritise where clinicians focus their efforts.
Comparing activity levels to clinical scores
In the lower-limb studies, the main clinical score that everyday activity levels were compared to was the K-level score [30, 43, 44, 49, 50, 54, 55, 95]. K-levels are the standard for classifying an individual’s current and potential functional status, particularly regarding ambulation. This classification was developed in 1995 and there is no gold standard method for establishing K-Levels [29, 119]. The common suite of tests used are clinic-based, providing information on the ability of the individual, rather than on their everyday functionality and needs. The ability level of the individual when ambulating in a clinical environment does not necessarily match how much they ambulate in their typical environment [55]. Nevertheless, the studies reviewed in this paper found that participants’ everyday physical activity mostly correlated with the K-level scores. However, monitoring participants in their daily life provided additional information that could complement the clinical measures to provide clinicians with a clearer picture of the individual’s capabilities and requirements. For example, the clinical K-level classifications were based on a physical examination at a single point in time, informed by clinical experience and subjective activity reports from the patient and family, whereas community activity monitoring increased the objectiveness in selecting suitable prosthetic components, adding a continuous element to assess changes in activity level over time [49, 54]. Community monitoring offers repeatable, objective criteria of functional level, based on the individual’s daily activities and environment.
Comparing different populations
The papers on comparing populations with lower-limb absence ranged in topic, but all demonstrated that there are significant differences between individuals and populations, so a one-size-fits-all method to providing prosthetics and services is not appropriate for meeting user needs. Of particular note in these studies was that people with vascular disease consistently showed lower levels of physical activity, so exercise-based interventions are possibly particularly important among this population. Few studies compared populations with BK limb absence to those with AK limb absence, but those that did found that individuals with AK limb absence walked fewer steps per day [93], walked for less time per day [28] and walked more slowly [52]. The characteristic of gait also differed, with AK individuals having a greater mean step width than those with BK limb absence [52]. Individuals with BK limb absence had higher activity levels on weekdays than weekends, but this difference was not observed in individuals with AK limb absence [77].
Activity monitoring for low-resource settings
Benefits of activity monitoring in community-based settings
Ultimately, the measures of success for any prosthesis are whether the user chooses to wear it and use it to perform the functions it was designed for. In the case of lower-limb prostheses, the primary function is safe ambulation and in the upper-limb, it is prehensile function and the ability to locate and orient the prehensor in the reachable workspace. While there are many lab-based assessment studies investigating the extent to which particular prosthetic devices restore gait quality or upper-limb function, this review suggests that further work is needed to understand real-world needs and physical activity practices of prosthesis users and the factors which influence them.
In low-resource settings, public limb fitting centres and NGOs can only provide prostheses with basic functionality for a limited number of people annually, leaving hundreds or thousands of people waiting [1]. Furthermore, prostheses often require replacement after about 3–5 years due to wear and changes in residual limb size and shape. There are challenges worldwide around collecting meaningful measures of prosthetic rehabilitation outcomes, with some clinicians overwhelmed by their outcome measure workload and others performing only subjective evaluation of functional activity at the time of discharge [120]. Activity monitoring and gait assessment can perhaps provide direct measures of prosthesis use and therefore help the decision-maker to decide the appropriate type of prosthesis for an individual and the point at which to replace the prosthesis. Over a longer term, with larger datasets, service providers may use this approach to provide a cost effectiveness assessment of different prosthetic devices, and appraise new components.
Beyond amount of use, activity monitoring methods also offer an objective insight into how a prosthesis has been used. Currently, users retrospectively report about their experiences when they are seen by a healthcare professional. In low-resource settings, factors including a shortage of prosthetists and a lack of transport for those based in rural communities mean that it can be a long time before feedback is given to the provider [121]. Consequently, often users will remember only the very bad experiences, biasing their reports. Conversely, in some cultures service users will provide little negative feedback, especially where prosthetic devices have been given to them free of charge. In both situations, they may experience worsened musculoskeletal health and soft tissue injuries by needing to wear an ineffective prosthesis for longer. In some facilities, NGOs have developed community outreach programs where they provide assessment of prosthesis use among other services. If it was possible to monitor prosthesis use remotely, this could help to inform the decision-making process, and provide earlier intervention for users experiencing problems, as well as identifying devices with high rates of successful use. Real-time monitoring may even allow identification of users experiencing acute injury or mental health problems, evidenced by sudden changes in activity level or type [122, 123].
Barriers to activity monitoring
Despite the potential benefits of activity monitoring, barriers remain, associated with cost, access, training and capacity, as well as technical and cultural aspects of their use. Although commercial activity monitors are readily available in high resource settings, their cost-benefit balance must be considered in low-resource settings, especially with regard to widespread, real-time assessment. Adding embedded sensors to prostheses, and arranging for mobile or periodic connectivity may make them unaffordable for the services and users needing them most [124]. Furthermore, this would also require robust access to communication and information technology in low resource settings and for people with disabilities. Barriers to communication and challenges with accessing clinics (e.g. due to cost or availability of transport, and ability to take time off work) mean that in low-resource settings, losses to follow-up are likely to be more common. The cost associated with the potential loss of sensors may be particularly significant for clinics and researchers in low-resource settings. This emphasises the need for low cost appropriate monitoring technologies.
In both high and low-resource settings, many of the algorithms used to analyse gathered activity data are not user-friendly and require at least basic skills in programming and signal processing. Some fast, user-friendly analysis tools do exist, as in consumer activity monitors, but these represent an addition to busy clinicians’ workloads, where there are limited human resources and access to appropriate tools to undertake effective monitoring. More complex service-wide data analysis is in many instances time consuming, and requires specialist epidemiology training and statistics knowledge, which makes it impractical for clinical settings. Perceptions about how useful the information gathered from outcome measurements is for improving services for people with limb absence varies between countries, and countries that struggle to financially support systemic changes often see little value in gathering data on outcome measurements.
There is still a need to train clinicians in measuring outcomes, particularly in objective evidence assessment, multidisciplinary team integration and technology transfer. Client education is also essential for them to be able to participate fully and provide useful feedback, especially in low-resource settings. The training of clients must be accessible across varied literacy levels, as well as being culturally aware and co-designed using participatory research methods.
Evaluating prosthesis use in low-resource settings has challenges beyond access to measurement tools, limitations of current measuring tools, and the training of clinicians in how to use these tools. In Jordan, for instance, lower- and upper-limb prostheses are rarely evaluated, not due to the lack of awareness of the importance of the evaluation procedure, but rather due to the difficulty of implementing any rehabilitative intervention informed by the results of evaluation. The number of people with limb absence, and the lack of accessibility of limb fitting centres and of trained inter-disciplinary rehabilitation teams to deal with prosthetic training and problem solving were identified as key prosthetic and orthotic service access barriers by the WHO in 2011 [26, 125], and are issues in many countries.
Culture also affects the every-day use of activity monitors. It is important to have well-trained professionals and clinicians who understand the context and activities-of-daily-living of the assessed group to interpret the data. For example, considering activity types, the reviewed studies did not differentiate between social/community activities and work-related activities, nor did they evaluate the context of such activities. There are complex and nuanced links between disability, poverty and health [126]. In low-resource settings without social support systems, if a person is active because they must be active (work, school, child care, or other responsibilities), then activity may not relate to the function and comfort of the prosthesis. The reviewed studies showed diversity in location and type of data collected, but it would be useful to include in future studies the ethnography of participants to assess whether particular groups and lifestyles are more physically active, regardless of the prosthetic components available.
User-centred development of activity monitoring technology and methods must consider the prosthesis user as well as the clinician. In the present review, no articles reported on the user’s acceptability of the actimeters. It is important to understand the needs and ergonomic factors related to the use of actimeters. Monitoring tools which are bulky or not concealed within the prosthesis may be intrusive for users if they raise questions about what they are. Furthermore, monitoring an individual’s activities may be seen in some cultures as an invasion of privacy, so it is important for individuals to consent to what data is collected, how their data is used and to who has access to it [127, 128].
Recommendations for future research utilising activity monitors to track prosthesis use
There have been few studies exploring psychological aspects, such as prosthesis embodiment [129], sensory preference [130] and attitudes of communities to disabilities [131] on wear and use of prostheses. It would, therefore, be useful to collect long-term data on community-based activities, particularly in regard to community participation and isolation, which is a common issue found amongst prosthesis users, and has been linked to quality of life scores [17, 132]. Physical activity monitoring in the community may also enhance knowledge of the links between physical activity and other factors, including prosthetic socket fit for comfort, function and reducing energy consumption. Socket fit plays a significant part in successful rehabilitation and restoration of function and mobility, but tools to objectively evaluate socket fit are lacking [133].
Most studies did not report on factors such as the weather, time-of-week, season or whether a walking aid was used. Factoring in these other aspects can provide greater understanding of the variations in an individual’s activity, and provide better support for clinical scores and prosthetists’ decisions. Some upper-limb studies assessed wear-time [33, 46, 91, 92], but most lower-limb studies did not. Wear-time, in addition to the total amount of activity, could give a better indication of whether there are issues with prosthesis comfort and whether users find a prosthesis beneficial in all situations, or only in certain situations (for example, many may use a prosthesis in public but not in private, or only outdoors, not indoors). The studies also did not report on durability or waterproofing of the sensors, which has particular relevance for sanitation chores, such as hand washing clothes [134], and for long-term-use assessment in rainy or humid environments.
Development of algorithms that allow sensors to provide detailed movement analytics, including information on gait symmetry, stability for safe ambulation, stride length, compensatory movements and upper-limb movement analytics could provide additional information to inform clinicians as they plan rehabilitation and exercises for prosthesis users, to increase prosthesis functionality. When selecting sensors to monitor physical activity, it is recommended by the authors that sensors are used that allow access to the raw data, as this enables bespoke data processing and study reproduction without the limitations of specific manufacturers.
Long-term monitoring of prosthesis-use and developing shared datasets supported by metadata standards may provide early warning of changes in socket fit and tissue health, enable comparisons to be made across studies to assist service providers in assessing prosthetic components, and help identify the unmet needs of prosthesis-users [135].