Secondary Logo

Journal Logo


Monitoring Prosthesis User Activity and Doffing Using an Activity Monitor and Proximity Sensors

Gardner, David W. MSME; Redd, Christian B. MSE; Cagle, John C. BSE; Hafner, Brian J. PhD; Sanders, Joan E. PhD

Author Information
Journal of Prosthetics and Orthotics: April 2016 - Volume 28 - Issue 2 - p 68-77
doi: 10.1097/JPO.0000000000000093
  • Free


People with lower-limb loss are often prescribed a prosthesis (i.e., artificial limb) to perform basic functional activities, such as standing, transferring, and walking.1,2 Use of a prosthesis also promotes a broad range of desirable patient outcomes, including self-reliance, safety, quality of life, independence, community access, participation in desired life activities, and engagement in recreational and vocational activities.3 Conversely, nonprovision or nonuse of a prosthesis is often associated with poor health outcomes.2,4,5 Prescription and successful use of a prosthesis is therefore considered a primary goal in the rehabilitation of persons with lower-limb amputation.6

Given the aforementioned health benefits, monitoring when (i.e., how much) and how (i.e., in what way) a prosthesis is used may be of significant clinical value. Monitoring prosthesis use in clinical practice may help clinicians monitor patients' progress during early rehabilitation, select appropriate prosthetic components at prosthetic transitions, or diagnose fit issues when skin breakdown appears.7 Monitoring use may also aid in evaluating and documenting the effectiveness of prosthetic and/or therapeutic interventions.8

A variety of strategies to monitor prosthesis use have been attempted. Subjective methods, including focus groups,9 interviews,2,10 surveys,11,12 or diaries,13 have been used previously to elicit information about individuals' prosthetic use. However, subjective methods of data collection are often limited by factors such as generalization, comprehension, perception, honesty, and recall,14–18 and they may not accurately characterize a patient's use of a prosthesis. Objective approaches to activity monitoring, such as foot switches,19–21 pedometers,8,22 activity monitors,23–25 and global position sensors,26,27 have been advocated to address concerns with subjective methods. However, the information provided by sensing technologies such as these may also be limited.

A fundamental challenge with using pressure sensors, accelerometers, or global position sensors to measure prosthesis use is that these devices are generally unable to distinguish a stationary prosthesis from a doffed prosthesis. Because the prosthesis is inert, subtle twitches or other movements may not be present or detected when the user is still. Thus it may be impossible to discern a prosthesis that is worn for sitting or standing, compared with one that is not worn at all. Clinically, distinction of these situations is needed to accurately characterize a patient's habitual activities or to diagnose reasons for socket fit changes throughout the day. For example, sitting with the prosthesis doffed can affect residual limb fluid volume differently than donned.28 Prosthesis users who doff their prosthesis temporarily may recover limb fluid volume and retain it during periods of subsequent activity, offsetting their need to add a prosthetic sock. Changes in fit may also be affected by the amount of time a user spends in different postures, whether wearing or not wearing the prosthesis.29 Prosthesis users typically lose limb fluid at a higher rate during standing than during walking or sitting.

Given the clinical importance of differentiating a doffed prosthesis from static postures, and differentiating between static postures (e.g., sitting vs. standing), a strategy to address the inherent limitations to contemporary monitoring solutions is needed. The goal of this research was therefore to develop a method to identify activities and distinguish a donned from a doffed prosthesis. A noninvasive monitoring technology was developed and evaluated with prosthetic limb users in both laboratory tests and in field trials.


A small, lightweight monitoring system that could be easily integrated into a lower-limb prosthesis was developed. The monitoring system hardware consisted of two accelerometers, two proximity sensors, and a signal processing and storage device. Postprocessing software converted collected data to a list of time-stamped activity states including doffed, walking (includes any type of movement), standing, and sitting.



Two commercial three-axis accelerometers (Actilife Actigraph [AG] GT3X+, Pensacola, FL, USA) were used, one mounted on the prosthesis and the other strapped to the thigh (Figure 1A). The accelerometers were set to record at 40 Hz (range, 30–100 Hz), a sampling rate which demonstrated a battery life of up to 3 weeks. Measured acceleration limits were −6 G to +6 G with a sensitivity of 0.003 G. Raw data from the accelerometers were exported in a comma-separated values (CSV) format and processed in MatLab (Mathworks, Natick, MA, USA) using the algorithm described below. Each accelerometer had an internal clock that time stamped the recorded data.

Figure 1
Figure 1:
Monitoring system. A, AG monitors positioned on thigh and lower leg. B, Proximity sensor mounted on the lateral brim of the socket. C, Electronics and battery for proximity sensors housed in custom box. D, Socket prepared for data collection. The electronics/battery box for the proximity sensors is within a small box on the lateral aspect of the socket, covered with tape.


Two proximity sensors were mounted to the medial and lateral proximal socket brim to determine if the prosthesis was donned or doffed (Figure 1B). Infrared proximity sensors (Fairchild QRE1113; Fairchild Semiconductor Corp, San Jose, CA, USA) were chosen because of their relatively short detection distance. We set the threshold detection distance at 80% of the maximum analog-to-digital count so that the proximity sensors identified objects only if they were within approximately 1.0 cm of the inside socket surface. We considered this threshold distance acceptable because a user's residual limb would generally not be more than 1.0 cm from both sensors while it was in the socket. Conversely, if a spare liner and/or socks were placed within the socket while it was doffed, they would likely be more than 1.0 cm from at least one sensor. The proximity sensors were sampled at 10 Hz, a rate found to accurately detect events of interest while minimizing the power consumption of the system.

The two proximity sensors were housed within custom holders constructed using a three-dimensional additive fabrication system (Objet30 Pro; Stratasys, Eden Prairie, MN; Figure 1B). The two sensors were placed in two separate locations. The first sensor was placed on the brim of the socket, between the medial and posterior medial aspect. The second sensor was similarly placed on the brim between the lateral and anterior lateral aspect. These positions ensured that the residual limb was detected even if the participant's thigh was anterior of the socket (e.g., during standing with the prosthetic limb posterior to the contralateral limb) or posterior of the socket (e.g., during sitting with 90° of knee flexion). Sensor positions were adjusted slightly, if needed, to ensure they did not adversely affect each user's ability to flex or extend the knee.


An Arduino Pro Mini (Sparkfun Electronics, Boulder, CO, USA) was used to sample data from each proximity sensor's ADC output and store it to a micro SD card. The Arduino was run at 10 Hz. The Arduino microcontroller internally tracked the time elapsed from the researcher-initiated start time. The Arduino and data storage electronics (mass 60 g) were housed within a custom enclosure attached to the lateral surface of the socket (Figures 1C, D). The entire system was powered by a rechargeable 1000 mAh polymer lithium ion battery. The system was able to sample data continuously for approximately 72 hrs before needing to be recharged.

Data from the proximity sensors were stored to the micro SD card as a plain text CSV file. The data collection start time was recorded by the system, and each sensor reading was subsequently time stamped. After completion of the testing period, the SD card was removed and the data downloaded to a personal computer (PC) for analysis.

Because the Arduino was programed to record the start time of the microcontroller during initialization, we aligned the time stamps between the accelerometers and Arduino microcontroller during postprocessing.


An activity-detection algorithm was designed to distinguish a donned from a doffed socket and to characterize activity in detail—sitting, standing, and walking (includes any type of movement). A flowchart of the algorithm is shown in Figure 2.

Figure 2
Figure 2:
Flowchart for processing data to characterize posture and activity. For θ t, the 0° reference axis is vertical.

To distinguish a donned from a doffed socket, outputs from the two proximity sensors were summed and compared against a threshold for limb presence established during calibration for the participant. By summing the two sensor outputs, we eliminated false positives that may have occurred if the participant doffed and overlaid a sock or liner on the socket. Data with a sum less than the threshold were considered donned and the signal set to “low,” while data greater than the threshold were classified as doffed and the signal set to “high.”

If the algorithm determined the socket was donned, the next step was to determine if the activity was static or dynamic. The knee jolt rate (KJR), which was the absolute value of the first derivative of the angle between the lower leg and thigh acceleration vectors in the sagittal plane (θk), was calculated using the equations

where L and T were the acceleration vectors in the sagittal plane (i.e., included x and y components only).

The participant's activity was considered dynamic if KJR was above a threshold value experienced while the participant drove or rode in a vehicle. The vehicle-riding threshold was determined experimentally for each participant. After arriving back at the laboratory from field testing, participants were questioned as to if they drove or rode in a vehicle. Upon inspection of accelerometer (L and T) and KJR data versus time within or near periods of reported vehicle use, we were able to easily distinguish accelerations caused by traveling in a vehicle from those produced by other activities. Knee jolt rates from traveling in a vehicle were of lower magnitude than those from walking and higher magnitude than those from static activity. A threshold KJR value between vehicle riding and walking was selected, and only points with KJR values greater than the threshold were considered dynamic activity. For most participants, 15°/second served as the threshold, but some participants required slightly higher thresholds at 20°/second. This was likely due to a more aggressive driving style or stiffer suspension on the vehicle.

If the participant's activity was dynamic, we investigated if the movement was continuous. If an additional KJR point greater than the vehicle riding threshold occurred within 3 seconds, then the movement was deemed continuous and classified as walking. If there was no such point within 3 seconds, then the activity was characterized as static and the accelerometers were considered to behave as inclinometers and used to identify the participant's posture as described below.

The participant's activity was deemed static if KJR was below the vehicle-riding threshold. If the participant was static, then the variable θk represented the knee angle. Posture was classified as sitting if the knee angle was between 35° and 145°. If the knee angle was between 145° and 180° (inclusive) and the thigh angle relative to the ground plane was greater than 53°, then the activity was classified as standing. These angles were selected based from preliminary testing on people with limb loss. If the thigh angle was less than or equal to 53°, then the activity was classified as lying down or sitting in an inclined chair, such as a dental chair. If the knee angle was less than 35°, the data remained unclassified (“unknown”).

Additional processing was conducted to ensure coherency within bouts of activity. If a brief classification of activity less than 3 seconds in duration (e.g., <3 seconds of standing) occurred within a long and otherwise continuous classification of another activity (e.g., sitting), the brief classification was considered incorrect and was reclassified as unknown. In addition, any classification of activity with a duration of less than 0.5 second was reclassified as unknown. After this postprocessing, periods of continuous activity (i.e., segments with such as classification) were determined and their start and end times listed in a table. Activities were subsequently plotted as a series of color-coded bars with the width of the bar proportional to the duration of the activity.



Volunteers were eligible for inclusion in the study if they had a transtibial limb amputation and were a limited community-level ambulator or higher (>K-2) on the Medicare Functional Classification Level scale. Other inclusion criteria included the ability to participate in low to moderate activity for 60 minutes, including bouts of sitting, standing, and walking. Participants were required to use a prosthetic limb for an average of at least 4 hrs/day. Exclusion criteria included current skin breakdown. Participant suitability and inclusion/exclusion criteria were evaluated by a clinical prosthetist before the beginning of any study procedures. All study procedures and materials were reviewed and approved by the University of Washington Institutional Review Board. Informed consent was obtained from participants before any study procedures were initiated.

When participants arrived at the laboratory, they were asked about their health and medical history by the research prosthetist. Relevant anthropometric data, including height, weight, and residual limb dimensions, were collected. Participants then doffed their prosthesis, and researchers installed the study monitoring system on the prosthesis and participant (Figures 1A–D).


Participants underwent a scripted protocol in our laboratory for approximately 10 minutes. The protocol included specified periods of doffing, sitting, standing, and walking. The intent of this calibration protocol was to collect data under known postures (e.g., standing, shallow sitting, and deep sitting) and provide a ground truth comparison for subsequently collected data (i.e., to ensure that the selected proximity sensor threshold was appropriate and both accelerometers were installed with appropriate orientations). This test was particularly important for participants who had a conical thigh since positioning the accelerometer too far anteriorly tended to cause the measured angle at the knee to be greater than 180° (i.e., hyperextended), which confounded analysis. This calibration was performed before both scripted out-of-laboratory testing and field testing (below).


Participants next underwent a scripted activity protocol within and outside of the laboratory building. The controlled protocol lasted approximately 30 minutes' duration, but varied based on the participant's walking speed. Participants were first asked to complete a series of indoor activities that included walking through hallways, standing for brief periods, sitting in a chair, and doffing their prosthesis (while seated). Participants then walked outdoors around the building perimeter on sidewalks. Before reentering the laboratory, they were asked to sit for 1 minute, and stand for 1 minute. No guidance was provided for exact posture participants were to use during each type of activity. There were approximately 2 doffs, 12 sits, 6 stands, and 19 periods of walking over the course of the out-of-laboratory testing. Participants used an elevator when changing floors, unless they voiced a preference to use stairs. A researcher followed the participant, documented the ordering of activities, noting approximate times of activity changes using a stopwatch.


Field testing was conducted on participants to measure activity over longer periods. For field testing, participants visited the laboratory twice within 1 day. During the first visit, the participant and prosthesis were instrumented with the sensors described above and the calibration protocol conducted. Participants were asked to leave the laboratory for a minimum of 3 hrs and conduct activities consistent with their regular routine. Upon returning to the laboratory, we queried the participants about activities performed after leaving the laboratory. The researcher visually inspected the data, identified periods where there appeared to be inconsistency between the participant's description and the collected data (if any), and queried the participant for more detail about those times.


Data were collected from 11 participants during scripted out-of-laboratory testing and 23 participants during field testing. Participant characteristics are listed in Tables 1 and 2, respectively.

Table 1
Table 1:
Participant characteristics for scripted out-of-laboratory testing
Table 2
Table 2:
Participant characteristics for field testing

Typical accelerometer and proximity sensor data collected during the scripted out-of-laboratory test protocol are illustrated in Figures 3A to C. The calculated postures and activities are illustrated in a plot of knee angle and a bar graph, color coded to reflect the activities (Figure 3D).

Figure 3
Figure 3:
Data from a participant performing the out-of-laboratory scripted testing. Raw data from (A) socket-mounted accelerometer and (B) thigh-mounted accelerometer. C, Processed data from proximity sensors. D, Plot of angle between leg and thigh acceleration vectors (dynamic activity), or knee angle (static activity). The bar chart at the bottom is color coded to reflect the detected activity.

The activities classified by the system well matched those documented by the researcher during the scripted out-of-laboratory testing. All activities matched except for one 60-second stand posture for participant 9 that was classified by the system as sitting (Table 3). The number of times each activity was conducted varied among participants because some participants stood for several seconds between sitting and walking, while others did not. Some participants also took brief breaks during the walks, while others elected to walk continuously.

Table 3
Table 3:
Results from out-of-laboratory testing

Out-of-laboratory activity durations averaged 13.1 minutes for sitting, 5.8 minutes for standing, 13.8 minutes for walking, and 2.4 minutes for doffing. Mean absolute differences (±standard deviation) of total activity durations measured using the monitoring system, compared with those calculated from the paper notes, were 21 (±10) seconds for sitting, 20 (±18) seconds for standing, and 14 (±10) seconds for walking. These differences corresponded to 2.7% (±1.3%) of the total time for sitting, 5.7% (±5.2%) of the total for standing, and 1.7% (±1.2%) of the total for walking. There were an average of 43 transitions per session. Thus the average timing error per transition was 1.2 seconds. We expect these differences were due primarily to the researcher's criteria for a transition being different from the algorithm, and the criteria being executed less consistently by the researcher than by the algorithm. The limited resolution of the researcher recording times, 1 second, further reduced accuracy of the paper notes.

The average (±standard deviation) length of field testing was 4.6 (±0.5) hrs. Results from the monitoring system well matched participant's descriptions of activities away from the laboratory, as illustrated in Figure 4. In two cases, participants needed to be queried because of inconsistency between the verbal description and collected data. One of these participants reported that he drove to work while out of the laboratory. Collected data indicated that the participant doffed during this time. When questioned, the participant replied that he always doffed while driving but forgot to report doing so. In another case, a participant reported that she had used her prosthesis for the entire day without removing it for any period. Collected data indicated that the participant had doffed for a 50-minute period in the evening. During follow-up questioning, the participant recalled that she had gone out for a massage, and doffed her prosthesis for the duration.

Figure 4
Figure 4:
Field testing results. Exemplary data from a 3.7-hr test session away from the laboratory (participant E). Activities included the following: 1) walked from laboratory to bus stop; 2) riding the bus; 3) walking from bus stop to a building for personal business; 4) riding the bus; 5) walk from bus stop to restaurant near the laboratory; 6) lunch at the restaurant; 7) walk from restaurant to a grocery store to purchase a beverage; 8) undocumented sitting; 9) walk back to the laboratory; 10) resting in lobby outside the laboratory; 11) walk into laboratory; 12) sit in laboratory.

Doffing events detected by the monitor system during field testing matched descriptions for 32 of the 37 doffs conducted by the 23 participants (Table 4). Incorrect classifications occurred as a result of incomplete recall by the participant for three of the five misclassifications (participants C, L [first of two misclassifications], and T). For the other two misclassifications (participants L [second of two misclassifications] and Q), whether the participant doffed briefly or not is unclear. We conclude that the monitoring system detected at least 95% of doffs (35 of 37) correctly. We note that participants I and O reported partially doffing their socket (i.e., they unlocked the pin suspension, but did not withdraw their limb from the socket) during their sessions. The monitoring system operated correctly, and did not interpret those actions as doffs. We also noted that participant I during his lunch break doffed his prosthesis and then propping up his residual limb within the doffed socket. The monitoring system operated correctly and did not interpret the propped up limb as a donned socket, though we did note that the proximity sensor signal intensity approached the threshold value characterizing donning.

Table 4
Table 4:
Results from field testing: number of doffs reported and captured


The strategy of combining information from proximity sensors and an accelerometer was effective in accomplishing one of the goals of this research—distinguishing a doffed prosthesis from a sitting or standing prosthesis. In 100% of the scripted out-of-laboratory protocol tests and at least 95% of the field tests, doffs were correctly detected.

Proximity sensors, such as those used in this study, offer advantages over force-sensing resistors (FSRs) (thin pressure sensors)30,31 that might otherwise be used for distinguishing a donned from a doffed prosthesis. The proximity sensors performed consistently well on each participant tested. Once positioned on the socket flap during calibration, the sensors did not need to be adjusted. The output signal was low noise and did not drift from baseline during testing. In preliminary tests using FSRs, we found it difficult to position the sensors such that they were under pressure for all the sitting and standing postures in which a prosthesis user engaged. Force-sensing resistor positions often needed to be adjusted for each individual participant tested. Problems with FSR drift and noise have been recognized.30–32

Once the don/doff distinction was made, the monitoring system effectively distinguished standing from sitting based upon the angles of the thigh and knee. Participants tended to sit with their thigh oriented at less than the 53° angle threshold, and their knee either near the 35° end of the spectrum with their prosthetic limb extended away from them, or near the 145° end of the spectrum with their knee flexed so that the prosthetic foot was under their thigh. If a participant stood with weight primarily on the contralateral limb, extending the prosthetic limb outward at a shallow angle, standing could be misidentified as sitting. This occurred once in the present study (Table 3, participant 9). Distinguishing standing from sitting is a recognized problem when placing accelerometers at the ankle or at the torso, though the literature suggests that specialized machine learning algorithms may be useful towards making this distinction.25,33,34 Until algorithms are perfected, providing users with probability estimates for unknown periods of activity (e.g., 25% likelihood of the activity being standing, and 75% likelihood of the activity being sitting) may be a strategy for communicating uncertainties in the resultant data.

Results from field testing suggest possible enhancements to the monitoring system and processing algorithm. Moving the proximity sensors distally so that they are within the socket wall in the middle of the socket may reduce the likelihood that propping up the residual limb within a doffed socket is detected as a doff-to-don transition. Including all detected doffs within a specified time range (e.g., when a user doffs and dons several times within a few minutes to achieve a good fit) as a single doffing event may be clearer to the practitioner interpreting collected data. Incorporating power management strategies to the microcontroller so as to allow sampling for weeks rather than days may help practitioners better identify trends or changes in the prosthesis user's activity and prosthetic use behavior.

It is useful to clinically distinguish sitting and standing from a doffed prosthesis for several reasons. Temporarily doffing the socket during the day is a potentially effective accommodation strategy. Socket doffing facilitates limb fluid volume recovery that is retained during subsequent activity.28 Further, history of activity, independent of doffing, is relevant to current limb volume.35 Helping the prosthesis user keep track of when and for how long the prosthesis is doffed may improve socket fit and comfort. Potentially, algorithms could be developed to interpret data from the prosthetic activity measurement system and make recommendations to the prosthesis user to maintain fit. Communication to the user via a personal communication device such as a smartphone may facilitate communication of the need for temporary socket doffing.

Practitioners could use activity and prosthetic use information to diagnose prosthetic fit issues. Problems with prosthetic fit are common12 and have been reported as the primary cause of skin issues in people with lower-limb loss.36 Prosthetists typically rely on clinical evaluation of the residual limb and user's reports of their socket use to troubleshoot problems with socket fit. Accurate knowledge of types and periods of activity, socket doffing, and user's recall may facilitate a more thorough clinical assessment and help to mitigate and/or more rapidly address problems when they arise. Adding additional states, for example, leaning and reclining, may further facilitate clinical use of the data.

Accurate prosthetic activity measurement may also help to advance development of closed-loop controllers for active socket technologies, for example, adjustable sockets. A socket that increased in volume when the user sat and rested, and then returned to a smaller size when the person moved to standing would likely benefit limb fluid volume recovery and retention.28 Unlike step counters or pedometers used to gain clinical insight into the user's overall behavior, sensors used for automated prosthetic volume control must be extremely accurate. Activating a prosthesis adjustment based on a misread posture could have catastrophic outcomes. An adjustable socket, for example, that misread standing as sitting and enlarged the socket during standing might make the user unstable and injure them. This low tolerance for activity misclassification is an important aspect for activity-monitoring systems to be used in controller applications.

As the ability to measure and apply prosthetic activity data advances, prosthetics manufacturers should be able to incorporate activity monitoring systems into componentry and use the resultant data to effectively control socket adjustments. From a mechanical perspective, the monitoring system is well protected if contained within the prosthetic componentry thus should be durable and long lasting. Well-designed research investigations that inform on how to apply activity and doffing data to maintain and improve prosthetic fit should help move this innovation forward, enhancing quality of life of persons with lower-limb loss.


In this study, we developed and tested a novel prosthetic activity monitoring system for measuring prosthesis users' activities and socket use. This portable system is capable of identifying if a prosthesis user is wearing his/her socket, and determining if he/she is walking (or performing other types of movements), standing, or sitting over extended periods. Laboratory and field testing showed that the system accurately classified activities and socket use, when compared with researcher observation and user self-report. The ability to accurately measure activity and socket use, using a system like that shown here, has the potential to both inform clinical practice and advance development of novel prosthetic technologies.


1. Gauthier-Gagnon C, Grisé MC, Potvin D. Enabling factors related to prosthetic use by people with transtibial and transfemoral amputation. Arch Phys Med Rehabil 1999; 80(6): 706–713.
2. Pezzin LE, Dillingham TR, Mackenzie EJ, et al. Use and satisfaction with prosthetic limb devices and related services. Arch Phys Med Rehabil 2004; 85(5): 723–729.
3. Schaffalitzky E, Gallagher P, Maclachlan M, Wegener ST. Developing consensus on important factors associated with lower limb prosthetic prescription and use. Disabil Rehabil 2012; 34(24): 2085–2094.
4. Taylor SM, Kalbaugh CA, Cass AL, et al. “Successful outcome” after below-knee amputation: an objective definition and influence of clinical variables. Am Surg 2008; 74(7): 607–612 discussion 612–613.
5. Johannesson A, Larsson GU, Ramstrand N, et al. Outcomes of a standardized surgical and rehabilitation program in transtibial amputation for peripheral vascular disease: a prospective cohort study. Am J Phys Med Rehabil 2010; 89(4): 293–303.
6. Webster JB, Hakimi KN, Williams RM, et al. Prosthetic fitting, use, and satisfaction following lower-limb amputation: a prospective study. J Rehabil Res Dev 2012; 49(10): 1493–1504.
7. Hafner BJ, Sanders JE. Considerations for development of sensing and monitoring tools to facilitate treatment and care of persons with lower-limb loss: a review. J Rehabil Res Dev 2014; 51(1): 1–14.
8. Dudek NL, Khan OD, Lemaire ED, et al. Ambulation monitoring of transtibial amputation subjects with patient activity monitor versus pedometer. J Rehabil Res Dev 2008; 45(4): 577–585.
9. Schaffalitzky E, Gallagher P, Maclachlan M, Ryall N. Understanding the benefits of prosthetic prescription: exploring the experiences of practitioners and lower limb prosthetic users. Disabil Rehabil. 2011; 33(15-16): 1314–1323.
10. Pohjolainen T, Alaranta H, Kärkkäinen M. Prosthetic use and functional and social outcome following major lower limb amputation. Prosthet Orthot Int 1990; 14(2): 75–79.
11. Day HJ. The assessment and description of amputee activity. Prosthet Orthot Int 1981; 5(1): 23–28.
12. Gauthier-Gagnon C, Grise M, Potvin D. Predisposing factors related to prosthetic use by people with a transtibial and transfemoral amputation. J Prosthet Orthot 1998; 10(4): 99–109.
13. Stepien JM, Cavenett S, Taylor L, Crotty M. Activity levels among lower-limb amputees: self-report versus step activity monitor. Arch Phys Med Rehabil 2007; 88(7): 896–900.
14. Paulhus DL, Vazire S. The self-report method. In: Robins RW, Fraley RC, Krueger RF, (Eds.) Handbook of Research Methods in Personality Psychology. London: The Guilford Press; 2007: 224–239.
15. Gorin AA, Stone AA. Recall biases and cognitive errors in retrospective self-reports: a call for momentary assessments. In: Baum A, Revenson T, Singer J, (Eds.) Handbook of Health Psychology. Mahwah: Lawrence Erlbaum; 2001: 405–413.
16. Paulhus DP. Measurement and control of response bias. In: Robinson JP, Shaver PR, Wrightsman LS, (Eds.) Measures of Personality and Social Psychological Attitudes. San Diego: Academic; 1991: 17–59.
17. John OP, Robins RW. Accuracy and bias in self-perception: individual differences in self-enhancement and the role of narcissism. J Pers Soc Psychol 1994; 66(1): 206–219.
18. Moskowitz DS. Comparison of self-reports, reports by knowledgeable informants, and behavioral observation data. J Pers 1986; 54: 294–317.
19. Holden J, Fernie GR, Soto M. An assessment of a system to monitor the activity of patients in a rehabilitation programme. Prosthet Orthot Int 1979; 3(2): 99–102.
20. Holden J, Fernie G. Minimal walking levels for amputees living at home. Physiother Can 1983; 35(6): 317–320.
21. Holden JM, Fernie GR. Extent of artificial limb use following rehabilitation. J Orthop Res 1987; 5(4): 562–568.
22. Briseno GG, Smith JD. Pedometer accuracy in persons using lower-limb prostheses. J Prosthet Orthot 2014; 26(2): 87–92.
23. Coleman KL, Smith DG, Boone DA, et al. Step activity monitor: long-term, continuous recording of ambulatory function. J Rehabil Res Dev 1999; 36(1): 8–18.
24. Bussmann JBJ, Culhane KM, Horemans HLD, et al. Validity of the prosthetic activity monitor to assess the duration and spatio-temporal characteristics of prosthetic walking. IEEE Trans Neural Syst Rehabil Eng 2004; 12(4): 379–386.
25. Rosenbaum Chou TG, Webster JB, Shahrebani M, et al. Characterization of step count accuracy of Actigraph Activity Monitor in persons with lower limb amputation. J Prosthet Orthot 2009; 21(4): 208–214.
26. Hordacre B, Barr C, Crotty M. Use of an activity monitor and GPS device to assess community activity and participation in transtibial amputees. Sensors (Basel) 2014; 14(4): 5845–5859.
27. Jayaraman A, Deeny S, Eisenberg Y, et al. Global position sensing and step activity as outcome measures of community mobility and social interaction for an individual with a transfemoral amputation due to dysvascular disease. Phys Ther 2014; 94(3): 401–410.
28. Sanders JE, Hartley TL, Phillips RH, et al. Does temporary socket removal affect residual limb fluid volume of trans-tibial amputees? Prosthet Orthot Int. 2015 Feb 20 [Epub ahead of print].
29. Sanders JE, Cagle JC, Allyn KJ, et al. How do walking, standing, and resting influence trans-tibial amputee residual limb fluid volume? J Rehabil Res Dev 2014; 51(2): 201–212.
30. Pitei DL, Ison K, Edmonds ME, et al. Time-dependent behaviour of a force-sensitive resistor plantar pressure measurement insole. Proc Inst Mech Eng H 1996; 210(2): 121–125.
31. Buis AW, Convery P. Calibration problems encountered while monitoring stump/socket interface pressures with force sensing resistors: techniques adopted to minimise inaccuracies. Prosthet Orthot Int 1997; 21(3): 179–182.
32. Dabling G, Filatov A, Wheeler JW. Static and cyclic performance evaluation of sensors for human interface pressure measurement. Conf Proc IEEE Eng Med Biol Soc. 2012; 2012: 162–165.
33. Capela NA, Lemaire ED, Baddour N. Improving classification of sit, stand, and lie in a smartphone human activity recognition system. IEEE International Symposium on Medical Measurements and Applications 2015; 473–478.
34. Redfield MT, Cagle JC, Hafner BJ, Sanders JE. Classifying prosthetic use via accelerometry in persons with transtibial amputation. J Rehabil Res Dev 2013; 50(9): 1201–1212.
35. Sanders JE, Severance MR, Swartzendruber DL, et al. Influence of prior activity on residual limb volume and shape measured using plaster casting: results from individuals with transtibial limb loss. J Rehabil Res Dev 2013; 50(7): 1007–1016.
36. Dudek NL, Marks MB, Marshall SC. Skin problems in an amputee clinic. Am J Phys Med Rehabil 2006; 85(5): 424–429.

actigraphy; individual with amputation; artificial limb; movement; physical activity; posture; prosthetic socks; proximity sensor; accelerometer

© 2016 by the American Academy of Orthotists and Prosthetists.