This is an open-access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR mhealth and uhealth, is properly cited. The complete bibliographic information, a link to the original publication on http://mhealth.jmir.org/, as well as this copyright and license information must be included.
Early mobilization after surgery reduces the incidence of a wide range of complications. Wearable motion sensors measure movements over time and transmit this data wirelessly, which has the potential to monitor patient recovery and encourages patients to engage in their own rehabilitation.
We sought to determine the ability of off-the-shelf activity sensors to remotely monitor patient postoperative mobility.
Consecutive subjects were recruited under the Department of Neurosurgery at Columbia University. Patients were enrolled during physical therapy sessions. The total number of steps counted by the two blinded researchers was compared to the steps recorded on four activity sensors positioned at different body locations.
A total of 148 motion data points were generated. The start time, end time, and duration of each walking session were accurately recorded by the devices and were remotely available for the researchers to analyze. The sensor accuracy was significantly greater when placed over the ankles than over the hips (
We provide one of the first assessments of the accuracy and utility of widely available and wirelessly connected activity sensors in a postoperative patient population. Our results show that activity sensors are able to provide invaluable information about a patient’s mobility status and can transmit this data wirelessly, although there is a systematic underestimation bias in more debilitated patients.
Functional recovery refers to improvement in mobility and independence of activities of daily living (ADL) after hospitalization for surgery or acute illness. It is a widely used outcome measure, especially in postoperative patients and in those with neurological conditions. Mobilization is a cornerstone of rehabilitation therapy not only in the hospital and acute care settings, but also at home and in the community [
Early in-hospital mobilization reduces the risk of conditions related to prolonged bed rest—pulmonary embolism, atelectasis, pneumonia, decubitus ulcers—and is associated with improved survival, decreased length of hospitalization, and improved psychological well-being [
Commercially available activity sensors have tremendous potential to provide this data because recent technological advances have resulted in devices that are small, wearable, affordable, and able to relay their data wirelessly via patient mobile phones or wireless networks at home or in the hospital [
However, there is little consensus on how to use activity sensors to provide an
This study complies with the Declaration of Helsinki. This research protocol has been approved by the Columbia University Medical Center Institutional Review Board (IRB) (protocol number AAA-M6702).
A total of 27 consecutive subjects were prospectively recruited from a convenience sample of inpatients under the Department of Neurosurgery at Columbia University from November 2013 to July 2014. The patient subjects were a median of 3 days postoperative and were enrolled during their first or second inpatient physical therapy session provided by the Department of Rehabilitation and Regenerative Medicine. Of the 27 patients, 20 (74%) were postoperative spine patients (primarily laminoplasties, laminectomies, and microdiscectomies) and 7 (26%) had had craniotomies for tumors or vascular malformations. Additional patient characteristics are included in
A total of 10 healthy volunteer controls with no preexisting gait abnormalities were also included in the study as a comparison. Their characteristics can also be found in
The activity sensor used was the FitBit Zip (produced by FitBit, San Francisco, CA). The device records data such as the number of steps taken and the time stamps of when these steps occurred, and automatically syncs to mobile phones (and other devices) via Bluetooth. The recorded data is uploaded online to a user-friendly personalized account, and is easily searchable by date and time with a resolution of 15-minute time intervals. FitBit is considered one of the leaders in the market of wearable activity sensors, and at a cost of under US $60, the Zip model is far more affordable than comparable devices [
Two researchers (BT, EB) observed patients during each session with a physical therapist. Similar to methods used in previous studies [
Patients were seen for standard inpatient physical therapy sessions with a licensed physical therapist and two researchers for the study. The two researchers were blinded to the type of surgery and the postoperative day.
Four activity sensors were placed on each patient with one FitBit Zip at each of the following locations: on the right and left hips over the anterior superior iliac spine, as suggested by the manufacturer and by previous studies [
Placement of the activity sensor.
Each patient was asked to ambulate at a self-selected pace down a flat level course that was set up with the 0-meter, 4-meter, and 10-meter lines marked, and then further than 10 meters if deemed safe and appropriate by the physical therapist. If the patient walked further, this total distance was also recorded. Immediately after standing up from bed, the patients were asked to ambulate to the 0-meter starting line, which was always within 1 meter of the foot of their bed.
The gold standard number of steps was counted from the 0-meter to 4-meter line, 0-meter to 10-meter line, and the 0-meter line to the total distance if the patient ambulated further. A digital stopwatch with 1/10-second resolution was also used to record the time elapsed during the 0-meter to 4-meter and 0-meter to 10-meter intervals. The activity sensors recorded the number of steps taken for the total distance ambulated, and the reading from each of the four was documented. The 15-minute time interval corresponding to each physical therapy session was searched on the online account or the mobile app, and the number of FitBit-counted steps during that time was recorded; there were no overlapping intervals. Controls followed the same protocol except that they did not ambulate further than 20 meters.
The primary outcome was the accuracy of the sensors in terms of mobility assessment, which was assessed by comparing the total number of steps recorded by each tracker to the total number of steps counted by the researchers. We also verified the time accuracy of the sensor by comparing the recorded times—available on both the Web interface and mobile phone app—to those recorded by the researchers.
Recorded information included postoperative day, type of procedure (ie, spinal surgery or craniotomy), postoperative diagnosis, presence and degree of weakness on standard neurological exam, age, and gender (see
To assess the level of physical assistance that the patient required to safely ambulate, the 6-point, graded Functional Ambulation Category (FAC) (see
The total number of steps counted by the two researchers (gold standard) was compared to the steps recorded on the activity sensors using an intraclass correlation coefficient (ICC) [
To identify independent predictors of accuracy, a multivariate model was conducted that included the following variables: age, gait speed, step length, postoperative day (POD), and surgical group. All statistical analyses were performed with SPSS version 21.
There were a total of 148 motion data points generated from 37 individuals—27 patients and 10 controls—who met inclusion and exclusion criteria for enrollment in the study. Characteristics of the patient subjects are shown in
Characteristics of patient subjects (n=27).
Characteristic | Median (IQRa), n (%), or mean (SD) | |
Age in years, median (IQR) | 57 (44-68) | |
Gender (male), n (%) | 13 (48) | |
Walker used during session, n (%) | 14 (52) | |
Average gait velocity (m/s)b, mean (SD) | 0.260 (0.156-0.357) | |
Average step length (m)b, mean (SD) | 0.232 (0.169-0.278) | |
Total distance walked (m), median (IQR) | 50 (21-62) | |
Total steps ambulatedc, median (IQR) | 184 (127-255) | |
Postoperative day, median (IQR) | 3 (2-5) | |
|
|
|
|
0 | 1 (4) |
|
1 | 4 (15) |
|
2 | 9 (33) |
|
3 | 13 (48) |
|
|
|
|
Spine | 20 (74) |
|
Craniotomy | 7 (26) |
|
|
|
|
Right-sided only | 2 (7) |
|
Left-sided only | 5 (19) |
|
Both | 5 (19) |
aInterquartile range (IQR).
bCalculated during the 4- or 10-meter walk.
cAs determined by researchers using digital counting app.
dFAC is a measure of ambulation on a scale of 0 to 5; see
eDetermined by physician on standard neurological exam.
In the subject group, the ankle sensors were more accurate in counting steps than the hip sensors when compared to the gold standard number of steps counted by the observers (ICC .837 vs .326, respectively). This inaccuracy was due to undercounting, since the hip sensors significantly underestimated the number of steps by -81.4% on average compared to the -26.1% underestimate seen in the ankle sensors (
Intraclass correlation coefficient (ICC) and mean difference compared to the gold standard number of steps as counted by the researchers.
Sensor location and patient characteristic | ICC of number of steps (95% CI) | Mean difference, |
|
Hips—overall | .326 (-.214 to .684) | -81.4 (-93.2 to -69.5) | <.001 b |
Ankles—overall | .837 (.630 to .927) | -26.1 (-43.9 to -8.2) | .006 |
Ankles—without walker | .791 (.304 to .937) | -5.6 (-27.0 to +15.8) | .58 |
Ankles—with walker | .815 (.193 to .947) | -45.1 (-71.5 to -18.5) | .003 |
Ankles—with walker, with correction factor of +50% | .773 (.292 to .927) | -17.6 (-57.4 to +22.2) | .57 |
Ankles—step length >0.232 m | .973 (.902 to .993) | +3.5 (-13.5 to +20.6) | .65 |
Ankles—step length <0.232 m | .792 (.288 to .932) | -46.4 (-70.4 to -22.4) | .001 |
Ankles—step length <0.232 m, with correction factor of +50% | .734 (.238 to .907) | -19.6 (-55.5 to +16.3) | .29 |
Ankles—FACc=3 | .816 (.377 to .945) | +0.78 (-20.9 to +22.5) | .94 |
Ankles—FAC=0,1,2 | .801 (-.080 to .949) | -51.0 (-73.3 to -28.7) | <.001 |
Ankles—FAC=0,1,2, with correction factor of +50% | .803 (.387 to .937) | -26.5 (-59.9 to +7.02) | .15 |
aAnalysis of variance (ANOVA).
bValues in italics are statistically significant.
cFunctional Ambulation Category (FAC).
Mean differences in ankle and hip tracker recording in subjects versus controls (left); mean differences in ankle and hip tracker recordings in subjects with and without a rolling walker (right).
Functional Ambulation Category (FAC) in relation to ankle sensor mean differences from the gold standard.
To show that the undercounting bias could be adjusted in patients with an FAC< 3, step length <0.232 m, and those using a walker, we added 50% to the original step counts in these subgroups—the approximate underestimation in each case—and mean difference from the gold standard improved significantly (
Multivariate analysis of predictors of ankle sensor accuracy.
Variable |
|
Age | .81 |
Postoperative day (POD) | .55 |
Gait speed | .44 |
Step length | .03a |
Surgical group | .75 |
aValues in italics are statistically significant.
There were no significant differences observed between right-sided and left-sided trackers when comparing subjects with left-sided versus right-sided weakness. The total distance ambulated also did not significantly affect the accuracy of the ankle trackers (
Scatterplot of ankle sensor differences in relation to average step length.
We assessed the practicality and reliability of wearable, easy-to-use activity sensors in patients with limited mobility in the early postoperative period. Data from the rehabilitation sessions were remotely accessible by an online or mobile phone interface—an unprecedented technology that will provide health care professionals with the amount, duration, and timing of patient mobility at home and in the hospital. Although the activity sensors accurately tracked the time and duration of each session, in terms of step counting, our results highlight that ankle versus hip sensor placement, along with specific characteristics of patient mobility—use of an assistive device, step length, and FAC—affect the devices’ ability to accurately reflect patient functional recovery.
As mentioned earlier, mobilization generally improves survival and functional outcome in a wide variety of patients recovering from surgery, neurological illness, and cancer, but accurately tracking mobility, especially in the outpatient setting, has been challenging. A patient’s mobility can be graded by physical performance measures such as gait speed, which is a function of age, stature, and strength [
We observed an underestimation of step counts in the less mobile subject population, likely because these patients tended to have a lower FAC, shorter step length, and need for a walker. These common clinical characteristics made patients’ movements less pronounced, which were more difficult for the sensors to detect—especially those placed on the hips. On the other hand, the readings in the control group did not significantly differ from the gold standard, indicating that the sensor accuracy was greater than in the subject group. The higher accuracy in the control group, who had gait speeds in the normal range [
The tendency to underestimate in more slowly moving patients was also observed in a study where sensors undercounted by 19.1% to 32.1% when gait speeds were ≤0.800 m/s [
Within the subject group, sensor accuracy was strongest in patients who were more mobile, including those who ambulated without a walker or significant assistance from the physical therapist and, as mentioned, those with longer step lengths. This inverse relationship between amount of movement and degree of undercounting suggests that the sensors underestimate more as a patient’s mobility and functional status worsen. The relationship between decreased accuracy of other accelerometers and poorer functional status, such as in patients with congestive heart failure and chronic obstructive pulmonary disease (COPD), was noted in a systematic review on the validity of activity sensors in patients with chronic disease [
Some limitations of this study should be mentioned. Since this was a pilot study, our sample size was 27 subjects, although this is in the range of comparable previous studies [
In conclusion, activity sensors generate low-cost data on patient recovery in the hospital and in the home environment. We have shown that they are able to remotely monitor patient activity, although our results demonstrate that in order to develop research protocols which use activity sensors to reliably track patient mobility, the suitability of the sensors needs to be individually determined, as suggested by previous authors [
Since impaired ambulation and limited mobility may occur in a wide variety of other diseases [
Functional Ambulation Category (FAC).
activities of daily living
analysis of variance
chronic obstructive pulmonary disease
Functional Ambulation Category
Health Insurance Portability and Accountability Act
intraclass correlation coefficient
intensive care unit
interquartile range
Institutional Review Board
postoperative day
The source of funding for this study was the Department of Neurosurgery, Columbia University. Sponsors were not involved in the research, or in the writing or review of the manuscript.
None declared.