Original Paper
Abstract
Background: Algorithms estimating real-world digital mobility outcomes (DMOs) are increasingly validated in healthy adults and various disease cohorts. However, their accuracy and reliability in older adults after hip fracture, who often walk slowly for short durations, is underexplored.
Objective: This study examined DMO accuracy and reliability in a hip fracture cohort considering walking bout (WB) duration, physical function, days since surgery, and walking aid use.
Methods: In total, 19 community-dwelling participants were real-world monitored for 2.5 hours using a lower back wearable device and a reference system combining inertial modules, distance sensors, and pressure insoles. A total of 6 DMO estimates from 164 WBs from 58% (11/19) of the participants (aged 71-90 years; assessed 32-390 days after surgery; Short Physical Performance Battery [SPPB] scores of 3-12; gait speed range 0.39-1.34 m/s) were assessed against the reference system at the WB and participant level. We stratified by WB duration (all WBs, WBs of >10 seconds, WBs of 10-30 seconds, and WBs of >30 seconds) and lower versus higher SPPB scores and observed whether days since surgery and walking aid use affected DMO accuracy and reliability.
Results: Across WBs, walking speed and distance ranged from 0.25 to 1.29 m/s and from 1.7 to 436.5 m, respectively. Estimation of walking speed, cadence, stride duration, number of steps, and distance stratified by WB duration showed intraclass correlation coefficients (ICCs) ranging from 0.50 to 0.99 and mean relative errors (MREs) from –6.9% to 12.8%. Stride length estimation showed poor reliability, with ICCs ranging from 0.30 to 0.49 and MREs from 6.1% to 13.2%. Walking speed and distance ICCs in the higher–SPPB score group ranged from 0.85 to 0.99, and MREs ranged from –10.1% to –1.7%. In the lower–SPPB score group, walking speed and distance ICCs ranged from 0.17 to 0.99, and MREs ranged from 13.5% to 32.6%. There was no discernible effect of time since surgery or walking aid use.
Conclusions: In total, 5 accurate and reliable real-world DMOs were identified in older adults after hip fracture: walking speed, cadence, stride duration, number of steps, and distance. Accuracy and reliability of most DMOs improved when excluding WBs of <10 seconds and were higher for WBs of >30 seconds than for WBs of 10 to 30 seconds and for participants with higher physical function. DMOs capture daily gait as early as 1 month after surgery also in people using walking aids. However, as most WBs in this cohort were short, there was a trade-off between improving accuracy and reliability by excluding short WBs and losing a substantial amount of data. These results have important implications for establishing the clinical validity of DMOs and evaluating the effects of interventions on daily-life gait, thereby facilitating the design of optimal care pathways.
doi:10.2196/67792
Keywords
Introduction
Background
Hip fracture after a fall is one of the most serious injuries in older adults, with a high rate of morbidity and a 1-year mortality of 22% [,]. Survivors of hip fracture experience a substantial decline in quality of life and mobility []. Mobility, the ability to move or walk around freely and easily, is essential for functioning well, maintaining independence, and ensuring social and emotional well-being []. Between 40% and 60% of people after hip fracture do not recover their prefracture level of mobility and ability to perform instrumental activities of daily living [], and the incidence of nursing home admissions is high []. Hip fractures require surgery to restore mobility, but there is large variation in the average length of stay at the hospital across countries, ranging from 4 to 40 days after surgery [,]. The main goals of care in the first days are typically pain management and early mobilization, followed by rehabilitation targeting physical function and independent walking to prepare patients for returning to their daily life [,]. However, the different recovery trajectories and the factors determining mobility outcomes for various individuals are poorly understood. The ability to walk and regain mobility is an important indicator of recovery, making it a critical focus in the rehabilitation of patients after a hip fracture. However, getting patients back on their feet requires more than just a step in the right direction. Comprehensive mobility assessment is key to supporting effective recovery, providing valuable information about patients’ progress and informing rehabilitation strategies and optimal care pathways.
Assessing mobility requires objective, quantifiable metrics that accurately reflect a patient’s ability to move around efficiently and independently. While mobility can be described using broader measures such as upright time and number of sit-to-stand transitions [,], these primarily provide information about activity amount. In contrast, gait-related metrics provide information about the quality and efficiency of movement, reflecting musculoskeletal function, coordination, and stability. In addition, gait impairments are strong predictors of falls, disability, and overall health and inform about the status and progression of different health challenges [-], making walking a key aspect of mobility.
Gait after a hip fracture has mostly been assessed using patient-reported outcome measures that are prone to response bias or via standardized walking tests in clinical settings and laboratory assessments []. Traditional supervised gait assessments using sophisticated technology in laboratory settings, such as gold standard instrumented gait analysis, allow for accurate spatiotemporal measurements of gait in a controlled laboratory environment [,]. However, gait assessments under such conditions lack ecological validity for several reasons []. Laboratory assessments provide a snapshot in time, are limited by space and infrastructure, are prone to white coat effects, and often consist of isolated and structured tasks that do not necessarily reflect what people with and without mobility impairment do in their daily lives [-]. Importantly, they may also limit the inclusion of people from rural areas, potentially overlooking a key demographic []. In addition, in-laboratory assessments typically evaluate gait on even, uncluttered ground over short, straight distances only []. Therefore, we need more knowledge regarding gait in people’s real-world environments. In addition, there is a critical need to complement current short-term follow-up and infrequent clinical tests or assessments with more comprehensive and detailed knowledge about daily-life gait recovery throughout the first years after a hip fracture []. Hence, extended monitoring of gait is vital for understanding the long-term recovery process and optimizing intervention strategies [].
However, assessing accurate gait characteristics in the real world is challenging due to internal and external confounding factors. Real-world gait is highly variable and complex as it entails navigating different walking surfaces and uneven terrain and obstacles and consists of a wide variety of related activities such as turning, stopping, and starting []. Methods based on the use of a single inertial measurement unit (IMU) on the lower back are available to assess gait in the real world through the quantification of digital mobility outcomes (DMOs) []. However, until recently, algorithms to estimate DMOs lacked comprehensive and systematic validation [].
The Mobilise-D consortium [] has made significant strides to change this. Mobilise-D has used a rigorous and comprehensive validation process, including a multisensor wearable setup for real-world analysis [] and involving 1 healthy and 5 patient cohorts performing an extensive technical validation protocol [,]. This resulted in the identification and refinement of algorithms that enable robust gait sequence detection and subsequent estimation of key DMOs from a single inertial device worn on the lower back, such as initial foot contact, cadence, stride length, and walking speed. Previous Mobilise-D work has shown that these DMOs can be estimated accurately and reliably in the real world across a range of cohorts (healthy older adults and adults with Parkinson disease, multiple sclerosis, congestive heart failure, chronic obstructive pulmonary disease, and hip fracture), tasks, and contextual factors [,]. Analyses of all 6 Mobilise-D cohorts showed that the real-world algorithm performances were valid and accurate but more challenging in the case of short walking bouts (WBs; <10 seconds) and slower gait speeds (<0.5 m/s) [,], typical hallmarks of gait after a hip fracture. Analyses stratified by cohort indicated that the estimated walking speed and cadence in the hip fracture cohort showed moderate reliability and mean relative error (MRE) of 10% and –0.5%, respectively, and estimated stride length with poor reliability and an MRE of 11% []. However, these analyses did not consider hip fracture–specific factors that may affect the reliability and accuracy of the DMO estimates, which may be of critical relevance for the hip fracture cohort.
Objectives
Building on the previous Mobilise-D publications [,], this study investigated additional DMOs and delved deeper into potential factors that may affect the reliability and accuracy of the real-world DMO estimates in the hip fracture cohort. As short WBs are frequent after a hip fracture [] and walking distance can influence spatiotemporal parameters [], we investigated whether different WB duration categories affected the estimates. Furthermore, physical functioning is associated with variations in gait parameters and is highly compromised after a hip fracture [,], particularly in the early phases of recovery [], often necessitating the use of a walking aid [].
Hence, this study investigated the following research question: can the validated Mobilise-D real-world DMOs be accurately and reliably estimated in older adults after hip fracture considering WB duration, time since surgery, level of physical function, and potential walking aid use? To answer this question, we compared 6 DMOs estimated from a single wearable device against a reference system [,] and investigated whether WB duration, physical function, walking aid use, and days since surgery affected the accuracy and reliability of the DMO estimates.
Given earlier results, we expected that most real-world DMOs could be estimated accurately and reliably in the hip fracture cohort and that results would improve further when excluding WBs of short duration. In addition, as shorter WBs tend to have higher errors [,], we expected that DMO accuracy and reliability might be lower in older adults with low physical function than in those with better physical functioning. Moreover, as gait function is especially impaired early after a hip fracture, the accuracy and reliability of DMO estimates may be lower in the earlier stages of recovery than in later stages.
Methods
Study Design and Participants
This multicenter observational study used data from the technical validation study (TVS) of the Innovative Medicines Initiative 2 Joint Undertaking–funded Mobilise-D project [,,]. Data were collected between July 2020 and March 2022. The participants after hip fracture were recruited from the Robert Bosch Foundation for Medical Research (Germany) and Kiel University (Germany).
The participants were recruited within 13 months of surgical treatment (fixation or arthroplasty) for a low-energy fracture of the proximal femur (International Classification of Diseases, 10th Revision, diagnosis codes S72.0, S72.1, and S72.2), as diagnosed through x-rays of the hip and pelvis.
Participants had to be aged ≥65 years to be included. Participants were excluded if they were unable to walk 4 m independently with or without a walking aid; had a shoe size of <36 European size; had a Montreal Cognitive Assessment [] score of ≤15; or had an occurrence of any of the following within 3 months before inclusion: myocardial infarction, hospitalization for unstable angina, stroke, coronary artery bypass graft, percutaneous coronary intervention, or implantation of a cardiac resynchronization therapy device []. We aimed to recruit 20 participants. Due to challenges posed by the COVID-19 pandemic, recruitment was much slower than anticipated and further hampered by us not being allowed to recruit participants during their hospital stay. As a result, we needed to almost double the recruitment period to be able to enroll 19 participants. Of these 19 participants, 6 (32%) had to be excluded because of technological issues with the pressure insoles of the reference system. Data from the reference system from another 11% (2/19) of the participants had insufficient signal quality. These 2 participants were relatively young (aged 70 and 66 years), had a very low Short Physical Performance Battery (SPPB) score of 3, and were assessed 28 and 64 days after surgery. Data from the remaining 58% (11/19) of the participants were included in the final analysis.
Protocol and Equipment
Participants’ activities were monitored for 2.5 hours in a real-world setting within their preferred habitual environment (home, work, or community). The activities were unstructured, but to ensure sufficient variability in the collected data, participants were encouraged to complete several activities, such as rising from a chair and walking to another room; walking up and down a flight of stairs; walking to the kitchen and preparing a drink; walking outdoors (if possible for a minimum of 2 minutes); and, if walking outside, walking up and down an inclined path [].
Gait data were collected using a single wearable device (McRoberts DynaPort MM+; 100-Hz sampling frequency, –8 g to +8 g triaxial acceleration range and 1-mg resolution, and –2000 to +2000 degrees per second triaxial gyroscope range with a 70–millidegrees per second resolution) worn at the lower back with a Velcro belt. In addition, participants were equipped with a multisensor reference system consisting of IMUs, distance sensors, and pressure insoles named Inertial Module With Distance Sensors and Pressure Insoles (sampling frequency of 100 Hz) [,,,,] previously validated with excellent reliability in different cohorts, including hip fracture, across a complex set of different motor tests, including simulated daily activities []. Specifically, 2 magneto-IMUs were fixed to the instep using clips. A third IMU was attached to the lower back using Velcro. Asymmetrically positioned distance sensors were fixed above the ankles using Velcro, and 2 pressure insoles were inserted into the shoes. To ensure synchronization, time stamps of the reference system and the single wearable device were aligned (–10 to +10 ms).
Evaluation of DMOs
The evaluation of the DMOs was based on previous Mobilise-D work that selected the top-ranked algorithms to detect gait sequences and estimate initial contact events, stride length, and cadence within identified gait sequences []. The best-performing cadence and stride length algorithms for hip fracture were then used to estimate walking speed []. The validation of this processing pipeline was based on a minimum 95% CI intraclass correlation coefficient (ICC) threshold for performance metrics (ie, sensitivity, positive predictive value, and accuracy) of at least 0.7 and a relative error of <20%, as described in the studies by Micó-Amigo et al [] and Kirk et al []. All algorithms are available in the MobGap Python library (Python Software Foundation) [].
DMOs were evaluated at a WB level. A WB was defined as a continuous walking sequence comprising a minimum of 2 consecutive strides of both feet []. WBs were separated by breaks of >3 seconds, and for a stride to be included, it required a duration of 0.2 to 3 seconds and a minimum length of 0.15 m []. These criteria were applied to generate WBs for both the single wearable device and the reference system by initially filtering the identified strides according to the stride level definition and then assembling them into final WBs by identifying breaks in the stride sequence. Final DMOs for both systems were calculated as the average value over all strides within a WB. For a rigorous comparison of DMOs at a WB level, it was essential to focus on WBs concurrently detected by both systems using a true-positive analysis approach. Accordingly, we considered WBs with a time overlap exceeding 80% of their duration as true positives, as detailed in the work by Kirk et al [].
For each WB, 6 gait characteristics were obtained from the single wearable device and the reference system: cadence (steps per minute; the number of steps taken per minute), stride length (meters; the length of 2 consecutive steps), number of steps, stride duration (seconds) [], walking speed (m/s) [], and distance (meters). The walked distance was calculated by multiplying 2 validated DMOs []: average walking speed × WB duration.
Variables
Participant Characteristics
Age, height, weight, sex, cognitive function (Montreal Cognitive Assessment), fracture type, and surgical implant were collected for all participants. Pain while walking was assessed using a visual analogue scale (from 0 to 100, where 0 is no pain and 100 is worst pain imaginable).
WB Duration
Given the larger single wearable device DMO errors observed in shorter WBs [,], we assessed whether the accuracy and reliability of DMO estimates differed when excluding WB durations of <10 seconds [,]. In addition, we divided WBs of >10 seconds into 2 subcategories: WBs of 10 to 30 seconds and WBs of >30 seconds. Thus, we analyzed the DMOs in 4 categories of WB duration: all WBs combined, WBs of >10 seconds, WBs of 10 to 30 seconds, and WBs of >30 seconds. This was performed at both a WB level and a participant level.
Physical Function
The participants’ physical function was assessed using the SPPB 7 days before or after the 2.5-hour activity monitoring. The SPPB consists of a static balance test, a 5-time chair rise test, and a 4-m walk test at a comfortable gait speed. The total SPPB score ranges from 0 to 12, where higher scores indicate better mobility capacity and a score of <8 points indicates impaired activity of daily living functions []. Due to the larger DMO errors in cohorts with more impaired gait and slower walking speeds [,], we divided participants into 2 physical function groups based on the 8-point SPPB threshold: a lower–SPPB score group consisting of participants with a total score between 0 and 7 and a higher–SPPB score group consisting of participants with a total score between 8 and 12. This threshold resulted in 2 groups of 45% (5/11) and 55% (6/11) of the participants, respectively.
Time Since Surgery and Walking Aid Use
Physical function and mobility typically improve gradually in the first year after a hip fracture [,,]. We collected the number of days between the surgery and the 2.5-hour free-living testing date, aggregated participant DMOs, and sorted them by number of days since surgery. We also collected information on walking aid use (yes or no and type of walking aid).
Statistical Analyses
DMOs were evaluated at a WB level to quantify DMO estimation errors (accuracy) and reliability using the validation metrics detailed in previous Mobilise-D work [,]. For accuracy, absolute agreement was assessed by calculating the mean error, mean absolute error, and precision (limits of agreements) [] between DMO estimates from the single wearable device and the reference system. MREs and mean absolute relative errors (MAREs) were calculated by dividing the (absolute) errors per WB by the corresponding estimates from the reference system, expressed in percentage, as shown in the following formulas (SWD refers to the single wearable device, and RS refers to the reference system):
- Mean error = (1 / n) × Σ from i=1 to n ([DMO_SWDi– DMO_RSi])
- MRE = (1 / n) × Σ from i=1 to n ([(DMO_SWDi– DMO_RSi)/DMO_RSi] × 100)
- Mean absolute error = (1 / n) × Σ from i=1 to n (|DMO_SWDi – DMO_RSi|)
- MARE = (1 / n) × Σ from i=1 to n ([|DMO_SWDi – DMO_RSi|/DMO_RSi] × 100)
For reliability, the ICC(2,1) [] was computed to evaluate how closely each of the DMOs of the 2 systems were related. ICC values of <0.5 were considered poor, ICC values between 0.5 and 0.75 were considered moderate, ICC values between 0.75 and 0.9 were considered good, and ICC values of >0.90 were considered excellent [].
For analyses stratified by WB duration, all 6 DMOs were included. In analyses stratified by the 2 SPPB score groups, we included the 2 most relevant DMOs for clinicians and patients: walking speed [] and distance [].
Walking speed and distance were also evaluated at a participant level. Histograms, Q-Q plots, and the Anderson-Darling tests were used for both variables to assess normality. Given that the DMOs were not normally distributed and the relatively low number of participants, median DMO values were computed across all WBs for each participant for the single wearable device and the reference system. The median DMO values from both systems were visualized in bar graphs with IQRs.
Data preparation and visualization were conducted in MATLAB (R2022a; MathWorks), whereas statistical analyses were conducted in Stata (version 18.0; StataCorp).
Ethical Considerations
The protocol was approved by the Faculty of Medicine of the University of Tübingen (647/2019BO2) and the Faculty of Medicine of Kiel University (D540/19). All participants gave their written informed consent. The analysis was conducted using pseudonymized data. No financial compensation was provided to participants in this study.
Results
Participant Characteristics
The 11 participants included in the analyses (n=6, 55% men and n=5, 45% women) were community dwelling, and their characteristics are outlined in 
Table 1. Notably, the study sample included older adults after hip fracture assessed between 32 and 390 days after surgery. Across the 164 WBs, walking speed and distance ranged from 0.25 to 1.29 m/s and from 1.7 to 436.5 m, respectively, as measured using the reference system, and 22% (36/164) of these WBs were of <0.5 m/s. Of the 11 included participants, 9 (82%) had ≥2 WBs of <0.5 m/s, and 2 (18%) used a walking aid during the observation period, both using a single cane or crutch. Of the 11 participants, 5 (45%) had trochanteric fractures and were treated with intramedullary nails, whereas 6 (55%) had cervical fractures, of whom 4 (67%) received a hemiprosthesis and 2 (33%) received a total prosthesis.
| Participant | Median (IQR) | ||||||||||||
| 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | |||
| Days since surgery | 32 | 39 | 60 | 114 | 141 | 179 | 193 | 200 | 244 | 369 | 390 | 179 (60-244) | |
| Age (y) | 71 | 76 | 80 | 84 | 79 | 83 | 87 | 72 | 71 | 83 | 90 | 80 (72-84) | |
| Height (cm) | 180 | 182 | 169 | 180 | 158 | 159 | 174 | 174 | 158 | 165 | 174 | 173.5 (159-180) | |
| Weight (kg) | 78 | 85 | 66 | 69 | 54 | 44 | 71 | 95 | 53 | 66 | 96 | 68.5 (54-85) | |
| MoCAa score | 26 | 30 | 27 | 28 | 30 | 18 | 25 | 30 | 22 | 21 | 19 | 26 (21-30) | |
| Walking pain (0-100) | 5 | 3 | 2 | 4 | 0 | 30 | 19 | 4 | 52 | 2 | 1 | 4 (2-19) | |
| SPPBb score | 9 | 4 | 3 | 9 | 10 | 7 | 11 | 10 | 6 | 12 | 4 | 9 (4-10) | |
| Gait speed (4MWTc; m/s) | 0.79 | 0.39 | 0.41 | 0.89 | 0.80 | 0.83 | 1.34 | 1.05 | 0.75 | 1.08 | 0.62 | 0.80 (0.69-0.97) | |
| WBsd of <0.5 m/s, n (%)e | 2 (14)f | 0 (0) | 10 (83)g | 3 (14)h | 2 (7)i | 3 (25)g | 2 (14)f | 3 (18)j | 8 (23)k | 0 (0) | 3 (30)l | —m | |
| Sex | Male | Male | Female | Male | Female | Female | Male | Male | Female | Female | Male | — | |
| Fracture type | Cn | C | C | C | To | T | T | C | T | T | C | — | |
| Implant | Hp | TPq | TP | H | Nail | Nail | Nail | H | Nail | Nail | H | — | |
aMoCA: Montreal Cognitive Assessment.
bSPPB: Short Physical Performance Battery.
c4MWT: 4-m walk test.
dWB: walking bout.
eNumber of matched WBs with a speed of <0.5 m/s as measured using the reference system.
fn=14.
gn=12.
hn=21.
in=27.
jn=17.
kn=35.
ln=10.
mNot applicable.
nC: cervical.
oT: trochanteric.
pH: hemiprosthesis.
qTP: total prosthesis.
WB Duration
In total, 164 WBs met our true-positive approach [], of which 65 (39.6%) were of <10 seconds, 60 (36.6%) were between 10 and 30 seconds, and 39 (23.8%) were of >30 seconds (Table 2). Estimations of the number of steps and distance demonstrated good to excellent reliability in all WB duration categories, with ICCs exceeding 0.83, MREs ranging from –6.87% to 5.34%, and MAREs ranging from 4.96% to 19.79% (Table 3). Walking speed and stride duration showed moderate reliability for all WBs combined and for WBs of >10 seconds, with ICCs ranging from 0.64 to 0.71, MREs ranging from –5.68% to 10.35%, and MAREs ranging from 7.51% to 19.75%. Walking speed showed good reliability and less error in longer WBs of >30 seconds (ICC=0.80; MRE=5.09%; MARE=14.5%) compared to shorter WBs of 10 to 30 seconds (ICC=0.50; MRE=12.78%; MARE=20.81%). Similarly, stride duration showed good reliability and less error in WBs of >30 seconds (ICC=0.77; MRE=–4.98%; MARE=5.3%) compared to WBs of 10 to 30 seconds (ICC=0.54; MRE=–6.13%; MARE=8.94%). Cadence showed moderate to excellent reliability in all WB categories, with ICCs ranging from 0.74 to 0.98, low MREs ranging from –0.56% to –0.34%, and MAREs ranging from 1.75% to 5.62%. In contrast, stride length showed poor reliability, with ICCs ranging from 0.30 to 0.49, an overestimation in MRE varying from 6.05% to 13.23%, and MAREs ranging from 15.46% to 20.61%.
| WB durationa | Number of WBs, n (%) | True-positive WBs per SPPB score group, n (%) | True-positive WBs by days since surgery, n (%) | ||||||||||||
| Lower | Higher | 32 days | 39 days | 60 days | 114 days | 141 days | 179 days | 193 days | 200 days | 244 days | 369 days | 390 days | |||
| All | 164 (100) | 70 (42.7) | 94 (57.3) | 14 (8.5) | 1 (0.6) | 12 (7.3) | 21 (12.8) | 27 (16.5) | 12 (7.3) | 14 (8.5) | 17 (10.4) | 35 (21.3) | 1 (0.6) | 10 (6.1) | |
| >10 s | 99 (60.4) | 42 (25.6) | 57 (34.8) | 10 (6.1) | 1 (0.6) | 8 (4.9) | 15 (9.1) | 15 (9.1) | 9 (5.5) | 8 (4.9) | 8 (4.9) | 14 (8.5) | 1 (0.6) | 10 (6.1) | |
| 10-30 s | 60 (36.6) | 27 (16.5) | 33 (20.1) | 5 (3) | 0 (0) | 4 (2.4) | 8 (4.9) | 13 (7.9) | 7 (4.3) | 5 (3) | 1 (0.6) | 11 (6.7) | 1 (0.6) | 5 (3) | |
| >30 s | 39 (23.8) | 15 (9.1) | 24 (14.6) | 5 (3) | 1 (0.6) | 4 (2.4) | 7 (4.3) | 2 (1.2) | 2 (1.2) | 3 (1.8) | 7 (4.3) | 3 (1.8) | 0 (0) | 5 (3) | |
aAs described in the Methods section, WBs of <10 seconds (n=65) were included in the all WBs category and were not reported separately.
| WB duration | RS, mean (5% quantile, 95% quantile) | SWD, mean (5% quantile, 95% quantile) | Error, mean (LoA) | MRE (%), mean (LoA) | MAE, mean (5% quantile, 95% quantile) | MARE (%), mean (5% quantile, 95% quantile) | ICC(2,1), mean (5% quantile, 95% quantile) | ||||||||
| Walking speed (m/s) | |||||||||||||||
| Alla | 0.65 (0.37, 1.03) | 0.69 (0.45, 0.96) | 0.03 (–0.25 to 0.31) | 10.35 (–50.12 to 70.82) | 0.11 (0.01 to 0.33) | 19.75 (1.80, 66.32) | 0.67 (0.58, 0.73)b | ||||||||
| >10 s | 0.68 (0.39, 1.05) | 0.71 (0.44, 1.02) | 0.03 (–0.24 to 0.30) | 9.75 (–53.53 to 73.03) | 0.10 (0.01, 0.33) | 18.32 (1.27, 66.32) | 0.71 (0.62, 0.79)b | ||||||||
| 10-30 s | 0.61 (0.38, 0.98) | 0.66 (0.43, 0.89) | 0.05 (–0.23 to 0.32) | 12.78 (–56.61 to 82.16) | 0.11 (0.01, 0.32) | 20.81 (1.74, 65.57) | 0.50 (0.32, 0.64)b | ||||||||
| >30 s | 0.77 (0.46, 1.19) | 0.78 (0.45, 1.10) | 0.01 (–0.25 to 0.26) | 5.09 (–46.99 to 57.16) | 0.09 (0.01, 0.39) | 14.50 (1.11, 79.76) | 0.80 (0.69, 0.88)c | ||||||||
| Cadence (steps per min) | |||||||||||||||
| Alla | 89.81 (64.25, 114.21) | 88.52 (65.76, 110.95) | –1.29 (–21.79 to 19.21) | –0.56 (–19.50 to 18.39) | 5.47 (0.15, 17.50) | 5.62 (0.16, 19.43) | 0.74 (0.68, 0.80)b | ||||||||
| >10 s | 88.28 (62.32, 111.14) | 87.42 (62.90, 104.36) | –0.86 (–14.73 to 13.02) | –0.39 (–13.81 to 13.04) | 3.58 (0.08, 12.36) | 3.84 (0.08, 13.61) | 0.86 (0.80, 0.89)c | ||||||||
| 10-30 s | 88.60 (64.83, 110.10) | 87.52 (64.83, 102.66) | –1.08 (–18.31 to 16.15) | –0.42 (–17.01 to 16.17) | 4.89 (0.16, 15.30) | 5.20 (0.20, 15.89) | 0.75 (0.64, 0.83)c | ||||||||
| >30 s | 87.77 (58.29, 111.48) | 87.26 (59.51, 108.71) | –0.52 (–6.42 to 5.39) | –0.34 (–6.45 to 5.77) | 1.57 (0.02, 10.59) | 1.75 (0.03, 10.36) | 0.98 (0.96, 0.99)d | ||||||||
| Stride length (m) | |||||||||||||||
| Alla | 0.88 (0.54, 1.24) | 0.93 (0.72, 1.19) | 0.05 (–0.33 to 0.43) | 11.42 (–49.51 to 72.35) | 0.15 (0.01, 0.41) | 20.40 (1.38, 64.86) | 0.45 (0.34, 0.55)e | ||||||||
| >10 s | 0.91 (0.57, 1.29) | 0.96 (0.72, 1.26) | 0.05 (–0.33 to 0.42) | 10.40 (–49.91 to 70.71) | 0.14 (0.01, 0.43) | 18.58 (1.03, 64.86) | 0.49 (0.36, 0.61)e | ||||||||
| 10-30 s | 0.83 (0.56, 1.23) | 0.90 (0.71, 1.11) | 0.07 (–0.30 to 0.43) | 13.23 (–47.34 to 73.80) | 0.14 (0.01, 0.42) | 20.61 (1.28, 63.57) | 0.41 (0.21, 0.57)e | ||||||||
| >30 s | 1.04 (0.60, 1.38) | 1.06 (0.89, 1.31) | 0.02 (–0.38 to 0.42) | 6.05 (–53.63 to 65.73) | 0.13 (0.01, 0.72) | 15.46 (0.80, 119.91) | 0.30 (0.04, 0.53)e | ||||||||
| Number of steps | |||||||||||||||
| All | 49.77 (6.00, 266.00) | 47.00 (5.00, 233.00) | –2.77 (–18.78 to 13.23) | –5.62 (–31.00 to 19.76) | 3.41 (0.00, 10.00) | 10.39 (0.00, 30.00) | 0.99 (0.99, 1.00)d | ||||||||
| >10 s | 76.32 (13.00, 317.00) | 72.10 (13.00, 306.00) | –4.22 (–24.21 to 15.77) | –5.80 (–25.88 to 14.28) | 4.87 (0.00, 21.00) | 8.61 (0.00, 25.81) | 0.99 (0.98, 1.01)d | ||||||||
| 10-30 s | 22.42 (13.00, 38.50) | 20.62 (12.50, 37.00) | –1.80 (–7.52 to 3.92) | –6.87 (–30.79 to 17.06) | 2.50 (0.00, 8.00) | 10.98 (0.00, 29.89) | 0.90 (0.79, 0.95)c | ||||||||
| >30 s | 159.26 (40.00, 479.00) | 151.31 (36.00, 423.00) | –7.95 (–37.77 to 21.87) | –4.17 (–15.72 to 7.39) | 8.51 (0.00, 56.00) | 4.96 (0.00, 16.86) | 0.99 (0.98, 1.00)d | ||||||||
| Stride duration (s) | |||||||||||||||
| All | 1.41 (1.06, 1.88) | 1.33 (1.03, 1.65) | –0.08 (–0.40 to 0.24) | –4.65 (–25.80 to 16.50) | 0.12 (0.00, 0.41) | 8.15 (0.15, 26.82) | 0.70 (0.55, 0.79)b | ||||||||
| >10 s | 1.43 (1.09, 1.95) | 1.33 (1.08, 1.65) | –0.10 (–0.41 to 0.22) | –5.68 (–24.00 to 12.64) | 0.12 (0.00, 0.42) | 7.51 (0.16, 24.25) | 0.64 (0.41, 0.77)b | ||||||||
| 10-30 s | 1.42 (1.10, 1.91) | 1.32 (1.09, 1.63) | –0.10 (–0.45 to 0.24) | –6.13 (–27.08 to 14.82) | 0.14 (0.01, 0.42) | 8.94 (0.52, 25.02) | 0.54 (0.29, 0.70)b | ||||||||
| >30 s | 1.43 (1.09, 2.07) | 1.34 (1.08, 1.67) | –0.09 (–0.35 to 0.18) | –4.98 (–18.43 to 8.47) | 0.09 (0.00, 0.42) | 5.30 (0.04, 20.19) | 0.77 (0.54, 0.88)c | ||||||||
| Distance (m) | |||||||||||||||
| All | 25.75 (2.36, 125.62) | 24.22 (2.44, 119.63) | –1.53 (–22.16 to 19.11) | 4.91 (–53.08 to 62.90) | 3.75 (0.06, 15.48) | 19.79 (0.92, 65.29) | 0.98 (0.97, 0.98)d | ||||||||
| >10 s | 40.08 (5.40, 203.72) | 37.50 (5.04, 168.91) | –2.58 (–28.93 to 23.77) | 3.14 (–54.18 to 60.46) | 5.66 (0.07, 26.48) | 17.72 (0.86, 58.11) | 0.98 (0.97, 0.98)d | ||||||||
| 10-30 s | 9.62 (4.84, 19.38) | 9.87 (4.69, 19.41) | 0.25 (–4.61 to 5.12) | 5.34 (–56.44 to 67.13) | 1.71 (0.05, 5.55) | 19.53 (0.86, 57.10) | 0.83 (0.74, 0.89)c | ||||||||
| >30 s | 86.93 (15.61, 253.43) | 80.00 (20.80, 233.73) | –6.93 (–47.31 to 33.44) | –0.25 (–49.97 to 49.47) | 11.74 (0.32, 59.36) | 14.93 (0.78, 79.93) | 0.96 (0.94, 0.98)d | ||||||||
aAs reported in the work by Kirk et al [].
bModerate ICC level (0.50-0.75).
cGood ICC level (0.75-0.90).
dExcellent ICC level (>0.90).
ePoor ICC level (<0.50).
Physical Function
The reliability of the walking distance estimates was excellent for both SPPB groups and all WB duration categories (ICCs≥0.94) except for WBs of 10 to 30 seconds in the lower–SPPB score group (ICC=0.63), which showed moderate reliability (Table 4). Distance was overestimated in the lower SPPB–score group for all WBs (MRE=20.05%; MARE=26.83%) and WBs of >10 seconds (MRE=20.34%; MARE=26.3%), and the error was lower for longer WBs of >30 seconds (MRE=13.52%; MARE=20.43%) than for shorter WBs of 10 to 30 seconds (MRE=24.13%; MARE=29.56%). Conversely, the distance was underestimated in the higher–SPPB score group for all WBs (MRE=–6.37%; MARE=14.55%) and WBs of >10 seconds (MRE=–9.53%; MARE=11.39%), and the error was lower for WBs of >30 seconds (MRE=–8.85%; MARE=11.49%) than for WBs of 10 to 30 seconds (MRE=–10.03%; MARE=11.32%).
In the higher–SPPB score group, walking speed reliability was good for all WB duration categories (ICC≥0.85), with a slightly underestimated walking speed and MRE ranging from –1.73% to –3.59% (MARE reaching 8.62% to 11.15%). Conversely, the lower–SPPB score group showed poor walking speed reliability and higher errors for all WBs combined, WBs of >10 seconds, and WBs of 10 to 30 seconds, with ICCs ranging from 0.17 to 0.44, MREs reaching 26.58% to 32.56%, and MAREs reaching 31.22% to 35.71%. Notably, the lower–SPPB score group had a moderate walking speed reliability for WBs of >30 seconds (ICC=0.66), with an MRE of 18.96% and MARE of 23.14%.
| WBa duration and SPPB group (score) | RS, mean (5% quantile, 95% quantile) | SWD, mean (5% quantile, 95% quantile) | Error, mean (LoA) | MRE (%), mean (LoA) | MAE, mean (5% quantile, 95% quantile) | MARE (%), mean (5% quantile, 95% quantile) | ICC(2,1), mean (5% quantile, 95% quantile) | |||||||||
| Walking speed (m/s) | ||||||||||||||||
| Lower (0-7) | ||||||||||||||||
| All | 0.59 (0.34, 0.91) | 0.70 (0.43, 0.94) | 0.12 (–0.19 to 0.42) | 26.58 (–49.26 to 102.41) | 0.15 (0.02, 0.39) | 31.30 (2.76, 95.98) | 0.44 (0.14, 0.63)b | |||||||||
| >10 s | 0.56 (0.34, 0.91) | 0.68 (0.43, 0.92) | 0.12 (–0.18 to 0.41) | 27.70 (–54.84 to 110.25) | 0.14 (0.02, 0.39) | 31.22 (2.76, 95.98) | 0.42 (0.01, 0.64)b | |||||||||
| 10-30 s | 0.53 (0.32, 0.66) | 0.67 (0.42, 0.87) | 0.14 (–0.16 to 0.43) | 32.56 (–54.61 to 119.72) | 0.16 (0.02, 0.35) | 35.71 (4.02, 95.98) | 0.17 (0.00, 0.42)b | |||||||||
| >30 s | 0.63 (0.34, 1.04) | 0.71 (0.44, 1.05) | 0.08 (–0.21 to 0.38) | 18.96 (–54.21 to 92.13) | 0.11 (0.01, 0.44) | 23.14 (0.77, 127.33) | 0.66 (0.33, 0.85)c | |||||||||
| Higher (8-12) | ||||||||||||||||
| All | 0.70 (0.39, 1.05) | 0.68 (0.48, 0.97) | –0.03 (–0.21 to 0.15) | –1.73 (–30.26 to 26.80) | 0.08 (0.01, 0.19) | 11.15 (1.24, 29.51) | 0.86 (0.80, 0.90)d | |||||||||
| >10 s | 0.76 (0.49, 1.19) | 0.72 (0.47, 1.08) | –0.04 (–0.20 to 0.12) | –3.49 (–23.62 to 16.65) | 0.07 (0.01, 0.19) | 8.82 (1.24, 22.25) | 0.89 (0.80, 0.93)d | |||||||||
| 10-30 s | 0.68 (0.44, 1.00) | 0.65 (0.46, 0.89) | –0.03 (–0.18 to 0.12) | –3.41 (–23.29 to 16.47) | 0.06 (0.01, 0.16) | 8.62 (1.27, 22.25) | 0.85 (0.73, 0.91)d | |||||||||
| >30 s | 0.86 (0.63, 1.19) | 0.81 (0.65, 1.10) | –0.05 (–0.22 to 0.13) | –3.59 (–24.49 to 17.32) | 0.08 (0.01, 0.21) | 9.09 (1.24, 18.61) | 0.87 (0.72, 0.93)d | |||||||||
| Distance (m) | ||||||||||||||||
| Lower (0-7) | ||||||||||||||||
| All | 22.80 (1.98, 125.62) | 24.31 (2.30, 130.75) | 1.50 (–9.72 to 12.72) | 20.05 (–49.57 to 89.67) | 3.10 (0.07, 15.48) | 26.83 (0.93, 93.85) | 0.99 (0.99, 0.99)e | |||||||||
| >10 s | 35.68 (5.55, 125.63) | 37.85 (5.75, 145.84) | 2.16 (–12.14 to 16.46) | 20.34 (–52.06 to 92.75) | 4.57 (0.15, 20.23) | 26.30 (0.86, 90.93) | 0.99 (0.98, 0.99)e | |||||||||
| 10-30 s | 8.79 (4.31, 16.80) | 10.66 (5.04, 16.80) | 1.87 (–3.29 to 7.02) | 24.13 (–50.53 to 98.80) | 2.31 (0.07, 6.03) | 29.56 (0.86, 90.93) | 0.63 (0.29, 0.81)c | |||||||||
| >30 s | 84.09 (14.35, 237.84) | 86.79 (27.33, 233.73) | 2.70 (–20.70 to 26.10) | 13.52 (–55.05 to 82.09) | 8.65 (0.32, 26.48) | 20.43 (0.33, 107.91) | 0.99 (0.96, 0.99)e | |||||||||
| Higher (8-12) | ||||||||||||||||
| All | 27.94 (2.56, 139.68) | 24.16 (2.49, 119.63) | –3.78 (–28.41 to 20.85) | –6.37 (–40.14 to 27.41) | 4.24 (0.04, 20.05) | 14.55 (0.87, 36.45) | 0.97 (0.96, 0.98)e | |||||||||
| >10 s | 43.32 (5.40, 203.72) | 37.25 (4.92, 168.91) | –6.07 (–36.94 to 24.80) | –9.53 (–30.31 to 11.24) | 6.46 (0.06, 34.81) | 11.39 (0.87, 25.97) | 0.97 (0.95, 0.98)e | |||||||||
| 10-30 s | 10.30 (5.29, 21.23) | 9.23 (4.46, 22.02) | –1.07 (–3.66 to 1.53) | –10.03 (–30.75 to 10.69) | 1.21 (0.04, 4.03) | 11.32 (0.87, 28.68) | 0.94 (0.80, 0.97)e | |||||||||
| >30 s | 88.71 (21.79, 253.43) | 75.76 (20.80, 194.07) | –12.95 (–57.48 to 31.58) | –8.85 (–30.07 to 12.37) | 13.67 (0.56, 59.36) | 11.49 (0.92, 23.42) | 0.96 (0.89, 0.98)e | |||||||||
aWB: walking bout.
bPoor ICC level (<0.50).
cModerate ICC level (0.50-0.75).
dGood ICC level (0.75-0.90).
eExcellent ICC level (>0.90).
Time Since Surgery and Walking Aid Use
To explore whether time since surgery influenced DMO accuracy, we plotted the participants’ walking speed and distance estimates from both systems sorted by the number of days since surgery, which ranged from 32 to 390. As shown in 
Figures 1 and , there was no evident pattern in DMO accuracy across the number of days since surgery for any of the WB duration categories. Similarly, no clear deviation was found for the 18% (2/11) of participants who used a walking aid.
Discussion
Principal Findings
This study is the first to investigate critical hip fracture–specific factors that potentially influence the accuracy and reliability of validated DMOs, building on the rigorous methods and data from the Mobilise-D TVS [,,]. We compared 6 real-world DMO estimates from a single wearable device against a validated multisensor reference system [] and investigated factors that may affect the accuracy and reliability of validated DMOs in older adults after a hip fracture: WB duration, physical function, time since surgery, and walking aid use. Overall, 5 of the DMOs (walking speed, cadence, stride duration, number of steps, and distance) showed moderate to excellent accuracy and reliability across WB durations. The sixth DMO (stride length) showed poor accuracy and reliability in this hip fracture cohort. Furthermore, walking speed and distance were more accurate and reliable in participants with better physical function. We did not observe a discernible effect of time since surgery or walking aid use on the walking speed and distance estimates. These results indicate that real-world gait assessment is feasible in a hip fracture cohort and that gait characteristics can be estimated accurately and reliably from as early as 1 month after hip fracture surgery.
Impact of WB Duration
Previous Mobilise-D studies that validated DMOs across 6 cohorts showed generally lower algorithm performances when including very short WBs (<10 seconds) and for cohorts walking slowly and relying more on walking aids, such as many older adults after hip fracture [,]. As these previous studies did not perform cohort-specific analyses stratified by WB duration, this study further examined the hip fracture cohort following the true-positive evaluation of DMO accuracy and reliability at a WB level as outlined in the work by Kirk et al [] and Micó-Amigo et al []. As expected, excluding shorter WBs increased the accuracy and reliability of walking speed, cadence, number of steps, stride duration, and distance, with WBs of >30 seconds having the highest accuracy and reliability. These lower errors are likely due to longer WBs capturing more continuous, consistent gait []. In addition, longer WB durations likely correspond to outdoor walks, which are generally not as slow and intermittent as indoor walking [,]. For short-duration WBs, contextual factors and turns might be more prominent [], likely leading to higher DMO errors. However, as patients after a hip fracture typically walk in short bouts, excluding these would result in the loss of a potentially large amount of data, jeopardizing the representativeness of the data.
In addition to slow gait in short bouts, the complexity and variability of real-world environments further challenge the performance of algorithms. Factors such as deviations from a straight path, turns and obstacles, limited visibility, crowded areas, and other mobility tasks such as navigating stairs or slopes may have influenced the DMO estimates. These conditions make stride length estimation particularly challenging as this DMO is sensitive to intermittent gait []. Stride length exhibited larger errors and poor reliability for all WB durations, contributing to the errors in walking speed and distance estimates. This lower performance in stride length estimation could be related to the algorithms assuming an inverted pendulum, which might not properly fit the hip fracture cohort due to their asymmetrical gait []. Further work is necessary to improve stride length estimation in older adults after hip fracture.
Impact of Physical Function
Our results show that walking speed can be estimated with good reliability in older adults with higher SPPB scores in all WB duration categories, with errors ranging from –3.59% to –1.73%. Conversely, the lower–SPPB score group exhibited larger errors, ranging from 26.58% to 32.56%, and demonstrated poor reliability for all WB duration categories except for WBs lasting >30 seconds, which showed moderate reliability and lower errors (18.96%). These group differences are likely due to the differences in participants’ real-world walking speed. As reported in our previous work, the performance of all algorithms decreased for speeds of <0.5 m/s [], a threshold distinguishing between slow and medium-speed walkers []. Those with lower SPPB scores walked slower on average than those with higher SPPB scores (0.59, SD 0.17 m/s vs 0.70, SD 0.20 m/s, respectively), as measured using the reference system. The lower–SPPB score group walked slower and had higher errors in 10- to 30-second WBs. Although 22% (36/164) of all WBs across participants were of <0.5 m/s, only 9% (1/11) of the participants had a median speed slower than this (Figure 1 and Table 1; participant 3). Notably, this participant was among those with the lowest errors in walking speed and distance estimates despite having an SPPB score of only 3. In general, errors in slow walkers may be due to lower amplitude in the acceleration signals, inconsistent gait cycles [,], irregular gait patterns [,], short step length, shuffling gait [], and less symmetrical gait []. Nevertheless, DMOs in the lower–SPPB score group demonstrated better accuracy and moderate reliability in WBs of >30 seconds, indicating that accuracy improves during longer WBs irrespective of physical function.
Patients have individual preferences regarding whether walking faster or further is more important [], and this often depends on the context. For instance, walking speed is essential for tasks such as crossing the road, whereas walking distance is more relevant for engaging in activities such as going to the grocery store or participating in the community. As most patients report walking further as more important [], a reliable distance estimate would provide clinicians with valuable information that is meaningful to their patients. Overall, distance estimates showed good to excellent reliability and low errors, especially for the longest WBs. When stratified by SPPB score, distance exhibited the same error pattern as walking speed, with errors being larger in shorter WBs. Although longer WBs could be more informative as these likely represent more consistent outdoor walking, for many patients, life space after a hip fracture may be confined primarily to their home environment, where furniture, walls, and doors are likely to result in significantly shorter WBs.
Distance was overestimated in older adults with lower SPPB scores and underestimated in those with higher SPPB scores, showing large variations in errors despite excellent reliability. This could be explained by the combination of overestimated average walking speed and underestimated WB duration [], the multiplication of which yielded the distance DMO. Furthermore, high errors in the lower–SPPB score group could be explained by an inclusive WB definition that allowed for brief pauses of up to 3 seconds []. Such brief pauses are included in the total duration of a WB but are not taken into account in the calculation of mean walking speed, leading to a potential overestimation of distance. Older adults with lower physical function after a hip fracture may be more likely to take brief pauses and make slower turns within a WB than those with higher physical function. Therefore, when interpreting walked distance, researchers and clinicians should focus on longer and more continuous WBs if possible.
Impact of Time Since Surgery and Walking Aids
Promisingly, our results showed no discernible change in the accuracy of walking speed and distance estimates across 32 to 390 days after hip fracture surgery, suggesting that these DMOs can be used to capture daily-life gait from as early as 1 month after surgery. As this study assessed DMO errors in the real world, we only included participants who were mobile in their home environment, which may have contributed to not having included patients within the first month after surgery. Therefore, we cannot conclude on the accuracy and reliability of DMOs in the acute phase after hip fracture. However, in the initial days and weeks following hip fracture surgery, when many patients are still in the hospital or rehabilitation center [,], clinical assessments may be more appropriate and informative than real-world gait monitoring. The primary focus during this period is typically first on pain management and early mobilization, subsequently followed by rehabilitation aimed at improving physical function and enabling independent walking to prepare patients for discharge to home environments [,]. Importantly, physical function usually improves the most during the first 3 months after hip fracture surgery [], with DMO accuracy for participants assessed within this period showing no discernible deviation from that of participants assessed after the first 3 months. Similarly, DMO accuracy for the 18% (2/11) of participants who used walking aids did not deviate visibly from that of participants who did not. Although we found no evidence that days since surgery or the use of walking aids systematically affect DMO accuracy and reliability, these results in our small sample size should be confirmed in a larger sample.
General Discussion
In patients after hip fracture with reduced walking ability, shorter walking duration, or slower walking speed, additional mobility outcomes may provide relevant information about their movement characteristics. The core DMOs from the Mobilise-D TVS allow for a broader assessment of additional DMOs, such as walking duration and number of WBs exceeding different duration thresholds. Other relevant metrics in this population may include the number of turns, time spent in an upright position, and the frequency of posture transitions (eg, sitting to standing). In particular, the number of sit-to-stand transitions may be an important mobility outcome for patients after hip fracture, especially in the acute phase, and ongoing research is exploring this further.
While the lower back remains a viable location for gait monitoring in the daily lives of patients after hip fracture, it may not be optimal for long-term recordings. Extended monitoring periods require sufficient battery life and may cause discomfort, particularly when sitting or lying down. In addition, adhesive attachments may irritate the skin, whereas belts used for fixation can introduce movement artifacts that compromise data quality. Other wearable device locations may help overcome some of these challenges. Although wrist-worn devices could be an option, wrist-based data are currently less accurate than lower back data [] and have shown lower accuracy in patients recovering from hip fracture []. In patients with frailty and slow and cautious gait, a more suitable location for a single wearable device may be the foot or shank, which is better suited for detecting initial contact events than the lower back []. Furthermore, promising alternative approaches for gait monitoring in patients after hip fracture include instrumented walking aids [] and even the prostheses themselves [], a recent area of innovative development.
Recent comprehensive validation studies represent a significant step toward advancing real-world gait monitoring in multiple cohorts by providing robust reliability and accuracy data for DMOs [,]. Currently, clinically relevant accuracy thresholds have not yet been established for these DMOs in the hip fracture cohort, and they may also differ for different DMOs, making it challenging to define their appropriateness for clinical use. Our comprehensive and transparent reporting of accuracy and reliability measures allows readers to interpret the results in the context of their own research or clinical practice. Furthermore, based on the DMOs in this study, ongoing research is investigating the minimal important difference using the Mobilise-D clinical validation study dataset []. Further research is needed to establish clinically relevant thresholds to strengthen clinical applicability.
Clinical Implications
As most walking activity happens in daily life, outside a laboratory or clinic, understanding gait recovery after hip fracture surgery should not only rely on snapshot tests in the clinic or assessments in the laboratory but be complemented with accurate and reliable real-world monitoring. This study shows that real-world gait monitoring can provide much-needed information about a patient’s functional recovery. Currently, the Mobilise-D consortium is undertaking a comprehensive clinical validation of all validated DMOs [], including construct validity, predictive capacity, and the ability to detect change. Together with the encouraging results of this study, this opens up the study of real-world trajectories of gait recovery after a hip fracture and investigation of what characteristics of the patient, treatment, or rehabilitation may predict these trajectories.
This study showed that, for the hip fracture cohort, the number of steps and cadence DMOs are the most accurate and reliable, as well as suitable for both short and longer walking periods. These DMOs can also account for participants with functional constraints preventing them from performing outdoor walking. Furthermore, we found that walking speed, stride duration, and walking distance can be used to provide additional information beyond the most frequently studied DMOs, such as number of steps, cadence, and walking time [-]. In terms of walking speed, we recommend excluding the shortest WBs of <10 seconds to ensure accurate and reliable estimates without disregarding a disproportionate amount of data. Furthermore, clinicians should consider the patients’ physical function when interpreting walking speed due to the overestimation in people with lower SPPB scores and underestimation in people with higher SPPB scores. Distance and stride duration showed promising results across all WB durations. However, in subgroups with lower SPPB scores (<8), WBs of >30 seconds are required for obtaining more accurate distance estimates, and this may be challenging for patients with the most frailty. We showed that the distance and walking speed DMOs can be used as early as 1 month after surgery also in people using a single cane or crutch, but more data on walking aid use are needed from a larger sample.
Strengths and Limitations
To the best of our knowledge, this is the first study to rigorously investigate the accuracy and reliability of real-world DMOs while considering critical hip fracture–related factors. A key strength is the sufficient number of 164 WBs [], which improves the statistical power. However, the number of available WBs decreased substantially when excluding short WBs. Although this improved the accuracy and reliability of several DMOs, it also excluded a large part of participants’ data, potentially jeopardizing their representativeness of daily life activities. Moreover, some participants, especially those with better physical function, contributed more WBs than others. This imbalance may have biased the analyses by overrepresenting the walking patterns of more mobile participants. Future studies could address this by accounting for within-subject variability and ensuring more balanced data collection across participants.
While our study sample included participants with a wide range of physical function and walking speeds, with SPPB scores ranging from 3 to 12 and 21.9% (36/164) of the WBs being of <0.5 m/s, our hip fracture sample was relatively well functioning compared to the general population after hip fracture, who may have SPPB scores ranging from 1 to 5 in the first year after surgery []. In addition, while we had enough WBs to compare sample by sample in the true-positive analysis [], our number of participants included in the analyses was small. As a result, formal analyses of the potential effect of days since surgery and walking aid use on DMO accuracy and reliability were not feasible.
Several factors might explain the specifics of our study sample. First, we only included participants who were mobile in their home environment. Second, as we recruited during the pandemic, we were prohibited from recruiting participants during their hospital stay, which consequently excluded those in the acute phase. Third, the complexity and thoroughness of the protocol likely also limited the recruitment of patients with the most frailty. Fourth, while we initially recruited 19 participants from a target sample size of 20, the sample size was reduced to 11 (58%) in the final analysis, primarily due to technical errors with the reference system. However, 11% (2/19) of the participants were excluded due to insufficient signal quality, which may have been related to their low physical function (SPPB scores of 3), potentially accompanied by shuffling or a very cautious gait, making it challenging to detect gait events. As a result, our findings may not fully generalize to patients with more severe impairments, and a larger sample is needed to provide stronger evidence. These limitations constrain our findings but offer a promising foundation for future research to improve the robustness and generalizability of DMOs in frail populations.
Conclusions
Considering specific factors critical for older adults after a hip fracture, our study identified 5 accurate and reliable real-world DMO estimates from a single wearable device worn on the lower back: walking speed, cadence, stride duration, number of steps, and distance. The accuracy and reliability of most DMOs improved when excluding WBs of <10 seconds and were higher for WBs of >30 seconds than for WBs of 10 to 30 seconds and for participants with higher physical function. DMOs can capture daily gait as early as 1 month after surgery also in people using walking aids. However, as most patients after hip fracture perform WBs of short duration, there is a trade-off between more accurate and reliable walking speed and distance estimates and the disregard of substantial amounts of data. Our results add more granularity to real-world gait assessments in populations with severe gait impairments. They have important implications for future research as they can provide a significant contribution to clinical validation studies, randomized controlled trials, and descriptive studies on gait recovery after hip fracture that is meaningful for clinicians and patients alike. Finally, these results support the use of these DMOs to assess intervention effects on real-world gait in detail, thereby aiding to the design of optimal care pathways.
Acknowledgments
This work was supported by the Mobilise-D project that has received funding from the Innovative Medicines Initiative (IMI) 2 Joint Undertaking under grant 820820. This Joint Undertaking receives support from the European Union’s Horizon 2020 research and innovation program and the European Federation of Pharmaceutical Industries and Associations. The authors thank the Mobilise-D technical validation study team of work package 2, all participants who devoted their time and energy to this taxing protocol in times of COVID-19, and Miriam Dinger at the Robert Bosch Foundation for Medical Research and Dr Tim Klüter at University Hospital Schleswig-Holstein for their indispensable help in recruiting patients. MAB acknowledges ChatGPT, an artificial intelligence model by OpenAI, for suggestions that helped improve the clarity of some sentences in this paper. SDD and LR were also supported by the Identifying Digital Endpoints to Assess Fatigue, Sleep, and Activities in Daily Living in Neurodegenerative Disorders and Immune-Mediated Inflammatory Diseases project that has received funding from the IMI 2 Joint Undertaking under grant agreement 853981. LR is a National Institute for Health and Care Research (NIHR) senior investigator. Both SDD and LR were supported by the NIHR Newcastle Biomedical Research Centre based at the Newcastle upon Tyne Hospitals National Health Service (NHS) Foundation Trust; Newcastle University; and the Cumbria, Northumberland, Tyne, and Wear NHS Foundation Trust. SDD and LR were also supported by the NIHR and Wellcome Trust Clinical Research Facility infrastructure at Newcastle upon Tyne Hospitals NHS Foundation Trust. SDD was supported by the UK Research and Innovation Engineering and Physical Sciences Research Council (grant reference EP/X031012/1 and EP/X036146/1). The Barcelona Institute for Global Health acknowledges support from grant CEX2018-000806-S funded by the Ministry of Science, Innovation, and Universities and the State Research Agency (MCIN/AEI/10.13039/501100011033) and support from the Government of Catalonia through the Research Centers of Catalonia program. The views expressed in this publication are those of the authors and do not necessarily reflect those of the IMI, the European Union, the European Federation of Pharmaceutical Industries and Associations, any associated partners, the NIHR, or the Department of Health and Social Care, who are not responsible for any use that may be made of the information contained herein.
Data Availability
The datasets generated or analyzed during the Mobilise-D technical validation study are available in the Zenodo repository [].
Conflicts of Interest
SDD reports consultancy activity with Hoffmann-La Roche Ltd outside of this study. All other authors declare no other conflicts of interest.
References
- Emmerson BR, Varacallo M, Inman D. Hip fracture overview. StatPearls. URL: https://www.ncbi.nlm.nih.gov/books/NBK557514/ [accessed 2024-04-29]
- Downey C, Kelly M, Quinlan JF. Changing trends in the mortality rate at 1-year post hip fracture - a systematic review. World J Orthop. Mar 18, 2019;10(3):166-175. [FREE Full text] [CrossRef] [Medline]
- Dyer SM, Crotty M, Fairhall N, Magaziner J, Beaupre LA, Cameron ID, et al. Fragility Fracture Network (FFN) Rehabilitation Research Special Interest Group. A critical review of the long-term disability outcomes following hip fracture. BMC Geriatr. Sep 02, 2016;16(1):158. [FREE Full text] [CrossRef] [Medline]
- Delgado-Ortiz L, Polhemus A, Keogh A, Sutton N, Remmele W, Hansen C, et al. Listening to the patients' voice: a conceptual framework of the walking experience. Age Ageing. Jan 08, 2023;52(1):afac233. [FREE Full text] [CrossRef] [Medline]
- Wahlsten LR, Smedegaard L, Brorson S, Gislason G, Palm H. Living settings and cognitive impairment are stronger predictors of nursing home admission after hip fracture surgery than physical comorbidities: a nationwide Danish cohort study. Injury. Oct 2020;51(10):2289-2294. [CrossRef] [Medline]
- Grigoryan KV, Javedan H, Rudolph JL. Orthogeriatric care models and outcomes in hip fracture patients: a systematic review and meta-analysis. J Orthop Trauma. Mar 2014;28(3):e49-e55. [FREE Full text] [CrossRef] [Medline]
- Werner M, Macke C, Gogol M, Krettek C, Liodakis E. Differences in hip fracture care in Europe: a systematic review of recent annual reports of hip fracture registries. Eur J Trauma Emerg Surg. Jun 08, 2022;48(3):1625-1638. [FREE Full text] [CrossRef] [Medline]
- Nijdam TM, Laane DW, Schiepers TE, Smeeing DP, Kempen DH, Willems HC, et al. The goals of care in acute setting for geriatric patients in case of a hip fracture. Eur J Trauma Emerg Surg. Aug 18, 2023;49(4):1835-1844. [FREE Full text] [CrossRef] [Medline]
- Reyes BJ, Mendelson DA, Mujahid N, Mears SC, Gleason L, Mangione KK, et al. Postacute management of older adults suffering an osteoporotic hip fracture: a consensus statement from the international geriatric fracture society. Geriatr Orthop Surg Rehabil. Jul 16, 2020;11:2151459320935100. [FREE Full text] [CrossRef] [Medline]
- Taraldsen K, Thingstad P, Sletvold O, Saltvedt I, Lydersen S, Granat MH, et al. The long-term effect of being treated in a geriatric ward compared to an orthopaedic ward on six measures of free-living physical behavior 4 and 12 months after a hip fracture - a randomised controlled trial. BMC Geriatr. Dec 04, 2015;15:160. [FREE Full text] [CrossRef] [Medline]
- Taylor NF, Peiris CL, Thompson AL, Prendergast LA, Harding KE, Hau R, et al. Association between physical activity and short-term physical function changes after hip fracture: an observational study. Physiother Res Int. Jan 2021;26(1):e1876. [CrossRef] [Medline]
- Dargent-Molina P, Favier F, Grandjean H, Baudoin C, Schott AM, Hausherr E, et al. Fall-related factors and risk of hip fracture: the EPIDOS prospective study. Lancet. Jul 20, 1996;348(9021):145-149. [CrossRef] [Medline]
- Thingstad P, Egerton T, Ihlen EF, Taraldsen K, Moe-Nilssen R, Helbostad JL. Identification of gait domains and key gait variables following hip fracture. BMC Geriatr. Nov 18, 2015;15(1):150. [FREE Full text] [CrossRef] [Medline]
- Hulleck AA, Menoth Mohan D, Abdallah N, El Rich M, Khalaf K. Present and future of gait assessment in clinical practice: towards the application of novel trends and technologies. Front Med Technol. 2022;4:901331. [FREE Full text] [CrossRef] [Medline]
- Taraldsen K, Polhemus A, Engdal M, Jansen CP, Becker C, Brenner N, et al. Evaluation of mobility recovery after hip fracture: a scoping review of randomized controlled studies. Osteoporos Int. Mar 06, 2024;35(2):203-215. [FREE Full text] [CrossRef] [Medline]
- Cappozzo A. Gait analysis methodology. Hum Mov Sci. Mar 1984;3(1-2):27-50. [CrossRef]
- Warmerdam E, Hausdorff JM, Atrsaei A, Zhou Y, Mirelman A, Aminian K, et al. Long-term unsupervised mobility assessment in movement disorders. Lancet Neurol. May 2020;19(5):462-470. [CrossRef] [Medline]
- Van Ancum JM, van Schooten KS, Jonkman NH, Huijben B, van Lummel RC, Meskers CG, et al. Gait speed assessed by a 4-m walk test is not representative of daily-life gait speed in community-dwelling adults. Maturitas. Mar 2019;121:28-34. [CrossRef] [Medline]
- Hillel I, Gazit E, Nieuwboer A, Avanzino L, Rochester L, Cereatti A, et al. Is every-day walking in older adults more analogous to dual-task walking or to usual walking? Elucidating the gaps between gait performance in the lab and during 24/7 monitoring. Eur Rev Aging Phys Act. May 3, 2019;16(1):6. [FREE Full text] [CrossRef] [Medline]
- Rochester L, Mazzà C, Mueller A, Caulfield B, McCarthy M, Becker C, et al. A roadmap to inform development, validation and approval of digital mobility outcomes: the Mobilise-D approach. Digit Biomark. Nov 26, 2020;4(Suppl 1):13-27. [FREE Full text] [CrossRef] [Medline]
- Rehman RZ, Klocke P, Hryniv S, Galna B, Rochester L, Del Din S, et al. Turning detection during gait: algorithm validation and influence of sensor location and turning characteristics in the classification of Parkinson’s disease. Sensors (Basel). Sep 19, 2020;20(18):5377. [FREE Full text] [CrossRef] [Medline]
- Mazzà C, Alcock L, Aminian K, Becker C, Bertuletti S, Bonci T, et al. Technical validation of real-world monitoring of gait: a multicentric observational study. BMJ Open. Dec 02, 2021;11(12):e050785. [FREE Full text] [CrossRef] [Medline]
- Connecting digital mobility assessment to clinical outcomes for regulatory and clinical endorsement. Mobilise-D. URL: https://mobilise-d.eu/ [accessed 2024-08-23]
- Salis F, Bertuletti S, Bonci T, Caruso M, Scott K, Alcock L, et al. A multi-sensor wearable system for the assessment of diseased gait in real-world conditions. Front Bioeng Biotechnol. Apr 21, 2023;11:1143248. [FREE Full text] [CrossRef] [Medline]
- Scott K, Bonci T, Salis F, Alcock L, Buckley E, Gazit E, et al. Mobilise-D consortium. Design and validation of a multi-task, multi-context protocol for real-world gait simulation. J Neuroeng Rehabil. Dec 16, 2022;19(1):141. [FREE Full text] [CrossRef] [Medline]
- Kirk C, Küderle A, Micó-Amigo ME, Bonci T, Paraschiv-Ionescu A, Ullrich M, et al. Mobilise-D consortium. Mobilise-D insights to estimate real-world walking speed in multiple conditions with a wearable device. Sci Rep. Jan 19, 2024;14(1):1754. [FREE Full text] [CrossRef] [Medline]
- Micó-Amigo ME, Bonci T, Paraschiv-Ionescu A, Ullrich M, Kirk C, Soltani A, et al. Mobilise-D consortium. Assessing real-world gait with digital technology? Validation, insights and recommendations from the Mobilise-D consortium. J Neuroeng Rehabil. Jun 14, 2023;20(1):78. [FREE Full text] [CrossRef] [Medline]
- Najafi B, Helbostad JL, Moe-Nilssen R, Zijlstra W, Aminian K. Does walking strategy in older people change as a function of walking distance? Gait Posture. Mar 2009;29(2):261-266. [CrossRef] [Medline]
- Kline Mangione K, Craik RL, Lopopolo R, Tomlinson JD, Brenneman SK. Predictors of gait speed in patients after hip fracture. Physiother Can. 2008;60(1):10-18. [FREE Full text] [CrossRef] [Medline]
- Prestmo A, Hagen G, Sletvold O, Helbostad JL, Thingstad P, Taraldsen K, et al. Comprehensive geriatric care for patients with hip fractures: a prospective, randomised, controlled trial. Lancet. Apr 25, 2015;385(9978):1623-1633. [CrossRef] [Medline]
- Thomas S, Halbert J, Mackintosh S, Cameron ID, Kurrle S, Whitehead C, et al. Walking aid use after discharge following hip fracture is rarely reviewed and often inappropriate: an observational study. J Physiother. 2010;56(4):267-272. [FREE Full text] [CrossRef] [Medline]
- Bertuletti S, Della Croce U, Cereatti A. A wearable solution for accurate step detection based on the direct measurement of the inter-foot distance. J Biomech. Mar 14, 2019;84:274-277. [CrossRef] [Medline]
- Nasreddine ZS, Phillips NA, Bédirian V, Charbonneau S, Whitehead V, Collin I, et al. The Montreal Cognitive Assessment, MoCA: a brief screening tool for mild cognitive impairment. J Am Geriatr Soc. Apr 2005;53(4):695-699. [CrossRef] [Medline]
- Salis F, Bertuletti S, Bonci T, Della Croce U, Mazzà C, Cereatti A. A method for gait events detection based on low spatial resolution pressure insoles data. J Biomech. Oct 11, 2021;127:110687. [CrossRef] [Medline]
- Caruso M, Sabatini AM, Knaflitz M, Gazzoni M, Croce UD, Cereatti A. Orientation estimation through magneto-inertial sensor fusion: a heuristic approach for suboptimal parameters tuning. IEEE Sensors J. Feb 1, 2021;21(3):3408-3419. [CrossRef]
- mobilise-d / mobgap. GitHub. URL: https://github.com/mobilise-d/mobgap [accessed 2024-08-28]
- Kluge F, Del Din S, Cereatti A, Gaßner H, Hansen C, Helbostad JL, et al. Mobilise-D consortium. Consensus based framework for digital mobility monitoring. PLoS One. Aug 20, 2021;16(8):e0256541. [FREE Full text] [CrossRef] [Medline]
- Baroudi L, Barton K, Cain SM, Shorter KA. Classification of human walking context using a single-point accelerometer. Sci Rep. Mar 06, 2024;14(1):3039. [FREE Full text] [CrossRef] [Medline]
- Schmitt AC, Baudendistel ST, Lipat AL, White TA, Raffegeau TE, Hass CJ. Walking indoors, outdoors, and on a treadmill: gait differences in healthy young and older adults. Gait Posture. Oct 2021;90:468-474. [CrossRef] [Medline]
- Guralnik JM, Simonsick EM, Ferrucci L, Glynn RJ, Berkman LF, Blazer DG, et al. A short physical performance battery assessing lower extremity function: association with self-reported disability and prediction of mortality and nursing home admission. J Gerontol. Mar 1994;49(2):M85-M94. [CrossRef] [Medline]
- Beckmann M, Bruun-Olsen V, Pripp AH, Bergland A, Smith T, Heiberg KE. Recovery and prediction of physical function 1 year following hip fracture. Physiother Res Int. Jul 24, 2022;27(3):e1947. [FREE Full text] [CrossRef] [Medline]
- Keppler AM, Holzschuh J, Pfeufer D, Gleich J, Neuerburg C, Kammerlander C, et al. Mobility improvement in the first 6 postoperative weeks in orthogeriatric fracture patients. Eur J Trauma Emerg Surg. Aug 2022;48(4):2867-2872. [FREE Full text] [CrossRef] [Medline]
- Giavarina D. Understanding Bland Altman analysis. Biochem Med (Zagreb). 2015;25(2):141-151. [FREE Full text] [CrossRef] [Medline]
- McGraw KO, Wong SP. Forming inferences about some intraclass correlation coefficients. Psychol Methods. Mar 1996;1(1):30-46. [CrossRef]
- Koo TK, Li MY. A guideline of selecting and reporting intraclass correlation coefficients for reliability research. J Chiropr Med. Jun 2016;15(2):155-163. [FREE Full text] [CrossRef] [Medline]
- Combs SA, Van Puymbroeck M, Altenburger PA, Miller KK, Dierks TA, Schmid AA. Is walking faster or walking farther more important to persons with chronic stroke? Disabil Rehabil. May 2013;35(10):860-867. [CrossRef] [Medline]
- Studenski S, Perera S, Patel K, Rosano C, Faulkner K, Inzitari M, et al. Gait speed and survival in older adults. JAMA. Jan 05, 2011;305(1):50-58. [FREE Full text] [CrossRef] [Medline]
- Hebenstreit F, Leibold A, Krinner S, Welsch G, Lochmann M, Eskofier BM. Effect of walking speed on gait sub phase durations. Hum Mov Sci. Oct 2015;43:118-124. [CrossRef] [Medline]
- Quintero D, Lambert DJ, Villarreal DJ, Gregg RD. Real-time continuous gait phase and speed estimation from a single sensor. Control Technol Appl. Aug 2017;2017:847-852. [FREE Full text] [CrossRef] [Medline]
- Soltani A, Aminian K, Mazza C, Cereatti A, Palmerini L, Bonci T, et al. Algorithms for walking speed estimation using a lower-back-worn inertial sensor: a cross-validation on speed ranges. IEEE Trans Neural Syst Rehabil Eng. 2021;29:1955-1964. [CrossRef]
- Taraldsen K, Thingstad P, Døhl Ø, Follestad T, Helbostad JL, Lamb SE, et al. Short and long-term clinical effectiveness and cost-effectiveness of a late-phase community-based balance and gait exercise program following hip fracture. The EVA-Hip randomised controlled trial. PLoS One. 2019;14(11):e0224971. [FREE Full text] [CrossRef] [Medline]
- Kim DW, Hassett LM, Nguy V, Allen NE. A comparison of activity monitor data from devices worn on the wrist and the waist in people with Parkinson’s disease. Mov Disord Clin Pract. Nov 2019;6(8):693-699. [FREE Full text] [CrossRef] [Medline]
- Kluge F, Brand YE, Micó-Amigo ME, Bertuletti S, D'Ascanio I, Gazit E, et al. Real-world gait detection using a wrist-worn inertial sensor: validation study. JMIR Form Res. May 01, 2024;8:e50035. [FREE Full text] [CrossRef] [Medline]
- Pacini Panebianco G, Bisi MC, Stagni R, Fantozzi S. Analysis of the performance of 17 algorithms from a systematic review: influence of sensor position, analysed variable and computational approach in gait timing estimation from IMU measurements. Gait Posture. Oct 2018;66:76-82. [CrossRef] [Medline]
- Fernandez-Carmona M, Ballesteros J, Díaz-Boladeras M, Parra-Llanas X, Urdiales C, Gómez-de-Gabriel JM. Walk-IT: an open-source modular low-cost smart rollator. Sensors (Basel). Mar 08, 2022;22(6):2086. [FREE Full text] [CrossRef] [Medline]
- Merle G, Parent-Harvey A, Harvey EJ. Sensors and digital medicine in orthopaedic surgery. OTA Int. Apr 2022;5(2 Suppl):e189. [FREE Full text] [CrossRef] [Medline]
- Mikolaizak AS, Rochester L, Maetzler W, Sharrack B, Demeyer H, Mazzà C, et al. clinical validation study (WP4) on behalf of Mobilise-D consortium. Connecting real-world digital mobility assessment to clinical outcomes for regulatory and clinical endorsement-the Mobilise-D study protocol. PLoS One. 2022;17(10):e0269615. [FREE Full text] [CrossRef] [Medline]
- O'Halloran PD, Shields N, Blackstock F, Wintle E, Taylor NF. Motivational interviewing increases physical activity and self-efficacy in people living in the community after hip fracture: a randomized controlled trial. Clin Rehabil. Nov 2016;30(11):1108-1119. [CrossRef] [Medline]
- Senserrick C, Lawler K, Scroggie GD, Williams K, Taylor NF. Three short sessions of physiotherapy during rehabilitation after hip fracture were no more effective in improving mobility than a single longer session: a randomised controlled trial. Physiotherapy. Sep 2021;112:87-95. [CrossRef] [Medline]
- Zusman EZ, Dawes M, Fleig L, McAllister MM, Cook WL, Guy P, et al. Older adults' sedentary behavior and physical activity after hip fracture: results from an outpatient rehabilitation randomized controlled trial. J Geriatr Phys Ther. 2019;42(2):E32-E38. [CrossRef] [Medline]
- Küderle A. Mobilise-D Technical Validation Study (TVS) dataset. Zenodo. URL: https://zenodo.org/records/13899386 [accessed 2024-10-21]
Abbreviations
| DMO: digital mobility outcome |
| ICC: intraclass correlation coefficient |
| IMU: inertial measurement unit |
| MARE: mean absolute relative error |
| MRE: mean relative error |
| SPPB: Short Physical Performance Battery |
| TVS: technical validation study |
| WB: walking bout |
Edited by A Mavragani; submitted 23.10.24; peer-reviewed by B Cornish, S Lord; comments to author 03.03.25; revised version received 31.03.25; accepted 02.04.25; published 20.05.25.
Copyright©Martin A Berge, Anisoara Paraschiv-Ionescu, Cameron Kirk, Arne Küderle, Encarna Micó-Amigo, Clemens Becker, Andrea Cereatti, Silvia Del Din, Monika Engdal, Judith Garcia-Aymerich, Karoline B Grønvik, Clint Hansen, Jeffrey M Hausdorff, Jorunn L Helbostad, Carl-Philipp Jansen, Lars Gunnar Johnsen, Jochen Klenk, Sarah Koch, Walter Maetzler, Dimitrios Megaritis, Arne Müller, Lynn Rochester, Lars Schwickert, Kristin Taraldsen, Beatrix Vereijken. Originally published in JMIR Formative Research (https://formative.jmir.org), 20.05.2025.
This is an open-access article distributed under the terms of the Creative Commons Attribution License (https://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work, first published in JMIR Formative Research, is properly cited. The complete bibliographic information, a link to the original publication on https://formative.jmir.org, as well as this copyright and license information must be included.