Objective Parkinson’s disease can impair gait and stability, leading to reduced independence and increased fall risk. While speed dependent treadmill training (SDTT) is clinically effective, the specific biomechanical and neurophysiological mechanisms driving these improvements remain unclear. The “StepuP” multicenter randomized controlled trial aims to elucidate these mechanisms and determine whether training enriched with virtual reality or mechanical perturbations (SDTT+) enhances gait efficacy and transfer to daily life. Methods We will recruit 126 individuals with Parkinson’s disease across four clinical sites and 21 healthy older adults as a reference group. Participants will be randomized to receive either standard SDTT or SDTT+ for 12 sessions. To capture the trajectory of recovery and retention, assessments will occur at three distinct timepoints: baseline, post-intervention, and a 12-week follow-up, each assessment including synchronized 64-channel electroencephalography (EEG), electromyography (EMG), and 3D kinematics. This multimodal setup allows for the quantification of cortical beta-band activity, corticomuscular coherence, and stability-related foot placement control. Furthermore, we will assess participant’s satisfaction, usability, and engagement through questionnaires and interviews to understand individual adherence and barriers to training. Significance The primary clinical endpoint is comfortable overground walking speed. We hypothesize that gait improvements are mediated by improved stability-related foot placement and cortical sensorimotor integration. By correlating lab-based mechanistic changes with real-world mobility patterns and participant experiences, this study seeks to identify specific pathophysiological mechanisms engaged during the treadmill training. These insights will help distinguish responders from non-responders, facilitating the development of personalized, acceptable, and effective rehabilitation strategies.

Freezing of gait is a common, debilitating symptom that affects many patients with Parkinson's disease. The lack of a standard, objective method to quantify freezing obstructs research and treatment. Wearable sensors combined with automatic detection algorithms have demonstrated increasingly promising results; nonetheless, video annotation remains the gold standard. After organizing a global machine learning contest to expedite the development of acceleration-based algorithms designed to automatically detect freezing, we tested the transferability of the winning models to a new dataset. Experts reviewed and annotated a test protocol conducted and videotaped in the homes of 12 patients. The models were applied to acceleration data from a lower back sensor worn by the patients. F1-scores, accuracy, recall, specificity, and precision were computed. Intraclass correlations quantified the agreement between model-estimated and annotation-based gold standard outcomes, including the percent time frozen, the number of episodes, and the total freezing duration. While there was a relatively large drop in performance for some of the models, the performance of the third place model showed good transferability to new data. Indeed, the agreement between the third place model and gold standard annotations was similar to or better than that seen when comparing two raters. These results further support the idea that if the goal is to detect freezing duration or percent time frozen, the combination of a single, lower back sensor and the third place model can be used to automatically detect freezing. Still, if the goal is to count episodes or detect freezing subtypes, additional sensors or other modelling approaches are needed.

OBJECTIVE:Gait affects knee loading. Modifying gait could reduce load and protect against cartilage loss. Our objective is to look for modifiable gait parameters and determine their relation to worsening cartilage damage. METHODS:We studied participants from the Multicenter Osteoarthritis Study (MOST) ages 45 to 90 years with, or at risk for, knee osteoarthritis (OA). Gait assessment used inertial measurement units (APDM, Inc) on the pelvis and ankles during a 20-m walk. Knee magnetic resonance imaging (MRI) was acquired at baseline and two years later. Cartilage damage worsening was assessed using MRI Osteoarthritis Knee Scores in 14 knee subregions. We examined change (yes/no) in each subregion. We used ensemble machine learning to discriminate subregions with and without cartilage damage. Predictors tested included gait variables, radiographic OA, baseline cartilage damage, age, sex, height, weight, depressive symptoms, and race/clinic site. Data were split 70% training and 30% test sets. We identified the 10 variables that, across 100 repetitions, most frequently contributed to risk of damage. We used G-computation to evaluate causal risk differences of worsening cartilage damage for each variable. RESULTS:We studied 1,703 participants (mean [±SD] age 61.4 [±9.4] years, 56% female). At two years, 46% had worse cartilage damage in at least one knee subregion. Of gait variables, longer step length was associated with increased risk of damage, especially in knees with more baseline damage. CONCLUSION:Longer step length was associated with worse cartilage damage over two years. Interventions to shorten step length might reduce risk of worsening cartilage damage.

Objective, continuous assessment of real-world mobility using wearables has significant potential to transform clinical research and practice, yet the field lacks standardised, open-source tools that enable reproducible algorithm real-world validation, across multiple clinical cohorts. This would improve transparency around definitions and performance, thereby enhancing interpretation and more meaningful comparison across studies. The Mobilise-D consortium validated a comprehensive analytical pipeline for estimating digital mobility outcomes from wearables, originally implemented in a combination of MATLAB, R, and Python codes. To overcome the licencing, reproducibility, and accessibility limitations of this implementation, the pipeline has been re-implemented and re-validated, against gold standards, as the open-source mobgap Python package. Here, we describe the mobgap ecosystem, detail how algorithms can be integrated and benchmarked in a reproducible way and present a re-validation of the pipeline against reference data across six clinical cohorts under real-world conditions. Validation results showed that across all cohorts, walking speed was estimated with an absolute error of 0.10 m/s and an intraclass correlation coefficient (ICC) of 0.81, demonstrating comparable or superior performance to the original implementation. Mobgap (v1.2) is openly available and is intended to serve as a reproducible reference implementation and benchmarking platform for researchers developing or validating mobility analysis algorithms using wearable data.

The L-test is a performance-based measure to assess balance and mobility. Currently, the primary outcome from this test is the time required to finish it. In this study we present the instrumented L-test (iL-test), an L-test wherein mobility is evaluated by means of a wearable inertial sensor worn at the lower back. We analyzed data from 113 people across seven cohorts: healthy adults, chronic obstructive pulmonary disease, multiple sclerosis, congestive heart failure, Parkinson’s disease, proximal femoral fracture, and transfemoral amputation. The iL-test automatic segmentation was validated using stereophotogrammetry. Univariate and multivariate analyses were performed on 164 kinematic features derived from inertial signals to identify distinct patterns across different cohorts. The iL-test accurately recognized and segmented activities during the L-test for all cohorts (technical validity). A random forest classifier revealed that proximal femoral fracture and transfemoral amputation induced significantly different mobility patterns compared to healthy people with AUC values of 0.89 and 0.99, respectively. Strong correlations were found between kinematic features and clinical scores in multiple sclerosis, congestive heart failure, proximal femoral fracture, and transfemoral amputation, with consistent patterns of decreased movement ranges and smoothness with increasing disease severity. Furthermore, features derived from 90° and 180° turns were found to be important contributors to differentiation amongst cohorts, underscoring the need to evaluate different turn degrees and directions. This study emphasizes the iL-test potential to deliver automated mobility assessment across a wide range of clinical conditions, indicating a prospective avenue for improved mobility assessment and, eventually, more informed healthcare interventions.

ObjectiveTo assess the feasibility and, preliminarily, the effectiveness of long-term, personalized gait training using a digital wearable system (Gait Tutor) that provides real-time audio biofeedback to correct or reinforce gait behaviour.DesignOpen-label and non-controlled, with assessments before and after intervention.SettingReal-world.ParticipantsTwenty persons with Parkinson's disease.InterventionParticipants performed home-based gait training in their ON medication state for 30 minutes, 3 times per week, for 9 months using a Gait Tutor.Main measuresWe evaluated adherence (% of expected sessions), usability, and, preliminarily, efficacy by assessing the motor performance of the participants before and after the intervention.ResultsSeventeen participants (85%) completed the study, performing an average of 83 sessions. Adherence was higher for persons with an intermediate disease stage (80.5% of expected training sessions), compared to those with a more advanced disease stage (46.2%). All participants reported extremely positive scores on the questionnaire about ease of use and effectiveness (4.37 ± 0.42). The Movement Disorders Unified Parkinson's Disease Rating Scale motor scores remained stable after the training (mean 9 months). In people with an intermediate disease stage, clinical scores and physical capacity tended to improve.ConclusionsFor the first time, this study shows the feasibility of long-term real-world gait training for people with Parkinson's disease, providing preliminary evidence that personalized, technology-driven rehabilitation strategies can be sustained over extended periods and can assist clinicians in objectively assessing gait performance in the real world.

The response to a request to walk involves a motor planning phase followed by an execution phase. The initial phase of gait initiation, specifically the time to anticipatory postural adjustment (APA), can be viewed as a form of reaction time. However, it is not clear how to characterize the cognitive processes involved in this stage. To address this question, time-to-APA, simple and complex upper limb visuomotor reaction time (SRT, CRT), cognitive, and motor performance were evaluated in 27 people with Parkinson’s disease (PD), 31 older adults (OA), and 34 young adults (YA). Our results showed that time-to-APA was significantly longer than SRT in all three groups (p < 0.001), indicating a more complex cognitive process. In YA, time-to-APA was significantly shorter than CRT (p < 0.001). In the OA and PD, time-to-APA was not significantly different from CRT. Mixed-effects analysis showed significant time (p < 0.001), group (p = 0.037), and group × time interaction effects (p = 0.002). Among all subjects, time-to-APA, but not APA duration, was associated with the Color-Trails Test (part B: rs = 0.406, p < 0.001). In PD, APA duration was correlated with MDS-UPDRS-part 3 (motor) scores (rs = 0.535, p = 0.004), but time-to-APA was not (p = 0.892). These findings suggest that time-to-APA is a cognitive process that is more complex than a SRT task and shares properties of a CRT task, especially among older adults and people with PD. In PD, this initial movement planning stage is not related to motor impairment, in contrast to APA duration. Further research is necessary to identify the factors underlying this initial stage of gait initiation.

BACKGROUND:Wider step width and lower step-to-step variability are linked to improved gait stability and reduced fall risk. It is unclear if patients with spinocerebellar ataxia (SCA) can learn to adjust these aspects of gait to reduce fall risk. OBJECTIVES:The aims were to examine the possibility of using wearable step width haptic biofeedback to enhance gait stability and reduce fall risk in individuals with SCA. METHODS:Thirteen people with SCA type 3 performed step width training (single session) using real-time feedback. RESULTS:Step width increased post-training (19.3 cm, interquartile range [IQR] 16.3-20.2 cm) and at retention (16.6 cm, IQR 16.2-21.1 cm), compared to baseline (11.0 cm, IQR 5.2-15.2 cm; P < 0.001). Step width variability decreased during post-training (19.7%, IQR 17.4%-26.2%) and at retention (22.3%, IQR 18.6%-30.2%), compared to baseline (44.5%, IQR 28.5%-71.2%; P < 0.001). Crossover steps, another mark of instability, decreased after training (P < 0.031). CONCLUSIONS:These pilot results suggest that patients with SCA can use a novel, wearable biofeedback system to improve their gait stability. © 2025 International Parkinson and Movement Disorder Society.

Measuring ataxia severity is primarily conducted in-person using tests such as the Scale for the Assessment and Rating of Ataxia (SARA). However, given the motor and cognitive impairments of people with cerebellar ataxia (PwA), there are major limitations in ensuring the assessment is accessible and scalable. We aimed to develop and validate a novel test, enabling the remote assessment of ataxia severity, SARA-Le (SARA Live e-version). SARA-Le is a structured step-by-step test for administering the SARA through video conferencing. In two experiments, we administered SARA-Le to 106 PwA. In Experiment 1 (n = 23), we assessed concurrent validity by comparing SARA-Le and in-person SARA scores administered by an independent neurologist. In addition, we evaluated associations between nine gait measures and both SARA and SARA-Le scores. In Experiment 2 (n = 83), we assessed the efficacy, internal consistency, and correlations between SARA-Le and other related measures. First, we found a high correlation (r = 0.89, P = 0.001) between SARA-Le and in-person SARA scores, supporting convergent validity. Second, SARA-Le and SARA scores were both similarly associated with the nine gait measures, supporting construct validity. Third, SARA-Le’s Cronbach’s alpha was very high (0.831), supporting internal consistency. Fourth, SARA-Le scores exhibited a positive correlation with disease duration (r = 0.44, P < 0.001), and a negative correlation with MoCA scores (r = − 0.27, P = 0.007), supporting construct validity. SARA-Le can serve as a remote technology-based protocol, improving the accessibility and scalability of ataxia severity evaluation.

Impaired mobility increases falls and mortality risk. However, guidelines to reliably assess real-world walking activity and gait remain undefined. We aimed to (i) determine the minimum daily wear time during waking hours (7:00–22:00) for a valid measurement day, (ii) identify the minimum number of valid days, and (iii) weekend days, to reliably assess weekly walking activity and gait parameters, and (iv) provide recommendations for reliable real-world walking assessments. Participants with chronic obstructive pulmonary disease (n = 565), multiple sclerosis (n = 558), Parkinson’s disease (n = 543) or proximal femoral fracture (n = 487) from 10 countries were asked to wear a single wearable device on the lower back, 24 h/day for seven days, resulting in 13,191 measurement days. The Mobilise-D processing pipeline was used to obtain 24 daily walking activity and gait parameters. Minimum daily wear time was determined as the highest wear time category that did not statistically change parameter values. Intraclass correlation coefficients ≥ 0.80 determined the minimum number of valid measurement and weekend days. The minimum daily wear time varied between “no requirement” (13

Reserve is a physiological capacity used under demanding situations. The concept was developed to account for the discrepancy between pathology and clinical manifestation. In neuroscience, motor, brain and cognitive reserves are abstract measures, conceptually defined yet elusive to quantify. Reserve is indirectly assessed using proxies such as years of education and brain volume, limiting its utility. Moreover, the dichotomy in definitions of cognitive and motor reserves is artificial, as daily function requires an intricate network of connections between these domains. Here, we assessed the validity of a newly developed graded motor cognitive 'stress test' to quantify the combined motor and cognitive reserve (MCR). The study included 144 participants (ages between 18 and 85, 50% women) with a range of reserve capacities (i.e. healthy young and older adults and individuals with Parkinson's disease, Alzheimer's disease, dementia with Lewy bodies and mild cognitive impairment). The assessment included walking on a treadmill while negotiating motor and cognitive challenges delivered using virtual reality. To establish an MCR index score, we used a semi-supervised machine learning algorithm. The model includes performance measures from completing the stress test and measures obtained from wearable sensors used during the test. Validation of the proposed MCR index was examined through: (i) model face validity-reflecting decline of performance as challenge increased; (ii) known-groups validity-classification of scores according to neurological status; (iii) construct validity (convergent)-association with common MCRs proxies as well as MRI-derived regional brain volumes. The model's face validity revealed decreased performance with increased motor and cognitive challenges (both domains P < 0.001). The index accurately discriminated between healthy controls and those diagnosed with neurological conditions with an area under the curve of 0.89 [95% CI: 0.79-0.99] which was significantly higher than all other commonly used proxies. Statistically significant Spearman's ρ correlations were observed with all commonly used motor and cognitive proxies (0.56 ≤ r ≤ 0.79, after multiplicity correction all P < 0.05), reflecting construct validity. In addition, statistically significant correlations were observed between the MCR index and whole-brain grey matter and white matter volumes (r = 0.63 and 0.55), as well as the pre-defined left and right caudate nucleus (r = 0.56 and 0.68) and inferior-frontal gyrus (r = 0.47 and 0.58). This proof-of-concept study shows that the novel MCR index is valid, with high sensitivity to neurological deficits and is able to quantify reserve on an individual level. This new innovative tool can assist in screening for motor cognitive deficits and potentially, for predicting motor and cognitive decline associated with neurodegenerative disease.

Hemodynamic homeostasis is essential for adapting the heart rate (HR) to postural and physiological changes during daily activities. Traditional HR monitoring, such as 24 hour (h) Holter monitoring, provides important information on homeostasis during daily living. However, this approach lacks concurrent activity recording, limiting insights into hemodynamic adaptation and our ability to interpret changes in HR. To address this, we utilized a novel wearable sensor system (ANNE@Sibel) to capture time-locked HR and daily activity (i.e., lying, sitting, standing, walking) data in 105 community-dwelling older adults. We developed custom tools to extract 24 h time-locked measurements and introduced a “heart rate response score” (HRRS), based on root Jensen–Shannon divergence, to quantify HR changes relative to activity. As expected, we found a progressive HR increase with more vigorous activities, though individual responses varied widely, highlighting heterogeneous HR adaptations. The HRRS (mean: 0.38 ± 0.14; min: −0.11; max: 0.74) summarized person-specific HR changes and was correlated with several clinical measures, including systolic blood pressure changes during postural transitions (r = 0.325, p = 0.003), orthostatic hypotension status, and calcium channel blocker medication use. These findings demonstrate the potential of unobtrusive sensors in remote phenotyping as a means of providing valuable physiological and behavioral data to enhance the quantitative description of aging phenotypes. This approach could enhance personalized medicine by informing targeted interventions based on hemodynamic adaptations during everyday activities.

The response to a request to walk involves a motor planning phase followed by an execution phase. The initial phase of gait initiation, specifically the time to anticipatory postural adjustment (APA), can be viewed as a form of reaction time. However, it is not clear how to characterize the cognitive processes involved in this stage. To address this question, time-to-APA, simple and complex upper limb visuomotor reaction time (SRT, CRT), cognitive, and motor performance were evaluated in 27 people with Parkinson’s disease (PD), 31 older adults (OA), and 34 young adults (YA). Our results showed that time-to-APA was significantly longer than SRT in all three groups (p < 0.001), indicating a more complex cognitive process. In YA, time-to-APA was significantly shorter than CRT (p < 0.001). In the OA and PD, time-to-APA was not significantly different from CRT. Mixed-effects analysis showed significant time (p < 0.001), group (p = 0.037), and group × time interaction effects (p = 0.002). Among all subjects, time-to-APA, but not APA duration, was associated with the Color-Trails Test (part B: rs = 0.406, p < 0.001). In PD, APA duration was correlated with MDS-UPDRS-part 3 (motor) scores (rs = 0.535, p = 0.004), but time-to-APA was not (p = 0.892). These findings suggest that time-to-APA is a cognitive process that is more complex than an SRT task and shares properties of a CRT task, especially among older adults and people with PD. In PD, this initial movement planning stage is not related to motor impairment, in contrast to APA duration. Further research is necessary to identify the factors underlying this initial stage of gait initiation.

In people with Parkinson’s disease (PD), freezing of gait (FOG) can manifest as an absence of leg movement (akinetic) or a presence of high-frequency leg trembling. FOG is triggered most often during turning or dual-tasking when OFF-medication, but it is unclear whether the same holds true for akinetic and trembling FOG. To investigate the effects of dopaminergic medication and cognitive and motor tasks on trembling and akinetic FOG. Sixty-three PD patients with daily FOG performed a home-based FOG-provoking protocol OFF and ON-dopaminergic medication. FOG was video-annotated based on pre-specified definitions. We compared the

Mobility is a cornerstone of health and quality of life, particularly in older adults. Digital mobility outcomes (DMOs) from real-world walking data offer crucial insights into the functional status and early markers of mobility decline. This study provides reference values for walking activity, pace, rhythm, and gait bout-to-bout variability in community-dwelling older adults and evaluates the effects of age, sex, height, and weight on these parameters. Using data from 200 older adults (aged 65–94 years) from the InCHIANTI Study and applying the Mobilise-D computational pipeline, we analyzed real-world walking over a week. Significant differences by sex and age were found, with males showing higher walking activity in younger age groups (65–74 and 75–84 years) but not in the oldest group (85–94 years). Additionally, we observed non-linear trends in mobility metrics with age, indicating an accelerated reduction in mobility at certain age ranges. These findings underscore the importance of monitoring real-world walking data to pinpoint critical periods of mobility decline and guide targeted interventions. This work offers valuable benchmarks for clinical assessments and future research.

Wearable technology has rapidly advanced, opening new possibilities for context-aware applications in fields such as healthcare and gait analysis, where distinguishing between indoor and outdoor environments is essential. This is often accomplished through technologies like GPS, Wi-Fi, cellular, and Bluetooth which, however, come with privacy concerns, high power consumption, and dependency on external infrastructure. To address these challenges, recent studies have preliminary exploited the ambient magnetic field, though comprehensive validation with real-life data is lacking. This article seeks to validate machine learning techniques, i.e., random forest (RF), extreme gradient boosting, and stacked long short-term memory (LSTM) networks, for indoor-outdoor discrimination using exclusively magnetometer data from the daily activities of 20 participants in four cities across three countries. The study investigated the most effective magnetometer placement (feet, lower back, and nondominant wrist) and pre-processing techniques (e.g., features and window size). Reference data are obtained through a GPS-based algorithm coupled with a geographical database running on a smartphone. The extreme gradient boosting algorithm yielded the best results, with an accuracy of 0.91, an F1 -score of 0.90, and an area under the ROC curve of 0.94. These findings confirm the feasibility of accurately estimating indoor/outdoor context information from a magnetometer at the same update frequency as a common GPS, but with important energy savings. The proposed model can be integrated into state-of-the-art gait analysis systems, being able to discriminate the location to avoid misinterpreting gait deviations in real-world settings, thus supporting continuous and ubiquitous gait monitoring. Datasets and algorithm implementations have been made publicly available.

Step width is vital for gait stability, postural balance control, and fall risk reduction. However, estimating step width typically requires either fixed cameras or a full kinematic body suit of wearable inertial measurement units (IMUs), both of which are often too expensive and time-consuming for clinical application. We thus propose a novel data-augmented deep learning model for estimating step width in individuals with and without neurodegenerative disease using a minimal set of wearable IMUs. Twelve patients with neurodegenerative, clinically diagnosed Spinocerebellar ataxia type 3 (SCA3) performed over ground walking trials, and seventeen healthy individuals performed treadmill walking trials at various speeds and gait modifications while wearing IMUs on each shank and the pelvis. Results demonstrated step width mean absolute errors of 3.3 0.7 cm and 2.9 0.5 cm for the neurodegenerative and healthy groups, respectively, which were below the minimal clinically important difference of 6.0 cm. Step width variability mean absolute errors were 1.5 cm and 0.8 cm for neurodegenerative and healthy groups, respectively. Data augmentation significantly improved accuracy performance in the neurodegenerative group, likely because they exhibited larger variations in walking kinematics as compared with healthy subjects. These results could enable clinically meaningful and accurate portable step width monitoring for individuals with and without neurodegenerative disease, potentially enhancing rehabilitative training, assessment, and dynamic balance control in clinical and real-life settings.

BackgroundWrist-worn inertial sensors are used in digital health for evaluating mobility in real-world environments. Preceding the estimation of spatiotemporal gait parameters within long-term recordings, gait detection is an important step to identify regions of interest where gait occurs, which requires robust algorithms due to the complexity of arm movements. While algorithms exist for other sensor positions, a comparative validation of algorithms applied to the wrist position on real-world data sets across different disease populations is missing. Furthermore, gait detection performance differences between the wrist and lower back position have not yet been explored but could yield valuable information regarding sensor position choice in clinical studies. ObjectiveThe aim of this study was to validate gait sequence (GS) detection algorithms developed for the wrist position against reference data acquired in a real-world context. In addition, this study aimed to compare the performance of algorithms applied to the wrist position to those applied to lower back–worn inertial sensors. MethodsParticipants with Parkinson disease, multiple sclerosis, proximal femoral fracture (hip fracture recovery), chronic obstructive pulmonary disease, and congestive heart failure and healthy older adults (N=83) were monitored for 2.5 hours in the real-world using inertial sensors on the wrist, lower back, and feet including pressure insoles and infrared distance sensors as reference. In total, 10 algorithms for wrist-based gait detection were validated against a multisensor reference system and compared to gait detection performance using lower back–worn inertial sensors. ResultsThe best-performing GS detection algorithm for the wrist showed a mean (per disease group) sensitivity ranging between 0.55 (SD 0.29) and 0.81 (SD 0.09) and a mean (per disease group) specificity ranging between 0.95 (SD 0.06) and 0.98 (SD 0.02). The mean relative absolute error of estimated walking time ranged between 8.9% (SD 7.1%) and 32.7% (SD 19.2%) per disease group for this algorithm as compared to the reference system. Gait detection performance from the best algorithm applied to the wrist inertial sensors was lower than for the best algorithms applied to the lower back, which yielded mean sensitivity between 0.71 (SD 0.12) and 0.91 (SD 0.04), mean specificity between 0.96 (SD 0.03) and 0.99 (SD 0.01), and a mean relative absolute error of estimated walking time between 6.3% (SD 5.4%) and 23.5% (SD 13%). Performance was lower in disease groups with major gait impairments (eg, patients recovering from hip fracture) and for patients using bilateral walking aids. ConclusionsAlgorithms applied to the wrist position can detect GSs with high performance in real-world environments. Those periods of interest in real-world recordings can facilitate gait parameter extraction and allow the quantification of gait duration distribution in everyday life. Our findings allow taking informed decisions on alternative positions for gait recording in clinical studies and public health. Trial RegistrationISRCTN Registry 12246987; https://www.isrctn.com/ISRCTN12246987 International Registered Report Identifier (IRRID)RR2-10.1136/bmjopen-2021-050785

OBJECTIVE:The objective of this study was to identify gait alterations related to worsening knee pain and worsening physical function, using machine learning approaches applied to wearable sensor-derived data from a large observational cohort. METHODS:Participants in the Multicenter Osteoarthritis Study (MOST) completed a 20-m walk test wearing inertial sensors on their lower back and ankles. Parameters describing spatiotemporal features of gait were extracted from these data. We used an ensemble machine learning technique ("super learning") to optimally discriminate between those with and without worsening physical function and, separately, those with and without worsening pain over two years. We then used log-binomial regression to evaluate associations of the top 10 influential variables selected with super learning with each outcome. We also assessed whether the relation of altered gait with worsening function was mediated by changes in pain. RESULTS:Of 2,324 participants, 29% and 24% had worsening knee pain and function over two years, respectively. From the super learner, several gait parameters were found to be influential for worsening pain and for worsening function. After adjusting for confounders, greater gait asymmetry, longer average step length, and lower dominant frequency were associated with worsening pain, and lower cadence was associated with worsening function. Worsening pain partially mediated the association of cadence with function. CONCLUSION:We identified gait alterations associated with worsening knee pain and those associated with worsening physical function. These alterations could be assessed with wearable sensors in clinical settings. Further research should determine whether they might be therapeutic targets to prevent worsening pain and worsening function.