Inter- and Intra-Rater Reliability of a Wearable Inertial Sensor for Spatiotemporal Gait Assessment in Stroke Survivors
Inter- and Intra-Rater Reliability of a Wearable Inertial Sensor for Spatiotemporal Gait Assessment in Stroke Survivors
Background and Objective:
Stroke frequently impairs gait functionality, leading to significant limitations in functional independence and quality of life. Objective gait analysis is essential for evaluating clinical status and rehabilitation progress. While laboratory-bound, non-wearable systems are considered the gold standard, their high cost, complex infrastructure, and requirement for specialized personnel limit their routine implementation in daily clinical practice. Portable inertial measurement units have emerged as a practical, low-cost alternative for motion analysis outside traditional laboratory settings. However, evidence regarding specific commercial devices in pathological populations remains limited. This study aims to evaluate the inter-rater and intra-rater reliability of the commercially available Baiobit® lumbar-mounted inertial sensor for measuring spatiotemporal gait parameters in stroke survivors.
Methodology:
A prospective, single-group, repeated-measures design was conducted at a specialized neurorehabilitation clinic (Neuron Clinic, Madrid). A convenience sample of 22 adults diagnosed with subacute or chronic stroke who were able to ambulate independently was included. Gait assessments were performed by two trained evaluators across two separate sessions spaced 24 to 48 hours apart. During each session, the Baiobit® sensor was secured to the participants' sacral region (S1-S2) using an adjustable belt. Participants performed three 10-meter walking trials at their comfortable, self-selected pace for each evaluator. Reliability was quantified using Intraclass Correlation Coefficients (ICC), Standard Error of Measurement (SEM), and Minimal Detectable Change (MDC).
Expected Outcomes:
It is hypothesized that the Baiobit® inertial sensor will demonstrate adequate relative and absolute inter- and intra-rater reliability across spatiotemporal gait parameters, supporting its viability as an objective monitoring tool in routine stroke rehabilitation clinical settings.
Introduction and Rationale Stroke survivors commonly experience sensorimotor sequelae that heavily compromise walking patterns, typically presenting reduced speed, lower cadence, shortened step length, and prolonged double-support time. Objective quantification of these spatiotemporal parameters is critical to direct therapeutic goals and evaluate rehabilitation outcomes. To overcome the financial and logistical barriers of traditional motion laboratories, wearable inertial measurement units (IMUs) are increasingly proposed due to their portability and ease of use. However, substantial heterogeneity in sensor configurations and algorithms necessitates population-specific metric validation. The Baiobit® system is an emerging commercial IMU that integrates a triaxial accelerometer, gyroscope, and magnetometer operating at 200 Hz to estimate gait metrics via internal event-detection algorithms. To date, no study has established its measurement consistency and repeatability specifically for stroke survivors.
Study Population and Eligibility Participants were recruited via convenience sampling from the outpatient database at Neuron Clinics in Madrid, Spain. To ensure safety and protocol compliance, strict eligibility criteria were applied: Inclusion Criteria: (1) Adults aged 18 to 85 years; (2) a confirmed diagnosis of ischemic or hemorrhagic stroke with a time since onset $> 1$ month; (3) ability to ambulate independently (Functional Ambulation Categories FAC > 4) regardless of foot-up orthosis use; (4) capacity to walk 10 meters without stopping and repeat the trial six times across the protocol; and (5) cognitive ability to understand instructions and provide written informed consent. Exclusion Criteria: (1) Medical instability or acute conditions comprising gait safety; (2) concomitant neurological conditions other than stroke; (3) severe musculoskeletal disorders interfering with locomotion; (4) visual or auditory deficits severely impairing balance; (5) recent use of medications affecting gait or postural control (e.g., benzodiazepines, antipsychotics, or opioids); and (6) pregnancy. Randomization and Blinding Protocols To control for sequence and order biases, a 1:1 block randomization scheme was generated using a digital tool (randomizer.org) to determine the evaluation sequence between the two raters. Blinding was coordinated by a third independent researcher who managed the logistical workflows, handled independent password-protected software accounts for each rater, and exported raw data to a masked master database. This design strictly prevented the two evaluators from assessing participants simultaneously or accessing each other's data records.
Experimental Procedures The data collection protocol spanned two distinct evaluation sessions scheduled 24 to 48 hours apart to minimize clinical fluctuations while preventing recall bias. Familiarization: In the first session, after formal enrollment and collection of sociodemographic data, participants were familiarized with the environment, and the 10-meter walkway was demonstrated. Sensor Placement: The Baiobit® sensor was securely mounted on the sacral region corresponding to the S1-S2 vertebrae using the manufacturer's adjustable elastic belt. Data Acquisition: Participants performed three 10-meter walking trials at their comfortable, self-selected pace per evaluator. The first assigned evaluator positioned the sensor, acquired the recordings, and completely removed the device. Rest Period: A mandatory 10-minute seated rest period was enforced between evaluators to mitigate fatigue effects. Replication: The second evaluator repeated the exact preparation, placement, and trial recording steps. This identical sequence was replicated during the second session under unvaried environmental conditions.
Outcome Measures and Biomechanical Variables Spatiotemporal gait variables were recorded bilaterally by the sensor software algorithms, capturing: gait speed (m/s), cadence (steps/min), step and stride lengths (m), percentages (%) of stance and swing phases, single-leg and double-leg stance durations, and propulsion metrics. Statistical Analysis PlanData normality was verified using the Shapiro-Wilk test. Continuous demographics are reported as mean and standard deviation (SD). Relative reliability was analyzed using Intraclass Correlation Coefficients (ICCs) with 95% confidence intervals (95% CIs). A two-way mixed-effects consistency model was selected for intra-rater reliability, and a two-way random-effects absolute agreement model was utilized for inter-rater reliability. Absolute reliability and precision were quantified calculating the Standard Error of Measurement. The Minimal Detectable Change was calculated to define the true clinical change threshold. Bland-Altman plotting was performed to identify systematic biases and calculate the limits of agreement (LoA).
Inclusion Criteria:
Exclusion Criteria: