Wearable Sensors in Gait Assessment for Parkinson Disease and Stroke in Real-World Environments: Systematic Review and Meta-Analysis

Background: Real-world gait assessment has gained momentum in populations with walking impairments, offering insights beyond standardized tests and supporting the integration of remote monitoring into clinical care. However, its potential remains limited by the lack of validated population- and context-specific digital biomarkers. Objective: The primary objective was to summarize and critically evaluate the current state of real-world gait assessment using wearable sensors in Parkinson disease (PD) and stroke, addressing methodological approaches, sensor configurations, validation strategies, and outcome reporting. The secondary objective was to quantify methodological heterogeneity by reporting pooled means with 95% confidence intervals (CIs) and prediction intervals (PIs) of commonly reported gait parameters. Methods: PubMed, Scopus, Web of Science, and Embase were searched for English-language studies published up to April 20, 2026. Eligible studies used wearable sensors to assess gait quality parameters in real-world settings in at least 5 individuals with PD or poststroke. Laboratory- or rehabilitation-only studies, non-peer-reviewed articles, abstracts, and studies before 2014 were excluded. Methodological quality was assessed with a checklist adapted from Hubble et al. Gait features available in at least 5 reports were included in a meta-analysis using a random-effects model with Hartung-Knapp-Sidik-Jonkman adjustment, deriving 95% CIs and PIs to characterize the pooled estimate and its dispersion. Results: Of 2489 records, 43 reports were included: 37 on PD (2774 participants; mean age 67.89, SD 8.68 years) and 6 on stroke (217 participants; mean age 63.83, SD 11.84 years). Methodological quality was high, but external validity was consistently the weakest domain in both populations. Sensor configurations most commonly consisted of a lower back accelerometer. In total, 13 distinct walking bout definitions were identified, whereas 17 reports provided none, indicating substantial terminological heterogeneity. In PD, 4 gait parameters met the meta-analysis threshold, each showing a PI far wider than its CI, precluding generalization to future settings. For example, gait speed pooled at 0.84 m/s (95% CI 0.77-0.91; 95% PI 0.56-1.12), and number of steps per day at 7202 (95% CI 3975-10,429), the latter with a PI extending below zero, a mathematically impossible range reflecting extreme dispersion. No stroke parameter was reported consistently enough to permit meta-analysis, underscoring a critically underdeveloped evidence base. Conclusions: Unlike previous reviews focusing on single diseases or specific hardware, this review contrasts PD and stroke to highlight shared methodological challenges. Real-world gait assessment in PD is advancing but fragmented, whereas the stroke evidence base remains sparse. The observed variance is compounded by procedural and algorithmic inconsistencies that cannot currently be separated from genuine clinical differences. The methodological groundwork established in PD offers a foundation for early adoption of standardized terminology, validation procedures, and core outcomes, essential for transitioning wearable sensors from exploratory tools to reliable clinical and remote-monitoring instruments.