Accuracy and Reliability of Remote Shoulder Motion Capturing Methods: A Systematic Review and Meta-Analysis

Background:
The COVID-19 pandemic accelerated the demand for remote assessment tools in rehabilitation, spotlighting the need for accurate and reliable technologies to measure shoulder range of motion (ROM) outside of clinical environments. Emerging tools such as smartphone apps, wearable sensors, and markerless motion capture systems are increasingly being adopted, yet their accuracy and reliability compared to reference standards remains unclear.

Objective:
To systematically evaluate the accuracy and reliability of existing remote shoulder ROM measurement technologies, quantify measurement bias, and assess their agreement with reference standards.

Methods:
A systematic review and meta-analysis was conducted on 26 studies evaluating remote ROM measurement tools. Pooled mean bias (in degrees) was calculated as the primary effect size for agreement, with reliability assessed using intraclass correlation coefficients (ICCs). Subgroup analyses were performed by motion type, technology category, population health status, and data acquisition method. Risk of bias was assessed using the QUADAS-2 tool.

Results:
Remote measurement methods showed a small but consistent overestimation of ROM compared to reference standards (pooled mean bias = 2.63°, 95% CI: 1.52°, 3.74°), particularly in flexion, internal rotation, and external rotation. No significant bias was observed in abduction or extension. Both inertial measurement unit (IMU) and non-IMU technologies demonstrated comparable levels of overestimation. Pathological populations exhibited greater variability (bias = 4.33° vs. 2.37° in healthy subjects). Self-measurements showed lower and non-significant bias compared to assessor-guided methods. Reliability was generally high, especially for test-retest assessments (ICCs > 0.90), though more variable in inter-rater and pathological settings.

Conclusion:
Remote technologies for assessing shoulder range of motion show generally small differences compared to reference standards, but performance is inconsistent, with substantial heterogeneity and variable reliability across studies. Although average error is often below reported minimal clinically important differences (MCID), the wide variability observed suggests these tools may not reliably detect small but clinically meaningful changes. As such, they may have value for tracking overall trends in shoulder function, but their use in clinical decision-making—particularly in pathological populations—should be approached with caution.