Validation of Actigraphy for Identifying Isolated REM Sleep Behavior Disorder

Overview

This multicenter study validated the use of wrist actigraphy combined with machine learning to detect isolated REM sleep behavior disorder (iRBD) across diverse populations and devices. The sleep-based actigraphy model demonstrated robust performance (AUC 0.838–0.865) across four international cohorts and multiple actigraphy devices, while rest-activity rhythm features were less consistent. Incorporating prodromal synucleinopathy symptoms further enhanced screening specificity and positive predictive value.

Background

Isolated REM sleep behavior disorder (iRBD) is a parasomnia characterized by loss of muscle atonia during REM sleep, leading to potentially injurious motor behaviors. It affects approximately 1–1.5% of the general population and is often a prodrome to synucleinopathies such as Parkinson’s disease. Diagnosis traditionally relies on resource-intensive video-polysomnography, limiting widespread detection. Prior studies demonstrated that high-resolution wrist actigraphy combined with machine learning can accurately identify iRBD, but generalizability across devices and populations remained untested. This study aimed to validate the actigraphy-based detection model across multiple centers and devices with varying sampling resolutions.

Data Highlights

Key Findings

• The fully automated sleep-based actigraphy model generalized well across four international cohorts totaling 352 iRBD and 258 controls, achieving AUCs between 0.838 and 0.865.
• Rest-activity rhythm features showed variable performance (AUC 0.520–0.818) and were less reliable for iRBD detection.
• Actigraphy devices with lower sampling resolutions (Philips Actiwatch, MicroMini-Motionlogger) performed comparably to high-resolution Axivity AX6 after appropriate data conversion.
• Combining actigraphy with prodromal synucleinopathy features (RBD symptoms, hyposmia, constipation, orthostatic hypotension) improved screening specificity and positive predictive values, with PPVs up to 98.9% for hyposmia.
• Adjusting for real-world iRBD prevalence (~1.5%) indicated PPVs ranging from 6.3% to 100% depending on prodrome used, highlighting the importance of symptom screening.

Clinical Implications

Wrist actigraphy combined with machine learning offers a scalable, non-invasive method for early detection of iRBD across diverse populations and wearable devices. Incorporating key prodromal symptoms enhances screening accuracy, potentially enabling population-level identification of individuals at risk for synucleinopathies. This approach may facilitate timely clinical interventions and enrollment in neuroprotective trials without reliance on costly polysomnography.

Conclusion

This multicenter validation confirms that sleep-based actigraphy models reliably detect iRBD across different devices and populations. When combined with prodromal symptom screening, this method holds promise for scalable, precise population-level iRBD screening.

References

1. International RBD Study Group et al. 2024 -- Validation of Actigraphy for Identifying Isolated REM Sleep Behavior Disorder