🧠 neurology

Accelerometry-Derived REM Sleep Behavior Disorder Predicts Future Parkinson's Disease in the UK Biobank

This study demonstrates that applying a machine learning classifier to wrist accelerometry data in the UK Biobank effectively identifies individuals with REM sleep behavior disorder who have a significantly elevated, dose-dependent risk of developing Parkinson's disease, offering a scalable and superior alternative to questionnaire-based screening for prodromal risk enrichment.

Original authors: Mejia, G. R., Brink-Kjaer, A., Liu, L., Zhou, L., Gunter, K., Ryu, K. H., Wickramaratne, S. D., Parekh, A., Gan-Or, Z., During, E.

Published 2026-07-06

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Mejia, G. R., Brink-Kjaer, A., Liu, L., Zhou, L., Gunter, K., Ryu, K. H., Wickramaratne, S. D., Parekh, A., Gan-Or, Z., During, E.

Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). ⚕️ This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer

Imagine your body has a tiny, invisible alarm system that starts ringing years before a major storm (Parkinson's disease) actually hits. For a long time, doctors have only been able to hear the storm when the wind was already howling and the trees were falling (motor symptoms like shaking). But this new study suggests we might be able to hear the first rumble of thunder long before the storm arrives, using a simple wristband.

Here is the story of how the researchers did it, explained in everyday terms:

The "Sleeping Giant" Problem

Parkinson's disease is like a slow-growing weed in a garden. It starts growing underground (in the brain) years before you ever see the leaves pop up above the soil. Scientists know that a condition called REM Sleep Behavior Disorder (RBD) is a very strong sign that this weed is growing. People with RBD act out their dreams—they punch, kick, or yell while sleeping because their brain forgets to "turn off" their muscles during sleep.

However, checking for RBD usually requires a "sleep lab" test (polysomnography), which is expensive, uncomfortable, and impossible to do for millions of people. It's like trying to check every house in a city for a leaky roof by sending a team of roofers to climb up on every single one.

The Wristband Solution

The researchers used the UK Biobank, a massive database of 87,975 people who wore a smartwatch-style device on their wrist for a week. This device recorded their movements while they slept.

Think of the device as a night-time motion detector. The researchers fed this movement data into a computer program (an AI) that had been trained to recognize the specific "dance" of a person with RBD. The AI didn't need to see the person; it just looked at the shaking and jerking patterns on the wrist to give everyone a "RBD Risk Score."

The Big Discovery: The "Tipping Point"

The study found that this wristband score was a crystal ball for the future.

The Low Risk Group: Most people had low scores. They were like people living in a calm neighborhood; the chance of them developing Parkinson's in the next 10 years was very low.
The High Risk Group: A tiny slice of people (the top 1%) had very high scores. These were the people whose wristbands were "shaking" the most during sleep.

The Result: The people in that top 1% group were five times more likely to develop Parkinson's disease over the next decade compared to the low-risk group. It wasn't a gradual increase; the risk jumped up sharply at the very top end of the scale, like a light switch flipping on rather than a dimmer slowly turning up.

Why This Matters (The "Extra Clues")

The researchers wanted to make sure the wristband wasn't just picking up on general "old age" or bad sleep. They checked:

Genetics: They looked at the participants' DNA. Even if someone had a "genetic lottery ticket" for Parkinson's, the wristband still found extra risk that the DNA missed. It's like the DNA tells you the potential for a fire, but the wristband tells you if the smoke is already rising.
Other Symptoms: People with high wristband scores also started showing other early warning signs of Parkinson's, like constipation, depression, or low blood pressure when standing up. The wristband seemed to be catching the "body" of the disease before the "brain" (motor skills) was affected.
Specificity: The wristband was very good at spotting Parkinson's specifically. It didn't just flag people who were getting Alzheimer's or having general health issues. It was tuned to the specific "signature" of Parkinson's.

The "Super-Scanner" Analogy

Imagine you are trying to find a few specific needles in a giant haystack.

Traditional Questionnaires: Asking people "Do you act out your dreams?" is like asking the haystack if it has needles. It's okay, but many people forget or lie, and it misses a lot.
The Wristband: This is like using a metal detector that beeps only when it finds the specific metal of the needle. The study found that this "metal detector" was three times better at finding the needles than just asking people questions.

The Bottom Line

This study shows that a simple, cheap wristband can act as a super-sensor for the earliest stages of Parkinson's disease. It can identify a small group of people who are at very high risk years before they get sick.

Important Note from the Paper:
The authors are careful to say this is not a diagnosis tool yet. You can't wear a watch and get a "Parkinson's" label tomorrow. Instead, this tool is like a highly accurate filter. It helps doctors find the small group of people who should get extra attention, more testing, or be invited to join clinical trials to test new medicines that could stop the disease before it starts. It turns a needle-in-a-haystack problem into a manageable search.

Technical Summary: Accelerometry-Derived REM Sleep Behavior Disorder Predicts Future Parkinson's Disease in the UK Biobank

Problem Statement
Identifying individuals at high risk for Parkinson's disease (PD) years before the onset of classic motor symptoms remains a critical unmet need for enabling early neuroprotective interventions. While idiopathic REM sleep behavior disorder (iRBD) is a highly specific prodromal marker for $\alpha$ -synucleinopathies, current gold-standard diagnosis via polysomnography (PSG) is costly, labor-intensive, and impractical for population-scale screening. Questionnaire-based screening lacks sufficient specificity due to the difficulty of distinguishing true RBD from mimics. There is a need for a scalable, objective method to detect RBD physiology in general populations to enrich cohorts for prodromal PD risk.

Methodology
This prospective population-scale study utilized data from 87,975 participants in the UK Biobank who wore Axivity AX3 wrist-worn accelerometers for 7 days. The study applied an externally trained and validated machine-learning classifier (gradient-boosted decision trees) to the 7-day accelerometry data to generate a continuous, accelerometry-derived RBD risk score. The classifier was trained on PSG-confirmed iRBD cases and controls and was applied without retraining.

Participants were stratified into three risk groups based on the score distribution: Low (0–90th percentile), Intermediate (90–99th percentile), and High (99–100th percentile). The primary outcome was incident PD, identified via linked Hospital Episode Statistics (ICD-10 G20) and UK Biobank algorithmically defined outcomes over a median follow-up of 10 years.

Statistical analyses included:

Survival Analysis: Cox proportional hazards models adjusted for age, sex, BMI, smoking, and alcohol use to estimate hazard ratios (HR) for incident PD across RBD strata.
Dose-Response Modeling: Restricted cubic splines were used to model the non-linear relationship between RBD scores and PD risk.
Genetic Interaction: Analyses were conducted in a genetic subcohort (N=66,524) to assess independence from PD polygenic risk scores (PRS) and to evaluate synergistic effects between RBD scores and PRS.
Prodromal Enrichment: Logistic regression was used to evaluate the association between RBD risk strata and longitudinal prodromal features (e.g., constipation, depression, autonomic dysfunction) in non-converters.
Specificity and Robustness: Associations with Alzheimer's disease and vascular dementia were examined to test specificity. Sensitivity analyses included lagging (excluding cases within 2 years of assessment) and competing-risk adjustments.

Key Results

Prediction of Incident PD: There was a strong, dose-dependent association between accelerometry-derived RBD scores and incident PD. Participants in the High-risk stratum (>99th percentile) exhibited a 4.69-fold increased hazard of developing PD compared to the Low-risk group (HR 4.69, 95% CI 3.21–6.87). The Intermediate group showed a 1.87-fold increased hazard.
Non-Linear Risk Concentration: The risk of PD was not uniform across the score distribution; it accelerated sharply at the extreme upper tail. High-risk participants who converted to PD were diagnosed approximately one year earlier than those in lower strata.
Prodromal Enrichment: Among non-converters, higher RBD scores were associated with a dose-dependent enrichment of prodromal features, particularly depression (OR 6.39 for High vs. Low) and constipation. Baseline cognitive deficits were also observed in higher-risk strata.
Independence from Genetics: The association between RBD scores and PD remained robust after adjusting for PD polygenic risk scores (HR 4.42 for High-risk group after adjustment). Furthermore, RBD scores and PRS demonstrated a synergistic interaction (Interaction HR 1.20), suggesting they capture complementary dimensions of risk (dynamic physiology vs. static genetic liability).
Specificity: The RBD score showed strong specificity for PD, with non-significant associations with Alzheimer's disease and vascular dementia.
Clinical Utility: The High-risk group achieved a positive likelihood ratio (LR+) of 7.91, approximately three times higher than questionnaire-based RBD screening (LR+ 2.8). While individual prodromal markers had weak discriminatory value, the accelerometry-derived score significantly improved model discrimination (C-index 0.784 for RBD alone vs. 0.842 when combined with PRS).

Significance and Claims
The authors claim that wrist-worn accelerometry provides a scalable, passive, and biologically grounded approach for identifying individuals at elevated risk for prodromal PD. The study demonstrates that an accelerometry-derived RBD signature captures a distinct dimension of prodromal risk that is largely independent of, yet synergistic with, genetic susceptibility.

The paper positions this technology not as a standalone diagnostic tool, but as a highly effective "enrichment biomarker" for population screening. By identifying a high-risk subgroup with a positive likelihood ratio of 7.91, this approach could efficiently prioritize participants for longitudinal monitoring, multimodal biomarker assessment, or enrollment in neuroprotective clinical trials. The findings support the feasibility of using wearable-derived physiological signals to address the challenge of identifying prodromal PD in large, unselected populations, potentially bridging the gap between the high specificity of PSG and the scalability required for public health applications.