Expanding Diagnostic Tools for iRBD Through Actigraphy-Based Classifiers

Emmanuel During, MD

Idiopathic REM sleep behavior disorder (iRBD) is a parasomnia in which patients physically act out vivid, often violent dreams during REM sleep. Normally, REM sleep involves temporary muscle atonia, but in iRBD this mechanism is impaired. In recent years, research has discovered that iRBD is one of the strongest early indicators of alpha-synuclein-related neurodegenerative disorders such as Parkinson disease (PD), dementia with Lewy bodies, and multiple system atrophy.

Diagnosing iRBD presents several key challenges–both clinical and logistical. Oftentimes, symptoms for iRBD are unrecognized or underreported, as well as may mimic other sleep disorders, leading to misdiagnosis. In addition, a polysomnography is required for confirmation; however, many sleep labels do not routinely score REM atonia, and access to PSG is limited in some areas due to cost, wait times, or lack of specialized centers. At the 2025 SLEEP Annual Meeting, held June 8-11, in Seattle, Washington, one talk, given by Emmanuel During, MD, focused on the use of a classifier to help diagnose iRBD.

During an associate professor of neurology at the Icahn School of Medicine at Mount Sinai, sat down to discuss his presentation in greater detail, giving insights on the external validation of such classifier. In the interview, he explained how the classifier, originally developed using high-resolution actigraphy data, was successfully adapted to a lower-resolution device and still achieved strong accuracy across a larger, international dataset. During also gave commentary on the therapeutic landscape for iRBD, available biomarkers, the pressing need for longitudinal monitoring solutions for patients with the condition.

NeurologyLive: Can you give an overview and details behind this classifier, how it came about, and just how it operates?

Emmanuel During, MD: Maybe we should go back to what iRBD is. It’s really an acronym—we say RBD for REM Behavior Disorder, with an "i" for idiopathic or isolated. REM Behavior Disorder is a condition that occurs during REM sleep and causes people to move, twitch, and sometimes act out their dreams. We call it isolated when there’s no obvious sign of Parkinson’s disease or dementia, but we know that in 80–90% of cases, it’s a precursor. That means it’s an early stage of disease, and almost all of these people will develop Parkinson’s disease or dementia with Lewy bodies within 10 to 20 years.

Why is it important to diagnose it? Obviously because of the very high risk—it’s an early marker. But we’re not very good at diagnosing it. It often takes years, and most patients aren’t being diagnosed because they don’t suspect anything wrong with their sleep. The current diagnostic method is a sleep study in the lab. It’s costly, inefficient, and we end up missing many patients. It’s hard to interpret and not scalable.

So the effort started a few years ago when we realized we needed a better, more scalable method—maybe something done at home. It made sense to look into devices people already wear, like sleep trackers, and see if those could detect subtle nighttime movements. That’s how it all started. We published early data showing this approach was feasible and surprisingly accurate, but we used a specific device with a specific dataset—the Stanford dataset, which was small but encouraging.

Since then, the main questions became: 1) Would it work in other datasets? 2) Would it work with a different device? That’s crucial if we want to scale this to diagnostic use. In this new study, we used data from the Hong Kong Sleep Center—a big dataset: 200 people with RBD, over 100 without. Every RBD diagnosis was confirmed in the sleep lab using gold-standard methods.

The twist was that they used a completely different device: the Philips Actiwatch, which is a low-resolution, widely-used actigraph. Our original work used the ActiGraph device, which is high-resolution and common in research. So we had to match how each device measured movement—basically, compare apples to apples. We asked volunteers to wear both devices for a few days and created a conversion between the two.

Once we had that, we tested whether our classifier, which was built using the ActiGraph data, could still predict RBD using the Philips data. And the results were exciting—though slightly less accurate, we still achieved 86% accuracy. That means our original model works across both populations and devices. That’s a big step forward.

What are the next steps to advance this research?

There are a few next steps. First, improving the model. The one we used is the original version—a basic machine learning model using engineered features. It had good accuracy, but we could improve that with more advanced techniques like deep learning, which might help detect more subtle patterns.

Second, the dataset was still relatively small—84 people total, evenly split between cases and controls. Imagine what we could do with a dataset five or ten times that size, with more diverse demographics. That would help create a more generalizable model—one that works better across different populations, genders, and age groups.

Third, we should try using additional devices—especially consumer wearables. There are many popular ones now that collect the kind of data we need. So far, we’ve only used movement data, but adding another signal, like pulse, could help. Heart rate variability is a known marker of neurologic health. A decrease in HRV is seen in Parkinson’s and dementia with Lewy bodies. If we could combine HRV with movement data, I think we could make these models even stronger.

What other types of research would you like to see from the field to advance understanding of iRBD, especially its link to neurodegenerative disorders?

Once we diagnose patients, we don’t yet have great tools to monitor them over time. And remember, this is not a rare condition—about a million people in the U.S. may have idiopathic RBD, and most of them don’t know it. Many will develop Parkinson or dementia with Lewy bodies. We know certain things help—like physical activity, avoiding cardiovascular risks—but as new neuroprotective drugs emerge, we’ll face questions.

Should all iRBD patients start therapy? Or only those showing signs of disease progression? There’s a need for scalable, low-cost monitoring tools. Wearables can help, but digital health overall—like apps, periodic cognitive tests, and patient-reported outcomes—can play a role too. That’s an area I hope to see move faster. It’s also something I’m personally working on.

What does the therapeutic landscape look like for iRBD?

It’s difficult. There are a lot of unknowns. At conferences, we talk about how some drugs that help in late-stage Parkinson’s or dementia might help earlier, but there are many options. Neuroinflammation likely plays a role, so drugs that reduce inflammation or alpha-synuclein deposition are of interest. For example, doxycycline has both effects. Ambroxol is another—it improves lysosomal function. GLP-1 drugs are also being looked at, though data is mixed. We don’t have a clear answer yet, but the hope is that starting treatment earlier could be more effective.

Are there any notable biomarkers for iRBD that are emerging?

Yes. Skin biopsy is a big one. It’s not new, but it’s clinically available now. It can detect alpha-synuclein in peripheral nerves with high specificity and about 80% sensitivity. That’s huge—it’s a game-changer.

Beyond that, dynamic biomarkers are the future—tools that tell you whether someone’s risk of progression is high, low, or intermediate. Another area that’s emerging is co-pathology. Many RBD patients have co-existing Alzheimer pathology, and that seems to predict faster cognitive decline. Blood-based Alzheimer biomarkers are now available and are starting to be studied in RBD populations.

Transcript edited for clarity. Click here for more SLEEP 2025 coverage.