SLEEP 2026 | Two-stage questionnaire and actigraphy-based screening approach for iRBD

We did two stages. The first stage was the questionnaire. This questionnaire had four parts. The first was dream enactment. That’s the core symptom of iRBD. We also included three other symptoms, hyposmia, autonomic instability, and constipation. So that was the first part of this screening pipeline. Then there was another stage, which was actigraphy. So this was a wrist-worn device worn for around 14 nights. And what it was collecting was the movement of the patient over their nights. And we ran this raw data through a machine learning algorithm to create movement features. And with both the first stage and second stage, we did another machine learning algorithm to classify the patients based on their results. So in combination, what it was was a questionnaire. And if they were positive in the first stage, then they would move on to the second stage, which is a more objective measure. And then they would do the actigraphy. And we would take the combined results. And that’s where we had our performance metrics. So what we found was a sensitivity of 73% and a specificity of 100%, which means that we were able to remove all the false positives from the population. And the patients that we trained all these models on and did the study on, it was a retrospective cohort, and it was 396 patients across five cohorts in two centers, Mount Sinai and Stanford. And the overall screening was done on a subset of that population where we had both actigraphy and questionnaire data. As we know, consumers are currently using wearable devices all throughout the world. These devices that are household names, Apple Watch, Fitbit, Aura, the list goes on. They are collecting the information that can be used for what we did in our study, just simple accelerometry data. So with this widespread adoption of these devices, what’s really missing is the algorithm that would be detecting the condition, which we have started the work on and is constantly improving as we find more participants and more training data. And also the framework, how these diagnoses will be made and distributed. So what we can imagine is a world where you’re taking a questionnaire, maybe it’s through your doctor’s office, maybe it’s through some app, some form, and then with the positive on the questionnaire, your device that you already own would screen you, and it would do what our study did, and if you were positive, then you’d be referred to your health system. What work is to be done is just improving the algorithm and then also finding a responsible way to deploy this technology as the diagnosis of this condition is something that should not be taken lightly due to its conversion into such serious neurodegenerative diseases that has been seen at levels of over 80% in some studies to either Parkinson’s disease or multiple system atrophy. So what that one might look like is something that would be needed to be developed over time you know maybe patients will you know consent will be made aware of what they are being screened for but I think that as the disease-modifying treatment for iRBD and those conditions that will eventually be seen improves, there will be an ethical responsibility to be screening these people and diverting them to treatment before they are even aware of their own symptoms. And I think this is where this technology really has a lot to offer.

This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.