A Two-Stage Questionnaire and Actigraphy Screening for Isolated REM Sleep Behavior Disorder in a Multicenter Cohort

This multicenter study demonstrates that a low-cost, two-stage screening protocol combining a brief symptom questionnaire with home-based wrist actigraphy achieves high specificity and good sensitivity for detecting isolated REM sleep behavior disorder, offering a scalable alternative to resource-intensive video-polysomnography.

The Gist

Imagine your brain has a "security guard" that usually turns off your muscles when you dream, so you don't act out your dreams. In a condition called Isolated REM Sleep Behavior Disorder (iRBD), that security guard takes a coffee break. People with iRBD might punch, kick, or yell while dreaming.

Why does this matter? Because iRBD is like a smoke alarm for a very serious fire: it is often the very first sign that a person will eventually develop Parkinson's disease or a related condition called Dementia with Lewy Bodies. Catching this "smoke" early is crucial, but right now, it's very hard to find.

Here is the problem: The only way to be 100% sure you have iRBD is to spend a night in a fancy sleep lab hooked up to wires (like a high-tech spider). This is expensive, scary, and hard to get. On the other hand, just asking people "Do you punch in your sleep?" isn't reliable enough because many people say "yes" but don't actually have the disorder.

The Solution: A Two-Stage "Funnel" Screening

The researchers in this paper came up with a clever, low-cost strategy to find these people without needing a sleep lab immediately. Think of it like a two-stage security checkpoint at an airport.

Stage 1: The "Radar Gun" (The Questionnaire)

First, you don't need a lab. You just need a short, four-question survey.

• The Questions: It asks about dream punching/kicking, loss of smell, constipation, and dizziness when standing up.
• The Analogy: Imagine a radar gun at a speed trap. It's not perfect, but it's fast and cheap. It scans everyone and flags the cars that might be speeding.
• The Result: This stage is designed to be sensitive (catching almost everyone who might have the problem), even if it flags a few innocent people by mistake. The "dream punching" question alone was surprisingly good at this.

Stage 2: The "Body Scanner" (The Watch)

If you pass Stage 1 (meaning you flagged as "maybe"), you move to Stage 2.

• The Tool: Instead of a sleep lab, you wear a simple smartwatch (or a specialized wrist sensor) for two weeks while you sleep at home.
• The Analogy: This is like the body scanner at the airport. It doesn't ask you questions; it watches your body. It looks for tiny, specific muscle twitches that happen only during REM sleep. A computer program (AI) analyzes the watch data to see if your "security guard" is actually asleep on the job.
• The Result: This stage is designed to be specific (making sure we don't flag innocent people). If the watch says "Yes, you have the disorder," it is almost certainly true.

How Well Did It Work?

The researchers tested this on nearly 400 people across different medical centers.

• The "Dream Punch" Question: Caught about 78% of the real cases.
• The Smartwatch: Caught about 80% of the real cases on its own.
• The Combo (The Magic): When they used the Question first, then the Watch for those who flagged, they achieved 100% specificity. This means zero false alarms. If someone passed both stages, they definitely had the disorder. They caught about 73% of the actual cases, which is a great balance for a screening tool.

Why This Is a Big Deal

Think of the current system as trying to find a needle in a haystack by looking at every single piece of hay with a microscope. It's slow and expensive.

This new method is like using a magnet.

1. The Magnet (Questionnaire): Pulls out the metal (potential cases) from the hay.
2. The Metal Detector (Watch): Confirms which pieces of metal are actually the needle.

The Bottom Line

This study shows that we can find people at risk for Parkinson's disease using a simple questionnaire and a smartwatch, without them ever needing to go to a sleep lab immediately. It's a low-cost, scalable way to screen thousands of people, find the ones who need help, and potentially slow down the progression of neurodegenerative diseases before they become severe.

In short: It turns a difficult, expensive medical mystery into a simple, two-step home check-up.

Technical Summary

1. Problem Statement

Isolated REM Sleep Behavior Disorder (iRBD) is a highly specific prodromal marker for synucleinopathies (e.g., Parkinson's disease, Dementia with Lewy Bodies), with over 80% of patients eventually developing neurodegeneration. However, a "diagnostic bottleneck" exists:

• Gold Standard Limitation: Definitive diagnosis requires in-laboratory video-polysomnography (vPSG), which is costly, resource-intensive, and not scalable for population screening.
• Questionnaire Limitations: Current screening questionnaires (e.g., asking about dream enactment) have low positive predictive values (~10%) in community settings due to low disease prevalence and high false-positive rates from mimics (e.g., sleep apnea, periodic limb movements).
• Need: A scalable, low-cost, two-stage screening protocol that combines high sensitivity (to catch cases) with high specificity (to rule out false positives) to enrich cohorts for confirmatory testing.

2. Methodology

Study Design and Cohorts

• Design: Multicenter, retrospective case-control study.
• Participants: 396 total participants (99 confirmed iRBD cases, 297 controls) aged 40–80 without overt neurodegenerative disease.
• Cohorts: Data drawn from five cohorts across two institutions (Icahn School of Medicine at Mount Sinai and Stanford University):
  • Sleep and Healthy Aging Study (SHAS)
  • VascBrain
  • Sinai Sleep Clinic
  • Stanford Sleep Center
  • Stanford Alzheimer's Disease Research Center (ADRC)
• Ground Truth: All iRBD cases were confirmed via overnight vPSG per ICSD-3 criteria.

Data Collection

1. Stage 1: Questionnaire (Symptom Screening)

   • A four-item questionnaire assessed:
     1. Dream Enactment: Validated Innsbruck summary question (kicking/hitting during sleep).
     2. Hyposmia: Reduced smell/taste.
     3. Constipation: Need for straining or laxatives.
     4. Orthostatic Symptoms: Suggestive of hypotension.
   • Responses: No (0), Don't Know (0.5), Yes (1).
   • Administered to 289 participants.

2. Stage 2: Actigraphy (Objective Movement Screening)

   • Device: Wrist-worn triaxial accelerometers (Axivity AX3/AX6).
   • Duration: Participants wore devices continuously; 236 participants completed ≥14 nights.
   • Preprocessing: Automated pipeline distinguished wear vs. non-wear nights based on sleep duration, non-wear hours, and skin temperature.
   • Feature Engineering: 113 candidate features derived from 1-second activity counts, focusing on four primary movement constructs:
     • Mean Motor Activity (MA)
     • Activity Index (AI)
     • Short and Long Immobile Bouts (SIB, LIB)
     • Twitch Activity (TA): A novel feature capturing low-amplitude movements (<0.5–1.5g) surrounded by zero-activity.

Machine Learning Pipeline

• Framework: Fully nested cross-validation (10 outer folds) with participant-level separation to prevent data leakage.
• Algorithms:
  • Questionnaire: XGBoost (compared against Random Forest, LightGBM, Elastic Net).
  • Actigraphy: XGBoost (chosen for high-dimensional tabular data performance).
• Optimization:
  • Hyperparameters tuned via Bayesian optimization (Optuna) within inner folds.
  • Thresholding Strategy:
    • Stage 1 (Questionnaire): Sensitivity-optimized threshold ($\tau_{sens}$) to minimize missed cases.
    • Stage 2 (Actigraphy): Specificity-optimized threshold ($\tau_{spec}$) to minimize false positives.
• Two-Stage Logic: A participant is classified as positive for iRBD only if they pass both Stage 1 and Stage 2.

3. Key Results

Demographics

• Total Cohort: 396 participants (Mean age 64 ± 11; 55% male).
• iRBD vs. Controls: Cases were slightly older, more frequently male, and had lower BMI.

Stage 1: Questionnaire Performance

• Single Item (Dream Enactment): AUC = 0.85; Sensitivity = 77.9%; Specificity = 92.3%.
• Four-Item Model (XGBoost):
  • AUC: 0.86 (95% CI: 0.81–0.91).
  • Feature Importance: Dream enactment was the dominant predictor (20.5 gain), followed by hyposmia (2.3).
  • Sensitivity-Optimized Threshold: Sensitivity 85.3%, Specificity 60.8%.
  • Specificity-Optimized Threshold: Sensitivity 76.8%, Specificity 92.3%.

Stage 2: Actigraphy Performance

• Dataset: 78 iRBD cases, 158 controls (3,819 nights total).
• Performance:
  • AUC: 0.88 (95% CI: 0.84–0.92).
  • Sensitivity: 80.8% (at default threshold).
  • Specificity: 84.8% (at default threshold).
  • Specificity-Optimized Threshold: Specificity 96.2% (Sensitivity 61.5%).

Two-Stage Screening Performance (Subgroup with both data types: 75 cases, 54 controls)

The sequential protocol (Questionnaire $\rightarrow$ Actigraphy) yielded:

• Using Dream Enactment Question Only (Stage 1):
  • Final Sensitivity: 67.6%
  • Final Specificity: 100%
• Using Four-Item Questionnaire (Stage 1):
  • Final Sensitivity: 73.3% (at optimized thresholds)
  • Final Specificity: 100%

4. Key Contributions

1. Novel Two-Stage Framework: Demonstrated a scalable, remote screening pipeline that successfully balances sensitivity and specificity, achieving 100% specificity in the validation cohort while maintaining ~73% sensitivity.
2. Integration of Subjective and Objective Data: Validated that combining a brief, self-administered prodromal questionnaire with passive wrist actigraphy significantly outperforms either modality alone in a case-control setting.
3. Feature Innovation: Introduced and validated "Twitch Activity" as a specific actigraphy feature for detecting REM sleep motor dysregulation, alongside standard metrics.
4. Rigorous Validation: Utilized a fully nested cross-validation framework with participant-level separation, ensuring that performance metrics are robust and not inflated by data leakage.
5. Scalability: Proposed a low-cost solution compatible with consumer wearables, potentially enabling large-scale community screening for synucleinopathy risk.

5. Significance and Limitations

Significance:
This study addresses the critical need for efficient iRBD screening. By achieving 100% specificity, the two-stage protocol effectively filters out false positives (common in single-questionnaire screening), reducing the burden on expensive vPSG resources. It offers a pathway to enrich clinical trials and community cohorts with high-risk individuals before confirmatory testing.

Limitations:

• Retrospective Case-Control Design: Performance metrics (especially sensitivity) may be overestimated compared to real-world community screening where disease prevalence is lower (~1.5%) and phenotypes are more heterogeneous.
• Selection Bias: Recruitment from sleep clinics and research cohorts may have biased the sample toward more severe or aware cases, potentially inflating questionnaire sensitivity.
• Data Completeness: Only a subset of participants (129/396) had complete data for both stages, limiting the statistical power of the final two-stage evaluation.
• Control Definition: Controls who denied dream enactment did not undergo vPSG, meaning some undiagnosed iRBD cases may have been mislabeled as controls, potentially underestimating sensitivity.

Conclusion:
The authors conclude that a two-stage, fully remote strategy is a feasible and highly specific method for identifying iRBD. They recommend prospective validation in community-based populations to confirm generalizability before clinical deployment.