By Dr Maria Gabriella Ceravolo
Parkinson’s disease (PD) is a complex, progressive neurological condition whose symptoms fluctuate not only across disease stages, but also from hour to hour in daily life. Traditionally, clinicians have relied on episodic in-clinic visits to assess motor symptoms, disease progression, and treatment response. While invaluable, these snapshots capture only a fraction of patients’ real-world functioning.
Over the past decade, wearable sensors—particularly inertial measurement units (IMUs), accelerometers, and gyroscopes—have emerged as powerful tools for continuous, home-based motor monitoring. These technologies are now widely used in Parkinson’s disease research and have demonstrated high feasibility, reliability, and patient compliance. Importantly, data collected in everyday environments often align closely with clinical assessments, highlighting their potential to complement traditional evaluations rather than replace them
Beyond Monitoring: The Promise of Predictive Algorithms
While continuous monitoring is already a major advance, its true transformative potential lies in what happens next: the integration of large-scale wearable data with predictive algorithms.
Rehabilitation is, by definition, person-centered. Effective personalization requires prediction and stratification—anticipating how individuals will progress, respond to treatment, or adhere to recommendations. Despite this, well-developed and validated predictive models for rehabilitation remain scarce across many health conditions and outcome domains, including Parkinson’s disease.
Wearable sensors generate rich, longitudinal streams of information on gait, tremor, bradykinesia, physical activity levels, and motor fluctuations. When combined with clinical observations, patient-reported outcomes, and contextual data (such as medication timing or daily routines), these datasets create an unprecedented opportunity for data-driven insight. Predictive algorithms based on machine learning (ML) can move us beyond describing what has already happened to anticipating what is likely to happen next.
Learning from Europe: The PREPARE Rehab Project
This vision is directly aligned with the goals of PREPARE Rehab, a HaDEA–Horizon Europe project bringing together 20 partners from nine countries. PREPARE addresses one of the core limitations of modern rehabilitation research: the lack of robust, generalizable prediction models.
The project applies advanced machine learning techniques to nine large-scale, real-world patient datasets, covering multiple health conditions and rehabilitation outcomes. Crucially, these data are analyzed in a federated manner, meaning that sensitive patient information does not need to be centralized—an approach that directly addresses ethical, legal, and privacy concerns.
As a proof of principle, PREPARE is developing a platform for sharing model results using the open-science EHDEN ecosystem and the Observational Medical Outcomes Partnership (OMOP) Common Data Model (CDM). This standardization enables predictive and stratification models to be reproducible, scalable, and transferable across healthcare systems. The resulting models will then be validated through demonstration studies across nine health conditions, setting a new benchmark for predictive rehabilitation research.
Advancing Our Understanding of Parkinson’s Disease Progression
Parkinson’s disease progression is highly heterogeneous. Some patients experience rapid functional decline, while others remain relatively stable for years. Continuous sensor data collected over weeks, months, or even years can reveal subtle motor changes that are invisible during brief clinic visits.
Predictive models—such as those envisioned within initiatives like PREPARE—can help:
- Identify early digital markers of disease progression,
- Detect transitions between disease stages sooner,
- Characterize individual progression trajectories rather than relying on population averages.
This shift from retrospective assessment to forward-looking prediction has the potential to fundamentally reshape clinical decision-making in Parkinson’s disease.
Measuring Real-World Adherence and Lifestyle Behaviors
Lifestyle recommendations—such as regular physical activity, adherence to exercise programs, or consistent engagement with rehabilitation—are increasingly recognized as essential components of Parkinson’s disease management. Yet adherence in daily life is notoriously difficult to measure.
Wearable sensors provide an objective window into real-world behavior. When analyzed using predictive and stratification algorithms, these data can help clinicians and researchers:
- Quantify adherence to prescribed activity and rehabilitation programs,
- Understand how symptoms, context, and disease stage influence compliance,
- Identify patients at risk of disengagement early.
This capability directly supports the person-centered rehabilitation paradigm promoted by PREPARE, where interventions are adapted not only to diagnosis, but to individual behavior and response patterns.
Toward Personalized Rehabilitation and Treatment
Perhaps the most compelling implication of predictive analytics is true personalization. Parkinson’s disease is not a one-size-fits-all condition—and rehabilitation should not be either.
By integrating wearable sensor data, clinical observations, and large-scale predictive models, future care pathways could support:
- Adaptive rehabilitation programs that respond to daily performance and fatigue,
- Stratified treatment strategies tailored to predicted trajectories,
- More responsive adjustments to therapy and medication based on real-world motor behavior.
In this framework, rehabilitation becomes a dynamic, learning system—continuously informed by data and refined through validated prediction models, as exemplified by the PREPARE approach.
Challenges and the Road Ahead
Significant challenges remain, including model interpretability, clinical integration, standardization, and equitable access to technology. However, initiatives like PREPARE Rehab demonstrate that these challenges can be addressed through federated data analysis, open standards, and rigorous validation across diverse populations.
Wearable sensors have already proven their value as reliable tools for real-world motor assessment. Predictive algorithms—built on large, standardized, and ethically managed datasets—represent the next critical step toward understanding Parkinson’s disease progression, improving adherence, and delivering truly personalized rehabilitation.
The future of Parkinson’s disease care is not only connected—it is predictive, federated, and person-centered.
.
Posted by: Nayara Güércio (Trilateral Research)