Overview¶
Early detection of Parkinson’s disease (PD) remains one of the most critical challenges in neurodegenerative research. In this study, we demonstrate that free-living wrist accelerometry, collected passively in everyday life, can detect prodromal gait changes years before clinical diagnosis.
This work leverages large-scale population data and rigorously validated machine learning pipelines to bridge the gap between wearable sensing and translational neurology.
Study Design¶
The analysis was conducted in two phases:
| Phase | Dataset | Objective |
|---|---|---|
| Phase 1 | Capture-24 (n=151) | Validate walking detection accuracy |
| Phase 2 | UK Biobank (n≈70,000) | Assess association between gait features and future PD |
Walking bouts were detected from raw wrist accelerometry, followed by extraction of clinically interpretable gait features.
Key Gait Features¶
- Arm swing amplitude (rigidity marker)
- Cadence (compensatory stride shortening)
- Jerk (movement smoothness)
These features were aggregated over stable walking periods ≥30 seconds.
Results¶
Predictive Performance¶
Survival random forest models achieved strong discrimination for future PD risk:
| Metric | Value |
|---|---|
| Time-dependent AUROC (Year 9) | 0.75 |
| AUPRC | 0.036 |
| Strongest predictor | Arm swing variability |
Spline analyses revealed non-linear risk relationships, with sharply increased hazard at low gait feature values.
Clinical Implications¶
This study demonstrates that:
- Single-wrist wearables can detect prodromal motor changes
- Only minutes of walking data per day are required
- Passive monitoring enables scalable population screening
These findings support the use of wrist accelerometry as a non-invasive, low-burden screening tool for early neurodegenerative risk stratification.
References¶
- Zhou et al., Detection of Gait Alterations Related to Parkinson’s Disease Prior to Diagnosis, SSRN Preprint.