Virtual exam for Parkinson’s disease enables frequent and reliable remote measurements of motor function

Parkinson’s disease (PD) affects millions globally and is expected to increase substantially by 2040. While disease-modifying therapies are being pursued, commonly used clinical endpoints like the MDS-UPDRS suffer from high within-subject variability and low test-retest reliability, and in-clinic assessments offer only a brief snapshot that may not reflect real-world functioning. Geographic and logistical barriers further restrict access to trials. These challenges have motivated development of digital endpoints using wearable sensors to enable objective, frequent, and ecologically valid measurement of motor function at home. Prior work has demonstrated feasibility of supervised, sensor-based assessment and early remote approaches, but long-term engagement, test-retest reliability, and sensitivity to clinically meaningful change (e.g., dopaminergic medication effects) have been underreported at scale. The study evaluates a smartwatch-based Parkinson’s Disease Virtual Motor Exam (PD-VME) to determine feasibility, analytical validity, reliability, and sensitivity to change using data from the Personalized Parkinson Project (PPP).

Prior studies showed that wearable sensors and smartphones can quantify PD motor signs and correlate sensor-derived features with clinical outcomes, especially under supervised conditions. Active assessments measure maximal capacity and complement passive monitoring, which captures real-time expression of signs. However, sustained engagement with remote active assessments has been challenging. Passive monitoring may not capture movement intent, which is crucial for bradykinesia. Systematic evaluation of remote measures’ sensitivity to dopaminergic medication effects and robust reporting of test-retest reliability have been limited. The literature suggests wrist-worn sensors can map to clinical ratings, but clinical scales are subjective and low-resolution, potentially limiting sensitivity; digital measures may detect subclinical motor signs and provide improved reliability with frequent sampling.

Study design: Data come from the Personalized Parkinson Project (PPP; NCT03363468), a prospective, longitudinal, single-center cohort of 520 people with early-stage PD (diagnosed within 7 years). Participants wore a Verily Study Watch up to 23 h/day for three years, passively collecting IMU, gyroscope, photoplethysmography, and accelerometer data. Two analysis subsets were defined. Set 1 (N = 198) had videos of in-clinic MDS-UPDRS Part III tasks scored by two independent raters plus an in-person rating; consensus was the median of the three. Set 2 (N = 370) enrolled in a substudy beginning May 2020 to perform PD-VME both in-clinic (starting July 2020) and remotely. PD-VME design: Patient-centric development defined eight active tasks targeting motor domains: rest/postural tremor; upper extremity bradykinesia (finger tapping, pronation-supination, repeated hand opening/closing); lower extremity bradykinesia (foot stomping); gait; and postural sway. Tasks included seated tests, arm raise/twist (20 s), and a timed up-and-go style gait task (60 s). During tasks, tri-axial accelerometer and gyroscope data were sampled at 200 Hz. Participants could log timing of symptomatic medication intake and were instructed weekly to complete two PD-VME exams on the same day approximately one hour apart: an off-state exam (at their typical worst time, before medication) and an on-state exam (later, at a typical best time), with medication tagging. In-clinic procedures: Participants in Set 2 also performed the PD-VME during clinic visits within 1 hour of the MDS-UPDRS Part III assessment and prior to dopaminergic medication intake, while an assessor observed (without providing feedback). MDS-UPDRS Part III exams were video-recorded for quality control and consensus scoring. Inclusion criteria included PD within the last 5 years, age ≥18 years, ability to consent, and no nickel allergy; comorbidities were not exclusionary. Evaluation metrics: Analytical validity was assessed via Spearman correlations between in-clinic sensor-derived measures and consensus MDS-UPDRS Part III task scores. Reliability was quantified as test-retest intraclass correlation coefficients (ICCs) for remote measurements, focusing on week-to-week and monthly-averaged values. Sensitivity to change was evaluated by comparing remote PD-VME measures between off and on states for participants on dopaminergic medications, using medication tags and timing rules (off: pre-scheduled time and ≥6 h after last medication; on: pre-scheduled time and 0.5–4 h post-tag). Effect sizes (Cohen’s d) and mean differences with 95% CIs were computed. Exams influenced by dopamine antagonists were excluded from on–off analyses. The representativeness of in-clinic versus at-home status was assessed by comparing in-clinic PD-VME values to the distribution (25th–75th percentile) of remote PD-VME measurements within 30–90 days around the clinic visit.

- Engagement and adherence: Median smartwatch wear time was ~21.1–21.2 h/day. The substudy achieved 59% completion of pre-protocol weekly PD-VME sessions over ~70 weeks (22,668 sessions). Dropout rate was 5.4%. In week 1, 80% completed at least one PD-VME; at week 52, 40% completed one PD-VME. - Usability: In clinic, 100% completed tremor and upper-extremity bradykinesia tasks; 98.5% completed gait. Protocol deviations were limited (e.g., 8.2% not placing hands on lap during rest tremor; 3.1% using both arms for arm-wrist task; 3.1% walking with arms crossed; 6.8% sitting during gait). - Analytical validity (in-clinic correlations with MDS-UPDRS Part III): Rest tremor lateral acceleration correlated with MDS-UPDRS 3.17 consensus (Spearman r ≈ 0.707; N = 138). Arm swing during gait correlated with MDS-UPDRS 3.10 (r = −0.46; N = 164). Abstract-level summary reported correlations with consensus ratings for rest tremor (r = 0.62), bradykinesia (r = 0.70), and gait (r = 0.46). - Sensitivity to dopaminergic medication (remote on vs off): Effect sizes ranged from small to moderate. Rest tremor showed a small effect (Cohen’s d ≈ 0.2) with mean difference ≈ 0.10 (95% CI 0.05–0.1). Across domains, abstract reported Cohen’s d = 0.19–0.54. - Reliability (test–retest ICC for remote measures): Weekly ICCs were good (e.g., rest tremor ≈ 0.71; gait arm-swing ≈ 0.743). Monthly-averaged ICCs were good-to-excellent (rest tremor ≈ 0.90; gait ≈ 0.75), with overall monthly ICC range 0.75–0.96. - In-clinic vs at-home representativeness: In-clinic PD-VME values fell within the interquartile range (25th–75th percentile) of each participant’s remote distribution only 39–41% of the time, indicating clinic exams often did not reflect typical at-home status.

Frequent, unsupervised smartwatch-based active assessments are feasible for people with PD and produce high-quality data. Engagement and wear-time were robust over prolonged periods, supporting scalability for longitudinal studies. The PD-VME’s digital measures demonstrated analytical validity via moderate-to-strong correlations with clinician consensus ratings, while frequent remote sampling improved test-retest reliability compared with traditional clinic-based scales. Sensitivity to dopaminergic medication confirmed responsiveness to clinically meaningful changes. Importantly, clinic-based assessments often failed to represent patients’ typical at-home condition, underscoring the value of ecologically valid, repeated measurements. The findings suggest that remote digital measures can enhance statistical power and potentially reduce the sample sizes required for trials assessing therapeutic effects or disease progression. The discussion also highlights limitations of subjective, low-resolution clinical ratings and shows that sensors may detect subclinical motor signs (e.g., tremor) not captured by visual exams, motivating a shift toward objective, responsive digital endpoints.

A smartwatch-based, self-guided virtual motor exam for PD enables frequent, reliable remote quantification of tremor, bradykinesia, and gait impairment. The system achieved high engagement, strong analytical validity, good-to-excellent test–retest reliability, and sensitivity to dopaminergic medication. Because clinic assessments often do not reflect at-home status, repeated remote measures provide a more accurate depiction of real-life motor fluctuations. These capabilities can increase power and reduce sample sizes for interventional and progression studies. Future work should expand to additional motor and non-motor domains, refine placement/asymmetry considerations, replicate findings in independent and newly diagnosed cohorts (including effects of dyskinesia), and leverage longer-term follow-up to assess sensitivity to disease progression.

- Device placement and asymmetry: The watch was worn unilaterally despite asymmetric PD symptoms, potentially affecting validity, especially for gait-related measures and side-specific motor signs. - Domain coverage: The study focused mainly on key motor domains; PD is multifaceted. Additional analyses leveraging other collected signals (e.g., PPG, EDA) are needed to assess broader motor and non-motor symptoms. - Generalizability: Results stem from a single-center cohort; replication in different populations, including newly diagnosed individuals and those with medication-induced dyskinesia, is needed. - Video quality constraints: Some in-clinic video recordings were insufficient for confident consensus ratings, reducing N for certain analyses. - Ongoing follow-up: Not all participants had completed the study at time of reporting; longer-term analyses are needed to confirm sensitivity to progression.