Leveraging long-term smartwatch data to inform Parkinson's disease progression, subtypes, and risk

Use of digital sensors to passively collect long-term, longitudinal data offers a step change in our ability to monitor Parkinsons disease (PD). However, to date the evaluation of long-term digital sensor data has been neglected in favour of evaluating short-term data collected in controlled settings. To address this, we combined longitudinal clinical and biological assessment data from the Parkinsons Progression Marker Initiative (PPMI) cohort with long-term (mean: 485 days) at-home digital monitoring data collected with the Verily Study Watch. We then derived digital timeseries components leveraging the long-term monitoring of the PPMI. We found three key findings: Firstly, that these digital timeseries components correlated with the rate of progression of motor (r = 0.23, p-value = 8.5x10-3, r = 0.26, p-value = 2.2x10-3) and autonomic symptoms (r = -0.23, p-value = 8.2x10-3), impairments in daily living (r = 0.26, p-value = 2.5x10-3), increase in medication requirements and complications (r = -0.25, p-value = 4.2x10-3), and rate of increase in cerebrospinal fluid (CSF) tau (ptau: r = 0.28, p-value = 2.6x10-3; ttau: r = 0.34, p-value = 1.2x10-4). Second, we derived digitally informed subtypes of PD and found higher similarity with CSF (0.35) and DaTscan (0.35) subtypes than has been found for previously published subtypes (CSF: 0.31{+/-}0.01, DaTscan: 0.31{+/-}0.02). Finally, we showed that long-term digital monitoring can inform PD risk and sensitively detect individuals with probable prodromal PD. Our findings highlight the wealth of application areas for digital sensors in PD research.