JMIR Mhealth Uhealth. 2026 Jun 24;14:e69273. doi: 10.2196/69273.
ABSTRACT
BACKGROUND: Patients with Parkinson disease (PD) along with subjective cognitive decline (PD-SCD) are considered an intermediate status between those with normal cognition (PD-NC) and those with mild cognitive impairment (PD-MCI). Wearable digital monitoring technologies and machine learning models offer significant potential for assessing cognitive impairment in patients with PD.
OBJECTIVE: We aimed to evaluate the utility of wearable technology and machine learning for identifying ordinal cognitive stages (OCS) in PD based on timed up and go (TUG) tests (including single-task TUG [TUGst] and dual-task TUG [TUGdt]).
METHODS: Patients with PD along with SCD, MCI, and NC were recruited in a movement disorder clinic. Patients performed TUGst and TUGdt gait trials wearing a motor function and motor symptom quantitative assessment system. In total, 209 kinematic parameters were synthesized for individual TUG to illustrate patients’ motion profiles. We constructed dual-task cost parameters (DTC), describing the magnitude of the effect of the cognitive challenge on motion performance. Covariate-adjusted ordered logistic regression was used to compare parameter differences among 3 groups. Multiple machine learning models were used to classify the participants into 3 cognitive impairment levels, with features being selected based on P values from intergroup statistical tests. The total population was randomly divided into a training set and an independent validation set in a 7:3 ratio, and 10-fold cross-validation was used in the training set. Furthermore, this study used permutation importance and Shapley Additive Explanations (SHAP) analysis (including summary plots, bar plots, and waterfall plots) to explain the feature importance of the final model.
RESULTS: The study included 65 age-matched patients (PD-NC: PD-SCD: PD-MCI= 14:21:30). Forty-five kinematic parameters were significantly different (P<.05) among the 3 groups, distributed across TUGst (n=25), TUGdt (n=12), and DTC (n=8) paradigms. Gait phase analysis revealed 35 parameters from walking phases, 9 from stand-to-sit transitions, and 1 from sit-to-stand transitions. Feature type distribution demonstrated predominance of variability features (n=20), followed by pace (n=12) and axial (n=8) characteristics. TUGdt paradigm analysis revealed pronounced movement differences between PD-MCI and both PD-NC and PD-SCD groups, particularly in variability, amplitude, pace, and axial domains. Cross-paradigm analysis identified consistent significant differences in specific features. These findings provide objective kinematic biomarkers for early cognitive state identification in Parkinson disease, with TUGdt parameters demonstrating superior discriminative capacity.
CONCLUSIONS: This suggests patients with PD-SCD could have early kinetic signs of cognitive impairment, positioning them between PD-NC and PD-MCI, and our multiclass support vector machine classification model with kinematic parameters achieved a recall rate above 0.70 in both training and validation datasets. The feature importance analysis revealed that DTC_Trunk-Right Rotation Max, DTC_Trunk-Max Transverse Angular Velocity, and dTUG_Lumbar-Right Sway Max Std were the most critical features for distinguishing cognitive states, providing scientific evidence for cognitive function screening based on kinematic parameters.
PMID:42341295 | DOI:10.2196/69273
