Objective: To investigate whether wearable sensor-derived kinematic data during walking and standing can capture medication-induced changes in motor symptoms in people with Parkinson’s disease (PwP) by reflecting clinical motor symptom severity.

Background: Gait impairments and postural instability in PD reduce quality of life. Monitoring therapy-induced changes in motor symptoms remains challenging, as standard clinical assessments are subjective and time-consuming. Wearable sensor data objectively distinguish between medication states, focusing on individual features at a time or using binary machine learning classification models. Here, we combine wearable-derived data into a single score of disease severity to provide more robust, objective measures for capturing treatment-induced changes in motor symptoms.

Method: As part of the OxQUIP study, a 2-minute walk task and a 30-second postural sway task were performed while wearing six inertial sensors on wrists, feet, sternum and lumbar area by (a) 91 PwP, tested longitudinally (7 visits) and (b) 48 PwP, tested in both ON and OFF medication states. A random forest model, composed of 122 gait and postural sway features derived from these sensors, was trained and internally validated on the longitudinal cohort to estimate disease severity in terms of MDS-UPDRS-III scores. The model was then used to calculate the ON/OFF digital scores of the medication cohort. We examined the effect of medication on the digital score and the association between digital and clinical scores.

Results: Digital scores were lower in the ON compared to the OFF medication state (p<0.003), consistent with changes in MDS-UPDRS-III (p<0.001) and were correlated with clinical motor symptom severity when both medication states combined as well as separately in the ON and OFF medication state (r=0.62-0.65, p<0.001). The model accurately estimated clinical scores (RMSE: ON=8.60, OFF=10.20, total=9.34) outperforming models based on individual features.

Conclusion: Wearable-derived gait and postural sway characteristics combined using machine learning can detect medication-induced changes in motor symptom severity in PD. These findings highlight the potential of wearable sensor kinematic characteristics for objectively monitoring motor fluctuations and evaluating treatment effectiveness in clinical practice.

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/assessing-the-effects-of-anti-parkinsonian-medication-using-wearable-sensor-derived-kinematic-data-and-machine-learning/