Objective: To quantify the extent to which contextual factors explain the variability in real-world gait outcomes in people with Parkinson’s disease (PwP) using machine learning (ML).

Background: Real-world gait assessments in PwP can provide important insights for clinicians regarding disease progression and the effectiveness of interventions. Current approaches typically summarize several days of recording into a single outcome measure. However, gait in daily life is influenced by a variety of contextual factors, such as medication, weather, the amount of walking and environment leading to fluctuations within and across days. The collective contribution of these factors to gait variability during real-world walking has not been quantified.

Method: A dataset of 20 PwP (13 male, age: 65.5 years, disease duration: 5.1 years, LEDD: 581.9 mg/day) was reanalysed in this study [1]. Participants wore an inertial measurement unit (Axivity AX6) on the lower back for 7 consecutive days to derive gait outcomes related to pace (stride length, stride velocity) and rhythm domains (stride time, cadence), walking volume (stride count) and pattern (bout duration). Medication intake was reminded and logged via a smartwatch, while additional contextual data (probability of being indoors/outdoors, weather) were extracted using a smartphone. A ML regression model was trained for each participant to predict the gait outcomes using 13 features extracted from the contextual data and to identify factors most strongly associated with gait variability.

Results: Contextual factors explained 30-40% of the variability in real-world gait, with the highest predictive performance observed for stride velocity (R² = 0.39 ± 0.16) and the lowest for cadence (R² = 0.30 ± 0.14). Feature importance analysis revealed that bout-specific features (stride count, bout duration), start time of the bout and time since last medication intake were the primary predictors of gait variability.

Conclusion: Our findings show that real-world walking in PwP is substantially influenced by contextual factors. Consequently, incorporating contextual data is important for a meaningful interpretation of real-world gait data and to distinguish clinically relevant motor changes from external fluctuations. Future research should prioritize the development of models that account for these factors to accurately quantify gait dynamics in the real world.

Understanding gait fluctuations using context

References: [1] Packer et al BMJ Open 2023

« Back to 2026 International Congress

MDS Abstracts - https://www.mdsabstracts.org/abstract/why-context-matters-the-influence-of-medication-and-environment-on-real-world-gait-in-people-with-parkinsons-disease/