Walking Asymmetry in Parkinson’s Disease Requires Multi-Layered Interpretation: A Longitudinal Single-Subject Analysis Using Apple Health Data
Abstract
Zhenghua Li and Kenichi Yamamura*
Background: Consumer-device gait metrics offer a promising means of continuously monitoring Parkinsonian gait in daily life. However, interpretation of algorithm-derived parameters such as walking asymmetry remains difficult because the absence of a recorded value may reflect either the absence of measurable asymmetry or failure of the walking episode to satisfy algorithmic recording conditions. In addition, our recent work has shown that Apple HealthKit-based digital phenotyping can detect pharmacological interference in Parkinsonian gait, demonstrating that this platform is useful not only for long-term monitoring but also for objective assessment of treatment-related motor deterioration. Objective: To re-evaluate the meaning of walking asymmetry data in Parkinson’s disease (PD) by separating the signal into three components: non-zero asymmetry values, days without measurable asymmetry, and zero-valued asymmetry events. Methods: A 77-year-old man with PD was monitored longitudinally using Apple Health data collected by iPhone during daily life. Raw XML files were examined directly to extract walking asymmetry, walking speed, step count, double-support percentage, and walking steadiness. Because asymmetry values of 0 were not retained in standard Auto Export output, XML-level analysis was required. Detailed analyses focused on October- December 2024 and October-December 2025, spanning the period before and after recurrent falls in November-December 2025.
Results: Walking asymmetry could not be interpreted as a single continuous variable. Instead, three distinct signals emerged. First, non-zero asymmetry values were observed, but their frequency markedly decreased in 2025 compared with 2024. Second, days without measurable asymmetry became more common during the fall-prone period, particularly in November 2025. Third, zero-valued asymmetry events, which were frequent in 2024, markedly decreased in 2025. Step count analysis showed that lack of asymmetry data was not explained solely by absence of walking; rather, in the more impaired state, greater walking volume was often required for asymmetry to be recorded. Walking steadiness showed only limited change during the fall-prone months, likely because this composite metric incorporates walking speed, step length, double-support percentage, and asymmetry, thereby attenuating the specific contribution of instability-related asymmetry changes.
Conclusion: In PD, Apple Health walking asymmetry should not be treated as a single metric. Its interpretation requires simultaneous consideration of non-zero asymmetry values, days without recorded asymmetry, and the frequency of zero-valued events. This multi-layered approach may provide a more informative framework for evaluating unstable gait and emerging fall risk in daily life than conventional averaging of asymmetry values alone.