From Preference to Performance: Patient-Centered Design of Multimodal Cueing in Parkinson's Disease Gait Training
Xinjin Li, Wenjie Wang, Kai Wang, Houzhen Tuo, Xiaolong Ma, Xiaohui Tan, Wei Sun, Feng Tian, Xiaojuan Ma · 2026 · Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems (CHI '26) · doi:10.1145/3772318.3791182
Summary
This CHI 2026 paper reports a patient-centered design and evaluation of a wearable multimodal cueing system for gait rehabilitation in early-stage Parkinson's disease (PD). The authors first run semi-structured interviews with 10 early-stage PD patients and 5 clinicians (3 neurologists, 2 rehabilitation specialists) to distill the CARES design framework - Centered on the patient, Adaptive and personalized, Reinforced and guided, Evidence-based and Safe - with 11 concrete subprinciples. They then build a prototype with ankle-mounted IMUs feeding a Microsoft Surface tablet that computes each patient's left/right gait cycle and step height in real time, then drives six cueing conditions across three sensory modalities and two paradigms: Periodic cueing (Periodic Green Line via HoloLens 2, Periodic Beats via open-ear headphones, Periodic Vibration via an ankle-mounted linear resonant actuator) and Process cueing (Virtual Character AR companion, Continuous Music with BPM synchronized to gait cycle, Continuous Vibration tracking step phase). Two within-subject user studies with 16 early-stage PD patients follow: User Study 1 compares periodic versus process paradigms across all three modalities; User Study 2 isolates modality effects using the best-performing process variants plus periodic vibration. Gait is measured via IMU-derived velocity, Gait Consistency Index, and Relative Difference Index; subjective data come from NASA-TLX and 5-point preference ratings. The paper releases both the CARES framework and evidence for a striking performance-preference dissociation.
Key findings
External cueing significantly improved affected-side gait velocity and bilateral symmetry across all tested conditions. Process cueing produced larger velocity gains than periodic cueing (9.15% vs 4.53% averaged across modalities) for visual and auditory channels, while periodic cueing held a modest advantage for somatosensory (periodic vibration +3.47% vs continuous vibration +1.21%). User Study 2 cleanly separated modalities: auditory process cueing (continuous music, BPM-synced) produced the largest objective gains - velocity +18.21% (d=1.41), strongest improvement in Gait Consistency Index (64.34% from 58.39%, d=2.09), and the biggest reduction in swing-phase asymmetry. Visual cueing (virtual character) was second on objective measures (+17.16% velocity, d=1.15) but yielded the lowest NASA-TLX workload and the highest user preference (mean 4.44/5, IQR 1.00). Somatosensory periodic vibration improved gait least (+9.48%) and imposed the highest workload. This produced a clear performance-preference dissociation: auditory won on outcomes, visual won on acceptance. Patients described the AR virtual character as "a partner practicing with me," surfacing companionship as a core psychosocial need. Clinicians endorsed the system but flagged safety concerns about HoloLens use outdoors and the risk that perceived improvement prompts patients to reduce medication prematurely.
Relevance
For practitioners working on motor-disability assistive technology, this paper is a concrete template for moving PD rehabilitation out of the clinic and into daily home-based training. Three takeaways stand out. (1) The CARES framework - particularly the principles of patient choice, dynamic personalization, and evidence-based safety - is directly reusable as a design-review checklist for any chronic-condition wearable. (2) The performance-preference dissociation is an important design warning: the cueing modality that optimizes short-term gait metrics (auditory) is not the one patients will actually adopt long-term (visual), so systems intended for sustained home use must offer selectable modalities rather than prescribing a single "optimal" one. (3) The finding that companionship (not just cueing) drove patient preference suggests that geriatric rehabilitation tools should be designed as socio-technical companions, not isolated biomechanical feedback devices. Limitations: the 16-participant sample is limited to early-stage idiopathic PD with preserved visual and cognitive function, sessions were short (five minutes per condition in a controlled lab), and the HoloLens 2 form factor is expensive and unlikely to scale to home deployment. Longer in-home studies and lighter-weight AR hardware are needed before clinical deployment.
Tags: parkinson's disease · gait training · multimodal cueing · rehabilitation · wearable technology · augmented reality · haptic feedback · motor disability · patient-centered design · IMU
Standards referenced: MDS-UPDRS · Hoehn and Yahr scale