Data and analysis code for "A virtual reality rehabilitation platform with EMG feedback for Parkinson's disease: design and a 6-week feasibility study"

Abstract

Data and analysis code for "A virtual reality rehabilitation platform with EMG feedback for Parkinson's disease: design and a 6-week feasibility study" This archive contains de-identified participant-level data and analysis code for a six-week supervised, single-arm feasibility study in Parkinson's disease. Forty-two participants enrolled and 37 completed 18 sessions. Datasets include characteristics and baseline measurements for 42 participants; session-level game records for 37 completers; paired secondary-game and EMG-derived values for 37 participants; user-experience and adherence summaries for 42 participants; and adverse-event records. A data dictionary, analysis scripts, table and figure code, source-file hashes and verification outputs accompany the data. Memory Market recall is the primary performance outcome and Balloon Burst accuracy is secondary. Their paired baseline/final comparisons use 37 participants and Holm adjustment across two tests. The hierarchy and multiplicity strategy were formulated after data collection. Other interpretable game measures and participant ratings are descriptive. A descriptive balloon-accuracy mixed model uses 671 recorded measurements from 42 participants, including five non-completer baselines. A completer-only sensitivity model and session-specific observation counts are supplied. Later non-completer measurements are unavailable and are not imputed. The slope summarizes performance during training; task difficulty was not modeled. The stored Garden composite, EMG-derived measurements and trajectory-smoothness index lack sufficient operational documentation for quantitative interpretation. Their values remain archived, with analytical use specified in the dictionary. Intermediate recall assessment conditions are undocumented, so recall analysis uses the baseline and final assessments. Raw EMG, study software, acquisition/feedback configurations and full questionnaire records are unavailable. These gaps limit reproduction of the intervention and measurement procedures. The seven input CSV files preserve the recorded observations. Run python run_all.py after installing requirements-tested.txt. README.md, DATA_DICTIONARY.md and ANALYSIS_SCOPE.md describe file coverage, analysis methods and limitations. The data and analysis code are provided through this Zenodo record. References to S1 Data in the archived documentation identify this same package. The uncontrolled study does not establish clinical efficacy or an independent effect of EMG feedback.