Remote Objective Monitoring of Adherence to a Home Rehabilitation Program for Upper Extremity Hemiparesis

Abstract:

Objective: Evaluate validity of applying artificial neural networks (ANN) to accelerometer signals for measuring adherence with an upper-extremity home neurorehabilitation program.

Design: Constraint-Induced Movement therapy patients wore an accelerometer above each wrist while they a) were treated in the laboratory and b) followed a home practice program using similar tasks. Therapists logged when patients performed tasks in the laboratory, and patients logged parallel information at home. An ANN was trained to discriminate periods when patients engaged in tasks from other periods using a randomly selected sample of laboratory data; accelerometer signals and therapist entries served as input and reference datasets, respectively. The trained ANN was tested on the remaining laboratory recordings and the home recordings.

Setting: Clinical research laboratory and patients’ homes.

Participants: Eight individuals more than 1-year post-stroke. Main Outcome Measure: Time in training as indicated by the ANN and therapist and patient logs.

Results: Across five randomly generated sub-samples of the laboratory recordings, the mean correlation between time in training as indicated by the ANN and therapist was .86. For home recordings, the corresponding correlation was .89.

Conclusions: Applying artificial neural networks to accelerometer signals might permit remote monitoring of adherence with upper-extremity home neurorehabilitation programs.

Key Words:

Arm, rehabilitation, adherence, ambulatory monitoring, stroke