Author(s): SCHURMANS, G., BACKES, A., FAYAD, J., MEYER, T., ECKELT, M., GRIMM, B., MOUTON, C., SEIL, R., MALISOUX, L. , Institution: LUXEMBOURG INSTITUTE OF HEALTH, Country: LUXEMBOURG, Abstract-ID: 2092

INTRODUCTION:
A new digital health solution consisting of pressure-sensitive insoles combined with inertial measurement units has been developed to provide clinicians with relevant information on the progress of their patients in their natural environment by detailing aspects of movement quality and quantity, and to monitor their patient’s everyday activities. This study aims to test the system’s ability to properly identify and quantify different types of daily life activities.
METHODS:
Healthy volunteers were invited to complete a series of 12 different usual daily life activities (i.e., sitting straight and relaxed, standing, shuffling, walking straight and non-straight, walking with crutches, hill climbing and descending, stairs climbing and descending, and indoor cycling). Participants were instructed to perform each of the activities twice, in the order and duration of their choice. Data from the system was compared with direct observation of video recording using software for behavioural research. Two observers analysed independently each video recording. Outcomes include both the total duration of each activity and the total counts of events, when appropriate (e.g., steps). The interrater reliability was assessed using Cohen’s Kappa statistic (K). The system’s accuracy to properly identify and quantify activities was examined by calculating the mean absolute percentage error (MAPE scores) and Bland Altman plots.
RESULTS:
A total of 100 participants were included (50 females (50%); median age = 29 years [IQR: 26; 43.3]; median body mass index = 24.7 kg.m^-2 [21.3; 25.9]). The total measurement time was approximately 10 min per participant (activities lasted from 5 to 60 seconds). The interrater reliability was excellent (K=0.85). Overall, MAPE scores between video analysis and the device ranged from 10.4% (total duration of level walking) to 381.1% (total standing count). The system seems to have a general good estimation for some walking duration variables (walking straight, level and non-level walking; MAPE score between 10-20%) and reasonable estimation for the total sitting duration and certain counts (total steps and stairs; MAPE score 20-50%). The performance to identify and quantify the other activities was lower (MAPE score > 50%), and it seems that the algorithm was particularly ineffective in identifying and quantifying standing count and duration (MAPE score > 100%).
CONCLUSION:
In a controlled laboratory environment, the system’s performance in classifying and quantifying accurately daily life activities was heterogeneous as some activities were estimated with good accuracy (e.g., walking straight duration) while other activities were erroneous (e.g., standing duration). Further development is needed to improve classification algorithms. Future studies will evaluate the acceptance and ease of use of the system in orthopaedic patients.