Prediction of Long-term Adherence to Smartphone Sensor-based Tests (Floodlight)
Marisa McGinley1, Gregoire Pointeau2, Mattia Zanon2, Letizia Leocani3, Giancarlo Comi3, Florian Lipsmeier2, Licinio Craveiro2, Helmut Butzkueven4
1Cleveland Clinic, Mellen Centre, 2F. Hoffmann-La Roche Ltd, 3Vita-Salute San Raffaele University, 4Department of Neuroscience, Monash University
Objective:
Understanding and predicting long-term adherence to Floodlight smartphone sensor-based tests in people with progressive multiple sclerosis (PwPMS), treated with ocrelizumab in CONSONANCE (NCT03523858).
Background:
Adherence remains a challenge for deploying remote digital health solutions, as their utility often depends upon obtaining sufficient data over a period of time. Understanding adherence patterns could enable patient-tailored interventions that boost long-term adherence.
Design/Methods:
Participants performed sensor-based tests, using a precursor to Floodlight™ MS, assessing their cognition (weekly), upper extremity function (daily), gait and balance (daily); and passive monitoring of gait (≥4 hours daily) for up to 3 years. Unsupervised Hierarchical Agglomerative Clustering was used to identify five adherence clusters. The drop-out date (the first day of the first period of 4 weeks of inactivity) was predicted via a linear model that used the first 2 months of adherence data as predictors.
Results:
Of 633 enrolled participants, 348 were included in the analysis. The five adherence clusters identified were: High Adherence (n=81; 23.3%), Slow Decrease (n=119; 34.2%), Fast Decrease (n=67; 19.3%), Low Adherence (n=50; 14.4%) and No Gait (n=31; 8.9%). No significant differences were observed in age (Mann–Whitney U test: all p>0.05) or sex (absolute Pearson residuals: all <0.40) of PwPMS between the clusters. Good correlation was found between the predicted and actual drop-out dates (Pearson r=0.80). Higher adherence was associated with later drop out, and most High Adherence participants (74/81; 91.4%) showed no inactivity period.
Conclusions:
Adherence during the first 2 months of using Floodlight technology can predict the time of inactivity of participants. Understanding adherence patterns may enable early, adherence-boosting interventions that focus on participants at risk of dropping out of remotely administered digital health studies, leading to an increase in data collection and improved engagement of PwPMS.