PresentationPDF Available

Trajectories of Engagement with a Digital Physical Activity Coach: Secondary Analysis of a Micro-Randomized Trial

Presentation

Trajectories of Engagement with a Digital Physical Activity Coach: Secondary Analysis of a Micro-Randomized Trial

Abstract

Context: Intervention components of a MobileCoach based smartphone app to promote walking were assessed in a seven-week micro-randomized trial (N = 274). In order to make a significant contribution to public health, the app must also engage those at higher risk for adverse health outcomes, i.e. elderly, less active or less healthy participants. Methods: In a secondary analysis, longitudinal trajectories of participants’ number of daily app sessions were clustered using the k-means algorithm with an alternating number of clusters. An app session was defined as any interaction between a participant and the app separated by at least five minutes between interactions. The final number of clusters was determined based on five different clustering quality criteria. Results: Two different clusters emerged: stable high engagement (31.3% of participants, 7.6 (SD = 2.9) mean daily app sessions) and stable low engagement (68.7% of participants, 1.5 (SD = 1.4) mean daily app sessions). Highly engaged participants were older (45.8 vs. 40.1 years, p < .001, d = 0.43) and accumulated more steps per day during the study (7373 vs. 5828 steps per day, p < .001, d = 0.57). Clusters did not differ with regard to participants’ baseline physical activity, gender, body mass index, self-reported health status or education. Conclusions: A chatbot-based walking app engaged participants of a micro-randomized trial over a period of seven weeks independent of their risk for adverse health outcomes. Thus, participants with low risk for adverse health outcomes at baseline do not drive high engagement with the app.
Trajectories of Engagement with a Digital Physical Activity
Coach: Secondary Analysis of a Micro-Randomized Trial
Name of presenting author: Jan-Niklas Kramer
Name of organisation of presenting author: University of St.Gallen
Country of presenting author: Switzerland
Email address of presenting author: jan-niklas.kramer@unisg.ch
Author details:
Jan-Niklas Kramer1, Florian Künzler2, Peter Tinschert1, Tobias Kowatsch1
1University of St. Gallen, Institute of Technology Management, St. Gallen, Switzerland
2ETH Zurich, Department of Management, Technology and Economics, Zurich, Switzerland
Preferred format of presentation (delete the option that does not apply):
Oral Presentation
Context: Intervention components of a MobileCoach1 based smartphone app to
promote walking were assessed in a seven-week micro-randomized trial (N = 274). In
order to make a significant contribution to public health, the app must also engage
those at higher risk for adverse health outcomes, i.e. elderly, less active or less healthy
participants.
Methods: In a secondary analysis, longitudinal trajectories of participants’ number of
daily app sessions were clustered using the k-means algorithm with an alternating
number of clusters2. An app session was defined as any interaction between a
participant and the app separated by at least five minutes between interactions. The
final number of clusters was determined based on five different clustering quality
criteria.
Results: Two different clusters emerged: stable high engagement (31.3% of
participants, 7.6 (SD = 2.9) mean daily app sessions) and stable low engagement
(68.7% of participants, 1.5 (SD = 1.4) mean daily app sessions). Highly engaged
participants were older (45.8 vs. 40.1 years, p < .001, d = 0.43) and accumulated more
steps per day during the study (7373 vs. 5828 steps per day, p < .001, d = 0.57).
Clusters did not differ with regard to participants’ baseline physical activity, gender,
body mass index, self-reported health status or education.
Conclusions: A chatbot-based walking app engaged participants of a micro-randomized
trial over a period of seven weeks independent of their risk for adverse health
outcomes. Thus, participants with low risk for adverse health outcomes at baseline do
not drive high engagement with the app.
1. Kowatsch T, Volland D, Shih I, et al. Design and Evaluation of a Mobile Chat
App for the Open Source Behavioral Health Intervention Platform
MobileCoach. Paper presented at: International Conference on Design Science
Research in Information Systems; 30 May - 1 June, 2017; Karlsruhe, Germany.
2. Genolini C, Alacoque X, Sentenac M, Arnaud C. kml and kml3d: R Packages to
Cluster Longitudinal Data. Journal of Statistical Software.65(4):1-32.
ResearchGate has not been able to resolve any citations for this publication.
Conference Paper
Full-text available
The open source platform MobileCoach (mobile-coach.eu) has been used for various behavioral health interventions in the public health context. However, so far, MobileCoach is limited to text message-based interactions. That is, participants use error-prone and laborious text-input fields and have to bear the SMS costs. Moreover, MobileCoach does not provide a dedicated chat channel for individual requests beyond the processing capabilities of its chatbot. Intervention designers are also limited to text-based self-report data. In this paper, we thus present a mobile chat app with pre-defined answer options, a dedicated chat channel for patients and health professionals and sensor data integration for the MobileCoach platform. Results of a pretest (N = 11) and preliminary findings of a randomized controlled clinical trial (N = 14) with young patients, who participate in an intervention for the treatment of obesity, are promising with respect to the utility of the chat app.
Article
Full-text available
Longitudinal studies are essential tools in medical research. In these studies, variables are not restricted to single measurements but can be seen as variable-trajectories, either single or joint. Thus, an important question concerns the identification of homogeneous patient trajectories. kml and kml3d are R packages providing an implementation of k-means designed to work specifically on trajectories (kml) or on joint trajectories (kml3d). They provide various tools to work on longitudinal data: imputation methods for trajectories (nine classic and one original), methods to define starting conditions in k-means (four classic and three original) and quality criteria to choose the best number of clusters (four classic and one original). In addition, they oer graphic facilities to "visualize" the trajectories, either in 2D (single trajectory) or 3D (joint-trajectories). The 3D graph representing the mean joint-trajectories of each cluster can be exported through LATEX in a 3D dynamic rotating PDF graph (Figures 1 and 9).