Shopping malls and digital health networks: predicting patterns of use
|Trevor van Mierlo||T.D.V.Vanmierlo@programme-member.henley.com|
Background: Digital behavior change interventions with social networks are common and have shown much promise in improving health outcomes. For interventions that have existed for several years, data analytics can give insight on demographic and psychographic characteristics of registrants, as well as patterns of use. Research methods in disciplines other than healthcare can assist in understanding relationships and trends. For example, in economics, it has been established that consumers are attracted to shopping malls with well-known anchor stores. When there is a lack of anchor stores, footfall decreases. Likewise, healthcare literature has identified Superusers, a small percentage of actors, as integral to the size and strength of a social network. In this study we hypothesize that Superusers function as anchors. To test the hypothesis we compared program usage patterns between actors and non-actors, and assessed the contribution of Superusers
Methods: Individual and network-level data were extracted from SQL databases of four free-to-consumer digital health interventions. Statistical analyses were conducted in SPSS 21 for Mac and Microsoft Excel. The interventions were publically available for different time spans (minimum 8.5 years, maximum 10.9 years), and addressed different topics (problem drinking, depression, panic disorder, and smoking cessation). All four interventions contained tailored exercises based on the theoretical principles of Structured Relapse Prevention, Cognitive-Behavioral Therapy, or the Stages of Change. Each intervention also contained a social network moderated by trained health experts, diaries, blogs, and brief interventions. Chi-squared tests were computed to investigate demographic characteristics and usage patterns of social network actors and non-actors. In order to examine patterns of social network engagement, the total number of posts written by each actor was plotted on logarithmic scales.
Results: 26% (n=16,880) of intervention registrants (N=64,986) were active users who authored one or more social network posts. When analyzing user behavior on an individual level, there were statistically significant associations between social network use and engagement with tailored exercises (P<0.01). On an aggregate level, 1% (n=169) of active users created over 73% of network content. Social network contributory patterns strongly resembled power curve distributions, with high R2 values (>.95), and significant Spearman Correlations (>.967, P=0.00).
Conclusions: To our knowledge, this is the first study to compare demographic characteristics and usage patterns within and amongst four distinct digital behavior change interventions. Results indicate that demographic characteristics cannot be used as strong predictors of intervention enrollment, engagement, or social network use. However, there appears to be significant relationships between social network use and intervention engagement. The structure of each social network resembled a power law, indicating that network engagement may be predictive. Superusers are integral to the size and strength of a network. They create positive externalities, increase network value, and are analogous to anchors. Through their consistent creation of fresh content, Superusers also function as patient experts, and actively engage other actors and non-actors. Moderators and managers can strategically encourage Superusers through principles of gamification, cooperative game theory, and behavioral economics. Practical examples, study limitations and future directions will be discussed.