Learning to be Proficient? A Structural Model of User Dynamic Engagement in eHealth Interventions

Abstract

eHealth interventions have transformed how individuals manage their health by offering prompt and accessible support for modifying their lifestyles. In this study, we investigate how individuals dynamically learn about the effectiveness of eHealth interventions to shed light on their continued participation. To capture the potential changes in individuals’ perceptions and associated behavior dynamics, we establish a hierarchical Bayesian learning framework to structurally characterize individuals’ decision-making processes. Our model addresses several unique characteristics of learning in the health-management setting, such as correlations between intervention choices and heterogeneity in delivered health signals. Through analysis of a 4-month dataset of users’ intervention participation in an online weight-loss platform, we demonstrate that individuals’ learning performance may vary significantly across intervention types. Our analysis reveals that individuals’ learning performance tends to be suboptimal for interventions with ambiguous instructions or those that prioritize short-term health improvements. Given that individuals’ learning performance is largely influenced by the level of noise contained in their intervention experiences, we identified possible noise sources and proposed several denoising strategies to improve user engagement and learning outcomes, with their efficacy analyzed through counterfactual analysis. The proposed strategies can be easily integrated into platform design to generate a positive impact on user engagement. Our empirical estimation also leads to interesting findings regarding heterogeneity in intervention experiences and users’ preferences for descriptive intervention features. These findings provide various implications for the personalization and design of online healthcare interventions.

Keywords: personal health management, wellness promotion, lifestyle change, eHealth interventions, online healthcare platforms, dynamic participation, Bayesian learning