BACKGROUND
Digital Health Interventions (DHIs) are widely used to manage users' health in everyday life through digital devices. The use of DHIs generates various data, such as records of intervention usage and the status of target symptoms, providing researchers with data-driven insights for improving these interventions even after deployment. Although DHI researchers have investigated this data, existing analysis practices have been carried out in a fragmented manner, limiting the comprehensive understanding of the data.
OBJECTIVE
We proposed an analysis task model to help DHI researchers analyze observational data from a holistic perspective. This model was then used to prototype an interactive visual analytics tool. Our objective is to evaluate the model’s suitability for DHI data analysis and explore task support through a visual analytics tool.
METHODS
We constructed an analysis task model based on data analysis practices from existing DHI research. Moreover, we designed 'Maum Health Analytics,' an initial prototype of an interactive visual analytics tool that supports the tasks included in the proposed model. To investigate whether our model adequately covers the DHI data analysis process, we conducted a preliminary user study with five groups of DHI researchers (n=15). During this process, we had them use Maum Health Analytics within given data analysis scenarios, providing analyzed results from in-the-wild data collected in a non-experimental setting through a mobile DHI service targeting depressive symptoms. After using the analytics tool, we interviewed the DHI researchers to determine whether the analysis tasks were appropriate and how the information provided by the tool could be utilized in practice.
RESULTS
Our analysis task model was created using three key components (i.e., user grouping criteria) for DHI data analysis: user characteristics, user engagement with DHIs, and the effectiveness of DHIs on the target symptom via pre-post comparisons. Furthermore, the prototype of interactive visual analytics was designed, with each feature mapped one-to-one to an analysis task described in the model. From the interview sessions, DHI researchers valued group-level analysis that enabled identifying users who need care, improving intervention content and recommendations, and understanding the effectiveness of DHIs in connection with user characteristics and engagement levels. They also noted several benefits of the model and tool, such as simplifying analysis tasks and supporting communication among diverse experts.
CONCLUSIONS
We proposed an analysis task model and designed an interactive visual analytics tool to support DHI researchers. Our user study showed that the model allows a holistic investigation of DHI data by integrating key analysis components, and the prototype tool simplifies analytic tasks and enhances communication among researchers. As DHIs grow, our model and tool could effectively meet the data analysis needs of researchers and improve efficiency.