Assessing Digital Phenotyping for App Recommendations and Sustained Engagement: Cohort Study (Preprint)

Abstract

BACKGROUND Low engagement with mental health apps continues to limit their impact. New approaches to helping match patients to the right app may increase engagement by ensuring the app they are using is best suited to their mental health needs. OBJECTIVE This study piloted how digital phenotyping, using data from smartphone sensors to infer symptom, behavioral and functional outcomes, could be used to match people to mental health apps and potentially increase engagement METHODS After one week of collecting digital phenotyping data with the mindLAMP app, participants were randomized to feedback and recommendations based on that data for selecting one of four predetermined mental health apps (related to mood, anxiety, sleep, fitness) or in the control arm, selecting the same apps but without any feedback or recommendations. All participants used their selected app for 4 weeks with numerous metrics of engagement recorded including objective screentime measures, self reported engagement measures, and Digital Working Alliance Inventory scores. The study offered participants no compensation. RESULTS 82 enrolled in the study and 17 dropped out of the digital phenotyping arm and 18 from the control arm. Across both groups, few participants chose or were recommended the insomnia or fitness app. The majority 78.7% used a depression or anxiety app. Engagement as measured by objective screen time and Digital Working Alliance Inventory scores were higher in the digital phenotyping arm. There was no correlation between self-reported and objective metrics of app use. Qualitative results highlighted the importance of habit formation in sustained app use. CONCLUSIONS Results suggest that digital phenotyping app recommendation is feasible and may increase engagement. This approach is generalizable to other apps beyond the four selected for use in this pilot, and practical for real-world use given the study was conducted without any compensation or external incentives that may have biased results. Advances in digital phenotyping will likely make this method of app recommendation more personalized and thus of even greater interest.