JMIR Publications

ABSTRACT

Background:

Digital health interventions (DHIs) are identified as a potential means of improving dietary health at scale. However, engagement and retention rates are often low, reducing their impact and effectiveness. In dietary DHIs, a range of application (app) features have been implemented to improve engagement. However, it remains unclear what the key, early predictors of engagement and retention are at scale, in real-world deployment. This is particularly important given that retention is often overestimated in controlled research studies, and most attrition occurs soon after initial engagement.

Objective:

This study aimed to assess predictors of 7- and 14-day retention with a free-to-use, personalized, dietary DHI, using large-scale, real-world data from a commercial launch. It also aimed to assess predictors of day-to-day retention across the first 14-days of use, given that factors affecting initial engagement may differ from those affecting ongoing use.

Methods:

Users of a free-to-use dietary DHI that analysed food characteristics for dietary health based on food photographs and packaging barcodes were given unique identifiers (n = 60,987). Demographic information, use of app features, and app usage patterns were included in logistic regression models to assess predictors of retention 7- and 14-days after enrolment in the app. Multilevel modelling was used to assess predictors of engagement on each of the first 14-days of app enrolment.

Results:

Specific app-related activities and user referral sources consistently predicted a greater likelihood of retention at both 7- and 14-days. While photographing and scanning food were key features, the strongest predictor of engagement was the subsequent logging of consumption, which yielded higher predictive value than the capture events alone, and enabling app notifications to allow engagement reminders. User referral sources were also among the strongest predictors of retention: Users who reported hearing about the app via related podcasts or via “friends and family” had higher odds of retention at 7- and 14 days. Those who heard about the app via television or radio were less likely to engage with the app at 7- and 14-days.

Conclusions:

This study suggests that immediate interaction with dietary DHI activities linked to real-life behaviors (e.g., logging food eaten) increases the likelihood of continued engagement. It also suggests that engagement with personal contacts who may also be engaged in app-related activities can enhance retention, as can engagement with related media such as podcasts. However, some sub-groups of users may engage with free-to-use dietary DHIs more casually, with potentially less committed motivation to continue. In these situations, greater initial personalization for groups identified as at risk of attrition could enhance engagement.