User Engagement and Attrition in an App-Based Physical Activity Intervention: Secondary Analysis of a Randomized Controlled Trial

Abstract

BackgroundThe success of a mobile phone app in changing health behavior is thought to be contingent on engagement, commonly operationalized as frequency of use.ObjectiveThis subgroup analysis of the 2 intervention arms from a 3-group randomized controlled trial aimed to examine user engagement with a 100-day physical activity intervention delivered via an app. Rates of engagement, associations between user characteristics and engagement, and whether engagement was related to intervention efficacy were examined.MethodsEngagement was captured in a real-time log of interactions by users randomized to either a gamified (n=141) or nongamified version of the same app (n=160). Physical activity was assessed via accelerometry and self-report at baseline and 3-month follow-up. Survival analysis was used to assess time to nonuse attrition. Mixed models examined associations between user characteristics and engagement (total app use). Characteristics of super users (top quartile of users) and regular users (lowest 3 quartiles) were compared using t tests and a chi-square analysis. Linear mixed models were used to assess whether being a super user was related to change in physical activity over time.ResultsEngagement was high. Attrition (30 days of nonuse) occurred in 32% and 39% of the gamified and basic groups, respectively, with no significant between-group differences in time to attrition (P=.17). Users with a body mass index (BMI) in the healthy range had higher total app use (mean 230.5, 95% CI 190.6-270.5; F2=8.67; P<.001), compared with users whose BMI was overweight or obese (mean 170.6, 95% CI 139.5-201.6; mean 132.9, 95% CI 104.8-161.0). Older users had higher total app use (mean 200.4, 95% CI 171.9-228.9; F1=6.385; P=.01) than younger users (mean 155.6, 95% CI 128.5-182.6). Super users were 4.6 years older (t297=3.6; P<.001) and less likely to have a BMI in the obese range (χ22=15.1; P<.001). At the 3-month follow-up, super users were completing 28.2 (95% CI 9.4-46.9) more minutes of objectively measured physical activity than regular users (F1,272=4.76; P=.03).ConclusionsTotal app use was high across the 100-day intervention period, and the inclusion of gamified features enhanced engagement. Participants who engaged the most saw significantly greater increases to their objectively measured physical activity over time, supporting the theory that intervention exposure is linked to efficacy. Further research is needed to determine whether these findings are replicated in other app-based interventions, including those experimentally evaluating engagement and those conducted in real-world settings.Trial RegistrationAustralian New Zealand Clinical Trials Registry ACTRN12617000113358; https://www.anzctr.org.au/ACTRN12617000113358.aspx

A Mobile App Specifically Designed to Facilitate Exercise in Parkinson Disease: Single-Cohort Pilot Study on Feasibility, Safety, and Signal of Efficacy

“…Regardless, 2 analyses (mixed factorial ANOVAs and correlational analyses) were conducted to determine if there was a dose effect of app usage on the outcome measures. The results of these analyses did not indicate that there was a dose effect and the very low correlations support this notion despite the fact that other exercise and smartphone app studies have linked exercise exposure to efficacy [ 51 ]. Further investigation into the dosing of the 9zest app-guided exercise using a more rigorously controlled design is warranted.…”

A systematic review of the dose-response relationship between usage and outcomes of online physical activity weight-loss interventions

“…Despite these advantages, persistent usage of online-based interventions is a concern ( Eysenbach, 2005 ). Previous studies have reported that usage of online-based interventions dropped under 50% within one to two month after starting the program ( Geraghty et al, 2013 ; Edney et al, 2019 ; Wangberg et al, 2008 ). Low intervention usage may not necessarily mean that participants have not improved their target behavior.…”

User Engagement With Smartphone Apps and Cardiovascular Disease Risk Factor Outcomes: Systematic Review (Preprint)

“…The study characteristics are summarized in Multimedia Appendix 2 . Of the 24 studies, 17 used an RCT design [16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32], 4 used a retrospective analysis [33][34][35][36], and 3 used a single-group pre-and posttest quasi-experimental design [37][38][39]. Although 20 studies used an RCT or single-group preand-posttest quasi-experimental design, it was the correlational studies nested within these larger parent studies that were of interest for this review.…”

“…Of the 17 RCT studies, 14 conducted an a priori power analysis. Of these 14 RCTs, 3 detected a significant difference in the primary outcome between groups [21,23,27], 9 did not detect a significant difference in the primary outcome between groups [16,17,22,[24][25][26]29,31,32], and 2 did not report the results of the difference in the primary outcome between groups [18,30]. One of the studies that did not detect a significant difference between groups did not achieve the target sample size [32].…”

“…Thirteen studies reported sample races [17][18][19][21][22][23][24][25][26][27][28]32,39], with 7 consisting largely of White participants [18,19,22,[24][25][26]28]. Seventeen studies reported on sample educational level, with the majority of the participants within each individual study having attended some college [16][17][18][19][20][21][22][23][24][25][26][28][29][30][31][32]38,39]. Ten studies limited their sample to participants who were overweight or obese [17,19,[21][22][23][24][25][26]28,35], and another 11 studies reported enrolling participants with a baseline mean or majority percentage BMI indicating overweight or obes...…”