Skip to main content

Estimated Effects of Behavior Change Components in mHealth Interventions for Physical Activity and Related Psychological Outcomes: Bayesian Component Network Meta-Analysis

In brief

Self-monitoring and gamification show small positive effects in mHealth trials

Across 130 randomized trials involving 25,369 people, self-monitoring and gamification had the clearest evidence of benefit, with small estimated effects on physical activity and related psychological outcomes. Feedback also had a positive estimate, but uncertainty included no effect; a weaker social-comparison result for psychological outcomes came from sparse evidence and needs confirmation.

Journal
Journal of medical Internet research (Q1)
Published
19 November 2026
Study design
Systematic review / meta-analysis of RCTs
Evidence level
Level 1, High (CEBM 1a)
Authors
Yimeng Gu, Yongheng Zhao, Wang Xi, Long Cheng
PMID
43056703
DOI
10.2196/88460

Why clinicians should know about it

  • Picked for Health Informatics (paper of the day, 20 November 2026): Meta-analysis estimates effects of digital intervention components

Abstract

BACKGROUND: Mobile health (mHealth) interventions for physical activity commonly combine several behavior change and digital design components, making it difficult to determine which components contribute to outcomes. Component estimates also depend on how repeated contrasts within trials and departures from additivity are handled. OBJECTIVE: This study aimed to estimate component-level contributions of 7 operationalized intervention components and examine whether their contributions differed between physical activity and related psychological outcomes. METHODS: We conducted a systematic review and Bayesian component network meta-analysis of 130 randomized controlled trials involving 25,369 participants after a final study-design eligibility reaudit. Component composition was coded for 310 study arms. The primary target-scope dataset contained 469 component-informative contrast-level observations from 121 trials; a broader 492-observation dataset was analyzed as a sensitivity. A Bayesian additive component metaregression used a multivariate normal likelihood by study, a shared-comparator covariance for multiarm outcome sets, a study-level random effect, and within-study residual heterogeneity. Sensitivity analyses varied working cross-outcome and cross-time correlations, restricted each study to a primary outcome group, excluded cluster-coded trials, compared outcome domains, and explored component interactions. RESULTS: In the primary hierarchical model, self-monitoring (Hedges g=0.153, 95% credible interval [CrI] 0.053-0.249) and gamification (g=0.102, 95% CrI 0.020-0.183) had CrIs above 0. Feedback retained a positive point estimate (g=0.127), but its 95% CrI included 0 (-0.004 to 0.259) after the final eligibility reaudit; intervals for the other 4 components also included 0. Across the 7 outcome-domain contrasts, social comparison was lower for psychological than physical activity outcomes (psychological-minus-physical activity difference g=-0.189, 95% CrI -0.325 to -0.052), including in the 66-trial same-study analysis (g=-0.201, 95% CrI -0.386 to -0.021), although social comparison was sparsely represented. For feedback, the corresponding differences were 0.084 (95% CrI -0.088 to 0.254) and 0.046 (95% CrI -0.163 to 0.254), respectively. The low-risk subset was rank deficient and did not support stable estimation of all 7 components. CONCLUSIONS: Self-monitoring and gamification had the clearest evidence of positive component contributions. Feedback retained a positive point estimate, but its CrI included 0 in the eligibility-locked analysis, and the data did not support a robust divergence in feedback by outcome domain. The isolated social comparison domain contrast arose from sparse evidence and should be considered hypothesis generating. Component rankings should be interpreted alongside small absolute effects, uncertainty, the pragmatic additivity assumption, uncorrected cluster designs in the source summaries, and limited low-risk evidence. TRIAL REGISTRATION: PROSPERO CRD420251182885; https://www.crd.york.ac.uk/PROSPERO/view/CRD420251182885.

Abstract as published, via PubMed.

View on PubMedFull text at the publisherOpen in the app

For healthcare professionals. The summary is generated by AI from the published abstract, and the evidence level is assigned automatically from the study design on the Oxford CEBM hierarchy. Neither is medical advice. Read the full paper before changing practice.