Abstract
BACKGROUND: Instant messaging delivered through social platforms is increasingly used to support diabetes self-management. However, evidence remains difficult to interpret because trials vary widely in platform choice, intervention design, outcome constructs, and measurement instruments. OBJECTIVE: To synthesize the effects of instant-messaging interventions for diabetes across pre-specified outcome domains, and to explore whether platform type, follow-up duration, and study size help explain variation in effect estimates. METHODS: We searched seven databases (2010-2025) for randomized controlled trials in which diabetes interventions were primarily delivered via instant-messaging/social platforms. Outcomes were organized a priori into six domains: health behaviors, diabetes knowledge, attitudes/self-efficacy, glycemic outcomes, other clinical outcomes, and diabetes-related complications. Continuous outcomes were pooled as standardized mean differences (SMDs) and binary outcomes as risk ratios (RRs) using random-effects models (REML). To improve interpretability, we prioritized domain-level synthesis and performed platform-stratified pooling only when at least three effect sizes were available within a given domain. Heterogeneity was summarized using τ(2) and I(2). Robustness was assessed using leave-one-out analyses; small-study effects were evaluated using funnel plots and Egger's test when ≥10 studies contributed to an analysis. Exploratory meta-regression examined follow-up duration and ln(sample size). RESULTS: Twenty-three trials contributed 236 effect estimates. Overall pooled effects across all continuous and binary outcomes were close to null and statistically non-significant, with substantial heterogeneity. Domain-specific synthesis showed clearer patterns: diabetes knowledge demonstrated the largest pooled improvement (SMD = 1.065, 95% CI 0.185-1.944), glycemic outcomes improved on continuous measures (SMD = -0.519, 95% CI -0.719 to -0.319), and behavioral outcomes showed a small but significant benefit (SMD = 0.359, 95% CI 0.010-0.709). Attitudes/self-efficacy and other clinical outcomes were more heterogeneous and did not show clear pooled benefits. For complications (binary outcomes), the pooled estimate suggested a potential reduction in risk (RR = 0.67, 95% CI 0.44-1.00) based on three studies and should be interpreted cautiously. Platform-overview pooling of continuous outcomes suggested variability across platforms, with more consistently positive pooled effects for Facebook Messenger-based interventions than for WhatsApp or WeChat; however, platform-by-domain pooling was often not estimable because many platform-domain combinations contributed fewer than three effect sizes. Meta-regression did not identify a clear linear association of follow-up duration or ln(sample size) with effect size, and explained little heterogeneity. CONCLUSIONS: Instant-messaging interventions for diabetes do not yield a clearly favorable overall pooled effect, but they show credible benefits for behavioral outcomes and selected clinical endpoints. Variation in effects appears more consistent with differences in intervention design and implementation than with platform labels alone. Future trials should report intervention components and maintenance strategies in greater detail and evaluate interactive, care-integrated messaging models. SYSTEMATIC REVIEW REGISTRATION: https://www.crd.york.ac.uk/PROSPERO/view/CRD420251079157, identifier: CRD420251079157.