Social Support Mechanisms in an Online Type 1 Diabetes Community: Social Network Analysis of Stakeholder Diversity...

Online health communities (OHCs) have emerged as critical platforms for patients with type 1 diabetes (T1D) to exchange informational and emotional support. However, how stakeholder roles and disease duration jointly shape support dynamics and influence formation remains underexplored.

This study aimed to examine network-based social support mechanisms in a large T1D OHC, focusing on how stakeholder diversity and disease duration are associated with social support behaviors, subnetwork structures, and user influence.

This retrospective observational study analyzed digital trace data from China's largest T1D online community (January 1-May 20, 2024), comprising 43,788 posts and 145,423 comments contributed by 1393 users. We manually annotated 2000 randomly sampled posts and fine-tuned a GPT-4o-mini (OpenAI) to classify support type (informational or emotional, and seeking or providing), yielding 20,384 support-related posts and 56,953 comments from 1224 users. We constructed weighted directed informational and emotional interaction networks and modeled predictors of a composite influence metric (Relative Centrality) using a gamma log-link generalized linear model (including demographics, identity, sentiment, disease duration, posting orientation, and cyclical activity time). Analyses were conducted in Python (version 3.11; Python Software Foundation). Statistical significance was set at P<.05.

Support predominantly flowed from longer-duration members (≥ y) to those at earlier stages (≤5 y). Both subnetworks exhibited multicentered, star-like structures; the informational subnetwork had broader participation (density 0.031, diameter 7), while the emotional network was denser (density 0.039, diameter 6). In the influence model, peer supporters had substantially higher influence than patients (exp(β)=34.79, 95% CI 18.94-64.08; P<.001), professionals lower (exp(β)=0.41, 95% CI 0.17-0.99; P=.055), and women higher than men (exp(β)=1.65, 95% CI 1.23-2.23; P=.001). Positive sentiment was associated with higher influence (exp(β)=1.91, 95% CI 1.22-2.97; P=.005), and negative lower (exp(β)=0.54, 95% CI 0.37-0.79; P=.001). Influence followed an inverted U-shaped trajectory over disease duration, peaking at approximately the 116th month (95% CI 43.25-188.91).

This study suggests that social support patterns and user influence in a T1D OHC vary by stakeholder role and disease duration. Users with shorter disease duration more often sought support, whereas longer-duration users more often provided support, and informational and emotional exchanges formed distinct interaction subnetworks. Peer supporters were the most influential users; influence was also associated with gender, sentiment, activity timing, and a nonlinear (inverted U-shaped) relationship with disease duration. These findings may inform peer-facilitated, stage-tailored community strategies, with professionals engaged in targeted, complementary roles. A patient-centered collaborative care approach integrating peer experience with multidisciplinary clinical input could be explored in future work.