Behavioral Dynamics of AI Trust and Health Care Delays Among Adults: Integrated Cross-Sectional Survey and Agent-Based Modeling Study

Comprehensive Summary

This study, presented by Cai et al., investigates whether excessive trust in artificial intelligence (AI) health tools leads adults to delay seeking professional medical care, particularly among patients with chronic illnesses. To do this, researchers conducted a cross-sectional survey of 2,460 Chinese adults from December 2024 to May 2025 using a 21-item questionnaire that measured AI trust, AI usage frequency, chronic disease status, and self-reported healthcare delays. The data was then analyzed using multivariable logistic regression and mediation analysis, followed by agent-based modeling (ABM) to simulate intervention strategies over a 14-day period with 2,460 virtual agents in a small-world network. Researchers found that higher AI trust was associated with a 9% increase in odds of delayed care-seeking, with AI usage frequency partially mediating this relationship by increasing delay odds by 24%. This also highlighted that chronic disease status amplified delay odds by 42%, and the ABM revealed a bidirectional trust erosion loop where population delay rates declined from 10.6% to 9.5% as trust decreased. The authors finish their claims by explaining that broadcast messaging interventions were most effective at reducing delays, while network rewiring paradoxically increased delays through a "trust polarization" effect, suggesting that AI health tools should prioritize calibrated decision support over full automation.

Outcomes and Implications

This research is important to the medical field because it reveals an unintended consequence of AI health tools where increased trust can paradoxically delay patients from seeking necessary professional care, particularly among vulnerable populations with chronic conditions. Unlike previous studies that only examined static associations, this work demonstrates dynamic feedback loops where AI trust and delay behaviors reinforce each other over time, creating a potentially dangerous cycle for patient outcomes. In terms of its clinical importance, understanding how frequent AI users and chronically ill patients are at heightened risk for care delays can help healthcare systems design better guardrails for AI health applications. The finding that broadcast messaging reduced delays while network-based interventions backfired suggests that population-level education campaigns may be more effective than peer-influence strategies when addressing AI-related healthcare delays. With the ABM demonstrating how trust erosion can actually reduce delays in some scenarios, the approach highlights the complex relationship between technology confidence and health-seeking behavior. While further research is needed to validate these findings in different cultural contexts and healthcare systems, this study lays the foundation for developing AI health tools that explicitly encourage appropriate professional care-seeking rather than serving as substitutes for medical consultation.