Investigating Patient Trust and Intention to Act in AI-Supported Clinical Decisions: Cross-Sectional Survey

Background: In health systems, AI is increasingly being implemented as a form of clinical decision support (CDS). Research on patient perspectives, however, lags behind clinician-focused studies. Before the introduction of AI-based tools, clinicians often used and continue to use rule-based guidelines and risk calculators (RCs) to inform decision-making. Prior research has shown that RCs for CDS have mixed effects on patient adherence to treatment recommendations, whereas for AI-based tools, the effect is still not clear. Objective: This study aims to assess differences in patients' behavioral intentions, trust, and preferences for clinical recommendations provided by doctors alone, doctors supported by RCs, and doctors supported by AI-based risk predictions. We also evaluate how sociodemographic characteristics affect these perceptions. Methods: We surveyed 1185 US-based adults. Participants rated their likelihood of taking recommended clinical actions (behavioral intentions) across 6 clinical scenarios, where they were randomized to a different recommendation source (doctor, doctor + RC, and doctor + AI). Participants also rated their trust in the recommendation, indicated their preferred source for recommendations, and provided qualitative feedback. Results: Behavioral intentions differed by recommendation source such that the doctor + RC condition had greater odds of behavioral intention than the doctor + AI condition. Behavioral intentions were also affected by the clinical condition. Participants had greater odds of trusting doctors alone over doctor + RC, and both doctors alone and doctor + RC had greater trust than doctor + AI. However, a plurality of participants ultimately preferred a doctor using an RC for making recommendations. Several sociodemographic differences were detected, namely, participants with higher subjective numeracy and literacy had higher odds of taking clinical action and higher trust across decision-support sources. We detected qualitative themes in the following categories: decision quality and reliability; trust and perceptions; patient-centeredness and autonomy; fairness, transparency, and ethics. Across categories, participants reported skepticism toward all decision sources and potential misunderstanding of AI's ability to personalize risk predictions. Some participants also stated that they had no additional concerns or were uncertain regarding their feelings. Conclusions: These findings reveal novel, nuanced patient preferences shaped by task and context, offering guidance for aligning AI-supported care with patient values. Patients were not aligned on the recommendation source across intentions, trust, and preferences. Taken together, our results indicate that patients prefer doctors working with easy-to-understand, explainable support but are still learning to trust these tools. Our findings also highlight the continued importance of the clinical context and its interaction with computer-assisted decision support. Future work should evaluate how we can transparently convey to patients the use of AI in clinical decision-making, embedded within the given clinical context. This may present an opportunity to bolster, rather than erode, patient trust.