AI and clinicians growing together: A cross-sectional survey of clinicians' attitudes toward AI-CDSS with comparison to 2020 data.

backgroundClinical Decision Support Systems (CDSS) - software programs that provide patient-specific recommendations to assist in clinical decision-making - have evolved considerably since 2020, with increasing integration of artificial intelligence (AI) and machine-learning features. Modern AI-powered CDSS (AI-CDSS) no longer rely on static algorithms but instead create adaptive, data-driven insights to be used for diagnostics, prognosis and ‍ ‌‍ ‍‌therapy. Exposure in the clinical setting and attention from regulators have increased, yet uncertainty persists about how clinicians view the risks, benefits and integration of such systems within their everyday practice.

objectivesOur objective in this research was to assess AI‑CDSS perception in 2025 and to explore cross-temporal patterns relative to earlier reports. Specifically, we sought to: (1) examine cross-temporal trends in perceived benefits and harms; (2) describe clinician subgroups by adoption intentions; and (3) examine professional and regulatory concerns following real-world experience with AI.

methodsA cross-sectional online survey was administered to UK and Italian clinicians (N = 215). The instrument maintained thematic continuity with a 2020 survey while incorporating AI-specific constructs. The questions spanned five themes: perceived advantages, hazards, regulation, clinical utility, and ethical concerns. Cluster analysis (Ward's method, z-scored items) was used to identify attitudinal clusters. Comparison through time with 2020 data prioritised thematic concordance and relative frequencies.

resultsAI‑CDSS are now more commonly used for diagnostic support (32.6%) compared to primarily administrative purposes in 2020. Greater endorsement was found for AI benefits such as improved diagnostics (63.3%) and medicine management (62.8%). Concerns moved from technological performance to professional issues, such as de-skilling of trainees (59.5%) and automation bias (67%). Regulatory concerns moved from device-focused agencies to evidence synthesis organisations (eg NICE: 31.2%). Hierarchical cluster analysis identified three distinct attitudinal profiles: The Optimists (n = 113), who reported high perceived benefits and low risk; The Balanced Sceptics (n = 83), with moderate scores across dimensions; and The Concerned (n = 19), who reported low perceived benefits and elevated risk perception.

conclusionClinicians tend to display more nuanced, context-specific views of AI‑CDSS following real-world exposure. Clinicians in 2025 reported moderate to high trust in AI-based tools; however, as trust was measured in 2025 only, no direct cross-temporal trust comparison can be made. Persistent concerns regarding ethics, explainability and professional education remain. Specialised regulatory frameworks and training models are needed to optimise safe, effective integration into modern practice.