Impact of AI-Based Clinical Decision Support Systems on Diagnostic Accuracy Among Healthcare Professionals: A Systematic Review and Meta-Analysis of Randomized Controlled Trials

초록

Background: Diagnostic errors affect approximately 5-15% of clinical encounters globally, contributing to significant patient harm. Artificial intelligence-based clinical decision support systems (AI-CDSS) are increasingly deployed to augment clinician diagnostic performance, yet rigorous evidence from randomized controlled trials (RCTs) remains limited. This systematic review and meta-analysis aims to quantify the effect of AI-CDSS on diagnostic accuracy among healthcare professionals. Methods: We systematically searched PubMed/MEDLINE, CINAHL, Embase, Cochrane CENTRAL, and Google Scholar from 2000 to 2026. Eligible studies were peer-reviewed RCTs comparing AI-CDSS with standard care. Risk of bias was assessed using the Cochrane RoB 2 tool. Random-effects meta-analysis was performed using standardized mean differences (SMD). Certainty of evidence was evaluated using GRADE. Results: Five RCTs (N = 12,657 participants) were included. The pooled SMD was 0.182 (95% CI: 0.003-0.362; p = 0.047; I2 = 68.6%), with the lower confidence bound approaching zero, indicating preliminary evidence of a modest, statistically marginal improvement with AI-CDSS. Subgroup analyses suggested greater effects for deep learning systems and chest radiology applications, though single-study subgroups preclude definitive comparative conclusions. No significant publication bias was detected (Egger's p = 0.18). GRADE certainty was rated MODERATE. Conclusions: This meta-analysis provides preliminary evidence that AI-CDSS may modestly improve diagnostic accuracy under specific conditions; however, the marginal statistical significance and near-zero lower confidence bound necessitate cautious interpretation. Implementation should prioritize contexts with demonstrated effectiveness and include ongoing outcome monitoring.

키워드