Researchers from the University of Michigan and University of Toronto argue that healthcare AI tools are being widely deployed despite limited data demonstrating improvements in patient outcomes (Wiens and Goldenberg, Nature Medicine, 2026).

The MIT Technology Review reports a shift in clinician attitudes leading to rapid adoption of ambient AI scribes, with early studies highlighting reduced burnout but omitting analysis of effects on clinical decision-making or long-term patient health (MIT Technology Review, April 24, 2026). A University of Minnesota study from January 2025 found 65% of hospitals using predictive AI with incomplete evaluations of accuracy and bias (Nong et al., 2025).

Wiens notes that even accurate tools may not improve outcomes due to variables in reliance, workflows, and potential cognitive effects on doctors (Nature Medicine, 2026). Related findings from a 2024 NEJM AI review confirm few AI interventions have undergone randomized controlled trials measuring patient-centered outcomes (NEJM AI, 2024).

While FDA has cleared over 700 AI medical devices since 2010 primarily on technical performance data, post-market outcome studies remain rare (FDA AI/ML database, 2025). Wiens advocates for context-specific evaluations to identify unintended consequences across hospital settings and clinician experience levels (Nature Medicine, 2026).