AI as second reader boosts chest CT orders for initially missed pneumothorax

AI-assisted CXR interpretation increased simulated CT-ordering intent from 8.1% to 26.7% (+18.6pp, p<0.001) in a nationwide simulation with 231 ED physicians, mainly by capturing initially missed pneumothoraces (diagnostic capture rate rose from 54.8% to 85.0% in subtle cases).…

• Design: Nationwide online simulation with 231 emergency physicians (51.5% GPs, 25.5% EM residents, 22.9% specialists) interpreting 20 chest radiographs each, yielding 4,620 response pairs.
• In subtle pneumothorax, AI converted 62.2% of initially missed diagnoses (339 cases where physicians answered "no pneumothorax") into CT orders (211/339).
• Limitation: Simulated CT-ordering intent, not real-world observed behaviour; single-country (Türkiye) study requiring prospective validation with actual CT-ordering data.

RadPigeon summaries are original and for information only. They are not clinical advice.