AI sensitivity gains in pediatric elbow fracture detection may be inflated by OR-rule combination, commentators argue

Theoretical AI assistance using an OR-rule (reader OR AI positive) overestimates clinical sensitivity gains by lowering the positivity threshold, sharply reducing specificity and PPV. Commentary calls for prospective real-time reader studies to assess true utility, including wor…

• Design: Methodological commentary on a retrospective external validation study of AI-assisted pediatric elbow fracture detection.
• Concerns: The reported sensitivity increase reflects both the algorithm's performance and the artificial lowering of the positivity threshold by the OR-rule; gains were partly driven by isolated joint effusions and 'doubt' outputs labeled as abnormal.
• Limitation: No new empirical data; the letter relies on logical critique of retrospective, theoretical AI assistance without prospective real-world evidence.

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