Participant-reported minimum acceptable sensitivity and specificity for artificial intelligence-based disease detection

Artificial intelligence (AI)-based diagnostic systems are increasingly integrated into clinical practice, yet the minimum diagnostic performance considered acceptable by patients and the public remains poorly understood. We conducted a nationwide survey of 1222 participants from the general population and a separate cohort of 106 long-term lung cancer survivors. Participants reported the maximum acceptable numbers of missed cases among 100 persons with disease and persons incorrectly classified as having disease among 100 persons without disease. These responses were converted to false-negative and false-positive fractions, from which minimum acceptable sensitivity and specificity were derived. Among 1328 participants, mean minimum acceptable sensitivity and specificity were both 0.90. Overall, 81.6% reported minimum acceptable sensitivity of at least 0.88, and 81.6% reported minimum acceptable specificity of at least 0.90, thereby exceeding the observed sensitivity of 0.874 and specificity of 0.895 reported for IDx-DR. In the combined analysis, cancer history was associated with higher minimum acceptable sensitivity and specificity after multivariable adjustment. These findings show that participants reported high minimum acceptable values for AI-based disease detection and support clear communication of diagnostic errors and predictive values in the intended-use population.