An AI program generated a response regarding CK-MB testing in 15 s, but clinical experts deemed it inaccurate and incomplete, indicating AI is not yet ready for answering clinical practice questions.
Does an AI program provide accurate and appropriate scientific content for laboratory medicine questions?
AI-generated responses to clinical laboratory questions currently lack the accuracy and nuance required for clinical practice.
Abstract Background Artificial intelligence (AI) promises to become an important tool in the practice of laboratory medicine. AI programs are available online that can provide concise medical and laboratory information within seconds after a question is submitted. Methods We posed the following contemporary and relevant question for clinical chemists using an AI program: “Should labs eliminate CK-MB testing?” The results of this inquiry were critically reviewed by a cardiologist, emergency department physician, and clinical laboratorians for accuracy and appropriateness. Results An AI report answering this question was generated within 15 s. The experts felt that the information was inaccurate with the statements that CK-MB results are released earlier than cardiac troponin (cTn), and are more useful for detection of myocardial injury in patients with renal insufficiencies. The summary omitted discussion on the ongoing debate for periprocedural detection of acute myocardial infarction (MI), the perceived value for detection of reinfarction, the value in risk stratification compared to cTn, the economic justification for testing both CK-MB and cTn, and medicolegal aspects of testing when results are discordant. Conclusions At this time, AI does not appear to be ready to be used by clinical laboratories for answering important practice questions.
Wu et al. (Fri,) conducted a other in Laboratory medicine. Artificial intelligence program was evaluated on Accuracy and appropriateness of AI-generated response. An AI program generated a response regarding CK-MB testing in 15 s, but clinical experts deemed it inaccurate and incomplete, indicating AI is not yet ready for answering clinical practice questions.