Key result
ChatGPT-4o matches specialists for initial endocarditis diagnosis in ~90% of cases but lags in antibiotic selection.
Why the study?
Does ChatGPT-4o provide a management plan of comparable quality and safety to local infectious diseases specialists in adults with infective endocarditis?
Cohort (n=50)
No
Does ChatGPT-4o provide a management plan of comparable quality and safety to local infectious diseases specialists in adults with infective endocarditis?
Answers are grounded in this paper and cite their sources.
While ChatGPT-4o shows high accuracy in diagnosing infective endocarditis using current guidelines, its lower appropriateness and potential harm in selecting antimicrobial therapy preclude its independent clinical use.
Background The role of artificial intelligence in managing complex infectious diseases (ID) remains poorly understood. This study assessed the quality and safety of the management plan recommended by ChatGPT-4o in patients with infective endocarditis (IE). Methods Prospective, single-center study enrolling adults with IE over an 8-months period. Clinical data for each patient, using ESC guidelines and Duke-ISCVID criteria as external knowledge or fine-tuning data, were provided to ChatGPT-4o asking for a management plan (diagnosis, diagnostic work-up, antimicrobial therapy, follow-up). The recommendations provided by local ID specialists (IDSs) and ChatGPT-4o were rated by three independent IDSs using a standardized scoring system. Evaluations were compared using McNemar-Bowker test and the inter-grader agreement was assessed. Results Fifty patients with IE were included, mostly with definite IE (86%) sustained by Gram-positive bacteria (80%). ChatGPT-4o initial diagnoses were identical to IDSs in ≥ 90% of cases with an optimal inter-grader agreement. Despite low inadequacy (0%-12%), the appropriateness of work-up proposed by ChatGPT-4o was significantly lower than that of IDSs because it was not specific enough in almost half of cases. The percentages of optimal antimicrobial therapy suggested by ChatGPT-4o were lower than those of IDSs (34%-66% vs. 72%-86%) and a trend towards lower adequacy was retrieved (inadequate 2%-18% vs. 2%-8%, harmful 0%-14% vs. 0%-2%), albeit the agreement between graders was moderate-low. The follow-up appropriateness was good (53%-82%). Conclusion While helpful in diagnosing IE, the application of ChatGPT-4o based on current guidelines requires further development in decision-making tasks, as antimicrobial therapy, and should not be employed independently.
No takes yet. Share an insight, caveat, or question.
Denti et al. (2026) conducted a cohort in Infective endocarditis (n=50). ChatGPT-4o management plan using ESC guidelines and Duke-ISCVID criteria vs. Local infectious diseases specialists (IDSs) consultation was evaluated on Quality and safety of the management plan (diagnosis, diagnostic work-up, antimicrobial therapy, follow-up) rated by three independent ID specialists. ChatGPT-4o matched infectious diseases specialists for initial infective endocarditis diagnosis in ≥90% of cases, but suggested optimal antimicrobial therapy less often (34%-66% vs. 72%-86%).
Synapse has enriched 10 related papers. Consider them for comparative context: