A multimodal AI system achieved 96.9% guideline adherence compared to 84.4% for standard clinical practice (p=0.046) and demonstrated 93.8% overall diagnostic accuracy in cardiac emergencies.
Observational (n=32)
No
Does a multimodal AI system improve diagnostic accuracy and guideline adherence compared to standard clinical practice in patients presenting with acute cardiac emergencies?
A multimodal AI system integrating specialized large language models demonstrated high diagnostic accuracy (93.8%) and significantly improved guideline adherence compared to standard clinical practice in the triage and management of acute cardiac emergencies.
Absolute Event Rate: 96.9% vs 84.4%
p-value: p=0.046
To evaluate the diagnostic accuracy and guideline adherence of a multimodal artificial intelligence (AI) system integrating specialized models for electrocardiogram (ECG) interpretation, clinical reasoning, and risk stratification in acute cardiac emergencies.We retrospectively analyzed 32 consecutive patients who presented to Etimesgut ehit Sait Ertrk State Hospital, Ankara, Trkiye, with cardiac emergencies between September 2024 and January 2025.A multimodal AI system integrating GPT-4o with vision capabilities for ECG interpretation, Claude 3.5 Sonnet for clinical reasoning, and Gemini 2.0 Flash for risk stratification was compared with standard clinical practice.The cohort comprised patients with acute coronary syndrome (ACS) (n=14), acute pericarditis (n=6), pulmonary embolism (n=5), aortic dissection (n=4), and myocarditis (n=3).Diagnostic accuracy, guideline concordance, and clinical outcomes were assessed.The multimodal AI achieved an overall diagnostic accuracy of 93.8% (30/32) and correctly classified all ACS subtypes (14/14, 100%), including ST-elevation myocardial infarction (STEMI) localization.Pericarditis, pulmonary embolism, aortic dissection, and myocarditis were correctly identified in 83.3% (5/6), 80.0% (4/5), 75.0%(3/4), and 66.7% (2/3) of cases, respectively.Differentiation between STEMI and pericarditis based on ST-segment morphology achieved an accuracy of 91.2%.Guideline adherence was 96.9% (31/32) for the multimodal AI versus 84.4% (27/32) for clinical practice (p=0.046).The AI correctly identified all high-risk patients requiring intensive monitoring.Multimodal AI systems integrating specialized models for ECG interpretation, clinical reasoning, and risk stratification demonstrate high diagnostic accuracy and guideline adherence in cardiac emergencies, approaching human expert-level performance.These pilot results support larger prospective validation trials examining multimodal AI integration in time-critical cardiovascular care.
Ekingen et al. (Thu,) conducted a observational in Cardiac emergencies (n=32). Multimodal AI system vs. Standard clinical practice was evaluated on Guideline adherence (p=0.046). A multimodal AI system achieved 96.9% guideline adherence compared to 84.4% for standard clinical practice (p=0.046) and demonstrated 93.8% overall diagnostic accuracy in cardiac emergencies.