Application of a Multimodal Large Language Model to Assist Diagnosis for Heart Failure With Preserved Ejection Fraction

What this Trial Is About

Heart failure with preserved ejection fraction (HFpEF) is a challenging heart condition that remains difficult to diagnose accurately, leading to delays in treatment. Researchers are evaluating a multimodal large language model designed to screen for HFpEF by analyzing diverse medical data such as patient records, images, and audio symptoms. This study compares the model's effectiveness with the traditional clinical assessment process used by cardiologists. The study involves two diagnostic approaches: the standard routine consultation following expert consensus guidelines and the multimodal large language model system called MedGuide-72B. During patient visits, the model collects and analyzes data from the routine clinical interaction to make its diagnosis. Both methods aim to identify HFpEF in patients suspected of having this condition, focusing on integrating and interpreting complex, multi-source medical information. Participants are cardiology inpatients aged 18 to 80 years suspected of HFpEF based on heart ultrasound and symptoms like breathlessness or swelling. Researchers will measure diagnostic accuracy, sensitivity, specificity, patient satisfaction, diagnosis time, economic costs, physician workload, and diagnosis efficiency over about eight months. The study collects data during routine care visits without altering treatment, and participants will be monitored for the quality and consistency of the diagnostic processes.

Application of Multimodal Large Language Model in HFpEF