Description

Heart failure is a major complication of various heart diseases and is the leading lethal cause of cardiovascular death worldwide. Based on the left ventricular ejection fraction (LVEF), heart failure can be divided into heart failure with reduced ejection fraction (HFrEF), heart failure with preserved ejection fraction (HFpEF) and heart failure with mildly reduced ejection fraction (HFmrEF). Heart failure rehospitalization rates and in-hospital complications did not differ between HFrEF and HFpEF. However, over the past two decades, the survival rate of HFrEF has improved significantly, whereas HFpEF has remained stagnant. One of the major reasons for this is that the diagnostic process of HFpEF is complicated, and it is easy to cause missed diagnosis in the clinic, resulting in delayed treatment.

Multimodal large language models are capable of integrating and analyzing medical data from different sources, including textual data (e.g., medical records, medical literature), image data (e.g., electrocardiograms, CT scan images), and audio data (e.g., symptoms narrated by patients). This multimodal data integration capability is crucial for understanding complex medical scenarios, as it provides a more comprehensive view of the condition than a single data source.

The diagnosis of HFpEF faces many challenges and requires clinicians to make judgments on multi-dimensional data, which can easily lead to the underdiagnosis and misdiagnosis of the disease. As a generative artificial intelligence tool, a large language model is able to integrate and analyze data from different sources and has the ability to learn and evolve from existing clinical evidence. Based on this, this study intends to evaluate the effectiveness of multimodal large language model for screening for heart failure with preserved ejection fraction (HFpEF), comparing it with the traditional clinical standard assessment process.