Large language model detects previously undiagnosed heart failure with preserved ejection fraction in patients with metabolic-associated fatty liver disease: A multicenter cohort study - PubMed
4 hours ago
- #fatty liver disease
- #artificial intelligence
- #heart failure
- A domain-tuned large language model (MedGuide-14B) identified previously undiagnosed HFpEF in 17.6% of patients with MAFLD, beyond the 12.7% clinically diagnosed.
- Model performance was high: AUC of 0.94, sensitivity 95.0%, specificity 92.3%, and 90.4% of newly identified cases confirmed in blinded validation (κ = 0.85).
- MedGuide-identified HFpEF patients had intermediate outcomes: 55.6% rehospitalization and 12.3% all-cause mortality at 48 months, between clinically diagnosed HFpEF and non-HFpEF groups.