De-identified numerical data supporting HFpEF detection using MedGuide-14B

This repository contains fully de-identified numerical data underlying the figures and tables of the manuscript: “Large language model detects previously undiagnosed heart failure with preserved ejection fraction in patients with metabolic-associated fatty liver disease: a multicenter cohort study” The dataset includes group-level baseline characteristics, model performance metrics, blinded clinician validation results, Kaplan–Meier survival curve values for all-cause and cardiovascular mortality, rehospitalization outcomes with unadjusted relative risks and odds ratios, calibration analysis data, and standardized inference prompts. All data are fully anonymized and contain no patient-level identifiers.