Diagnostic signature for heart failure with preserved ejection fraction (HFpEF): a machine learning approach using multi-modality electronic health record data

Table 2 Multivariable model performance using the fivefold cross-validation in derivation dataset

Model	Control set	f1_macro ± 95% CI	f1_weighted ± 95% CI	AUROC ± 95% CI
Structured signature	Non-HF	84.05 ± 2.7	84.18 ± 2.7	92.04 ± 1.4
Structured signature	HFrEF	75.75 ± 2.1	87.22 ± 1.42	90.31 ± 3.5
Unstructured signature	Non-HF	98.81 ± 1.3	98.82 ± 1.3	99.7 ± 0.5
Unstructured signature	HFrEF	78.59 ± 4.9	88.99 ± 2.1	94.38 ± 1.4
Combined signature	Non-HF	98.57 ± 1.4	98.59 ± 1.4	99.8 ± 0.3
Combined signature	HFrEF	83.03 ± 2.8	90.91 ± 1.6	95.67 ± 2.0

ISSN: 1471-2261