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

Table 5 Multivariable model performance in independent test cohort

	Performance metric	H₂FPEF, %	Combined HFpEF versus non-HF signature, %	Combined HFpEF versus HFrEF signature, %	Aggregate model HFpEF versus HFrEF/non-HF score, %	Scoring Agreement* (HFpEF versus HFrEF and HFpEF versus non-HF)
Test set	AUROC (95% CI)	0.77	0.86 (± 0.002)	0.85 (± 0.001)	0.90 (± 0.002)	0.3
Test set	AP	0.53	0.70	0.66	0.74	0.3

The 95% CI is reported using bootstrapping in a thousand of iterations
AUROC area under receiver operative curve; AP average precision; CI confidence interval in bootstrapped samples
*HFpEF annotation agreement between the two scoring systems using Cohen’s kappa statistics (python 3, Sklearn v.0.22)

ISSN: 1471-2261